Post on 19-Jan-2016
description
NASA Review (71007)
1
High Resolution Soil Moisture Estimation via Data Assimilation Using
NASA Land Information System
Valentine Anantharaj Georgy Mostovoy Anish Turlapaty and Jim Aanstoos
Mississippi State University - GeoResources Institute
NASA Review (71007)
2
LIS Evaluation Team amp Collaborators
bull RPC Teamndash Valentine Anantharaj Georgy Mostovoy Nicholas
Younan Jim Aanstoos and Anish Turlapaty (MSU)ndash Christa Peters-Lidard (NASA GSFC HSB)ndash Paul Houser (GMU CREW)ndash Bailing Li and Sujay Kumar (GSFC)
bull Collaborators and Consultantsndash USDA NRCSndash MSU DREC and USDA (Stoneville MS)
NASA Review (71007)
3
Identified Needs of USDA NRCS
bull Routine analysis soil moisture over the continental needs
watersoilssunweatherclimatevegetationterrain
watersoilssunweatherclimatevegetationterrain
observe model assimilateobserve model assimilate
NASA Review (71007)
4
Soil Moisture Data Sources in this RPC Experiment
bull In-situ observationsndash USDA Soil Climate Analysis Network (SCAN)
bull Remotely sensed and estimatedndash NASA and JAXA Aqua Advanced Scanning
Microwave Radiometer ndash EOS (AMSR-E)
bull Numerical Modelsndash The Noah model in the NASA Land Information
System
NASA Review (71007)
5
USDA NRCS SCAN
NASA Review (71007)
6
Anticipated Societal Benefits
1 provides critical information to support drought monitoring and mitigation
2 provides essential information for predicting droughts based on weather and climate predictions
3 supports irrigation water management4 supports fire risk assessment5 supports water supply forecasting and NWS flood forecasting6 supplies a critical missing component to assist with snow climate
and associated hydrometeorological data analysis7 supports climate change assessment8 enables water quality monitoring9 supports a wide variety of natural resource management amp research
activities such as NASA remote sensing activities of soil moisture and ARS watershed studies
NASA Review (71007)
7
An Integrated Framework forLand Data Assimilation System
ApplicationsInputs OutputsPhysics
TopographySoils
WaterSupply ampDemand
AgricultureHydro-ElectricPower
EcologicalForecasting
Water Quality
ImprovedShort Term
ampLong TermPredictions
Land Cover and Vegetation (MODIS AMSRTRMM SRTM)
Meteorology Modeled amp
Observed (TRMM GOES Station)
Observed Land States(Snow ET Soil Moisture Water
Carbon etc)
Land Surface Models (LSM)Physical Process Models
Noah CLM VIC SiB2 Mosaic Catchment etc
Data Assimilation Modules(EnKF EKF)Rule-based
Water Fluxes Runoff
Surface States
Moisture Carbon Ts
Energy FluxesLe amp H
Biogeo-chemistry
Carbon Nitrogen etc
(Peters-Lidard Houser Kumar Tian Geiger)
NASA Review (71007)
8
LIS Evaluations Purpose and Activities
NASA Review (71007)
9
Purpose of RPC Evaluations hellip
bull Primaryndash Evaluate LIS capabilities and NASA data to enhance
and extend USDA-NRCS SCANbull Approach
ndash Evaluate LIS performancendash Assimilate SCAN and AMSR-E observations and
evaluate LIS capabilities to enhance SCAN by means of Observation Sensitivity Experiments (OSE)
ndash Derive physically consistent soil moisture maps at a range of spatial resolutions from 25x25 km2 to 1x1 km2
ndash Quantify uncertainties at all scales
NASA Review (71007)
10
Team Activity
bull MsState Project Management RPC Integration Control Run MODIS-VF [SSURGO]
bull NASA GSFC LIS Support AMSR-E data assimilation science expertise
bull GMU CREW SCAN data assimilation science expertise
NASA Review (71007)
11
Data Assimilation and Observation Sensitivity Experiments
bull Evaluation of data assimilation techniquesndash EKF EnKF
bull Data assimilation (land state)ndash Soil moisture
bull Soil moisture stationsbull AMSR-E
ndash Temperaturebull MODIS LST []
bull Sensitivity studiesbull Expected Outcomes high resolution soil moisture
analysis product uncertainty characterization
NASA Review (71007)
12
Status of Current Activities
bull Preliminary evaluation of simulated soil moisture data ndash Georgy Mostovoy
bull Quality Assessment of soil moisture measurements AMSR-E and SCAN - Anish Turlapaty
NASA Review (71007)
13
Future Directions
bull Assimilate AMSR-E soil moisture datandash Evaluate AMSR-E impacts
bull Incorporate MODIS Vegetation Fraction (VF) and compare with control runndash Evaluate MODIS VF impacts
bull Assimilate SCAN soil moisture datandash Evaluate SCAN impacts
NASA Review (71007)
14
ASMR-E Soil Moisture Data Assimilation and Evaluation
Noah Land Surface Model of NASA Land Information System
Soil Moisture Data
Soil Climate Analysis Network
AMSR-Eon NASA
AQUA Satellite
Evaluation Study
Soil Moisture Data
Soil Moisture Data
Soil Moisture Data
No D
A
EnKF DA
NASA Review (71007)
15
Future plansAssimilation of AMSR-E soil moisture data
12 hour time step 3 hourly output and 5 ensemble members
00Z 03Z 06Z 09Z 12Z 15Z 18Z 21Z 00Z
12 hr forecast+obs 12 hr forecast+obs
Data assimilation frequency will be twice daily at 06Z and 18Z DADA will will not be ldquoturned onrdquo until observation is available not be ldquoturned onrdquo until observation is available We plan to take the ensemble mean as first guess for next time step initial conditions
NASA Review (71007)
16
Noah LSM RUN AMSR-E SM EnKF Assimilation(TEST2)
Scaled AMSR-E SM
Expected Result [Example Only]EnKF Assimilation of AMSR-E SM Retrievals
Noah LSM RUN
EnKF Assimilation of Scaled AMSR-E SM RetrievalsEnKF Assimilation (TEST2)
Example
Only
NASA Review (71007)
17
Preliminary Evaluation of Soil Moisture Simulated by the Noah
Land Surface Model
Georgy Mostovoy
Geographical distribution of SCAN sites
OBJECTIVE Validation of the Noah Land Surface Model (LSM) baseline runsversus SCAN soil moisture observations
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
2
LIS Evaluation Team amp Collaborators
bull RPC Teamndash Valentine Anantharaj Georgy Mostovoy Nicholas
Younan Jim Aanstoos and Anish Turlapaty (MSU)ndash Christa Peters-Lidard (NASA GSFC HSB)ndash Paul Houser (GMU CREW)ndash Bailing Li and Sujay Kumar (GSFC)
bull Collaborators and Consultantsndash USDA NRCSndash MSU DREC and USDA (Stoneville MS)
NASA Review (71007)
3
Identified Needs of USDA NRCS
bull Routine analysis soil moisture over the continental needs
watersoilssunweatherclimatevegetationterrain
watersoilssunweatherclimatevegetationterrain
observe model assimilateobserve model assimilate
NASA Review (71007)
4
Soil Moisture Data Sources in this RPC Experiment
bull In-situ observationsndash USDA Soil Climate Analysis Network (SCAN)
bull Remotely sensed and estimatedndash NASA and JAXA Aqua Advanced Scanning
Microwave Radiometer ndash EOS (AMSR-E)
bull Numerical Modelsndash The Noah model in the NASA Land Information
System
NASA Review (71007)
5
USDA NRCS SCAN
NASA Review (71007)
6
Anticipated Societal Benefits
1 provides critical information to support drought monitoring and mitigation
2 provides essential information for predicting droughts based on weather and climate predictions
3 supports irrigation water management4 supports fire risk assessment5 supports water supply forecasting and NWS flood forecasting6 supplies a critical missing component to assist with snow climate
and associated hydrometeorological data analysis7 supports climate change assessment8 enables water quality monitoring9 supports a wide variety of natural resource management amp research
activities such as NASA remote sensing activities of soil moisture and ARS watershed studies
NASA Review (71007)
7
An Integrated Framework forLand Data Assimilation System
ApplicationsInputs OutputsPhysics
TopographySoils
WaterSupply ampDemand
AgricultureHydro-ElectricPower
EcologicalForecasting
Water Quality
ImprovedShort Term
ampLong TermPredictions
Land Cover and Vegetation (MODIS AMSRTRMM SRTM)
Meteorology Modeled amp
Observed (TRMM GOES Station)
Observed Land States(Snow ET Soil Moisture Water
Carbon etc)
Land Surface Models (LSM)Physical Process Models
Noah CLM VIC SiB2 Mosaic Catchment etc
Data Assimilation Modules(EnKF EKF)Rule-based
Water Fluxes Runoff
Surface States
Moisture Carbon Ts
Energy FluxesLe amp H
Biogeo-chemistry
Carbon Nitrogen etc
(Peters-Lidard Houser Kumar Tian Geiger)
NASA Review (71007)
8
LIS Evaluations Purpose and Activities
NASA Review (71007)
9
Purpose of RPC Evaluations hellip
bull Primaryndash Evaluate LIS capabilities and NASA data to enhance
and extend USDA-NRCS SCANbull Approach
ndash Evaluate LIS performancendash Assimilate SCAN and AMSR-E observations and
evaluate LIS capabilities to enhance SCAN by means of Observation Sensitivity Experiments (OSE)
ndash Derive physically consistent soil moisture maps at a range of spatial resolutions from 25x25 km2 to 1x1 km2
ndash Quantify uncertainties at all scales
NASA Review (71007)
10
Team Activity
bull MsState Project Management RPC Integration Control Run MODIS-VF [SSURGO]
bull NASA GSFC LIS Support AMSR-E data assimilation science expertise
bull GMU CREW SCAN data assimilation science expertise
NASA Review (71007)
11
Data Assimilation and Observation Sensitivity Experiments
bull Evaluation of data assimilation techniquesndash EKF EnKF
bull Data assimilation (land state)ndash Soil moisture
bull Soil moisture stationsbull AMSR-E
ndash Temperaturebull MODIS LST []
bull Sensitivity studiesbull Expected Outcomes high resolution soil moisture
analysis product uncertainty characterization
NASA Review (71007)
12
Status of Current Activities
bull Preliminary evaluation of simulated soil moisture data ndash Georgy Mostovoy
bull Quality Assessment of soil moisture measurements AMSR-E and SCAN - Anish Turlapaty
NASA Review (71007)
13
Future Directions
bull Assimilate AMSR-E soil moisture datandash Evaluate AMSR-E impacts
bull Incorporate MODIS Vegetation Fraction (VF) and compare with control runndash Evaluate MODIS VF impacts
bull Assimilate SCAN soil moisture datandash Evaluate SCAN impacts
NASA Review (71007)
14
ASMR-E Soil Moisture Data Assimilation and Evaluation
Noah Land Surface Model of NASA Land Information System
Soil Moisture Data
Soil Climate Analysis Network
AMSR-Eon NASA
AQUA Satellite
Evaluation Study
Soil Moisture Data
Soil Moisture Data
Soil Moisture Data
No D
A
EnKF DA
NASA Review (71007)
15
Future plansAssimilation of AMSR-E soil moisture data
12 hour time step 3 hourly output and 5 ensemble members
00Z 03Z 06Z 09Z 12Z 15Z 18Z 21Z 00Z
12 hr forecast+obs 12 hr forecast+obs
Data assimilation frequency will be twice daily at 06Z and 18Z DADA will will not be ldquoturned onrdquo until observation is available not be ldquoturned onrdquo until observation is available We plan to take the ensemble mean as first guess for next time step initial conditions
NASA Review (71007)
16
Noah LSM RUN AMSR-E SM EnKF Assimilation(TEST2)
Scaled AMSR-E SM
Expected Result [Example Only]EnKF Assimilation of AMSR-E SM Retrievals
Noah LSM RUN
EnKF Assimilation of Scaled AMSR-E SM RetrievalsEnKF Assimilation (TEST2)
Example
Only
NASA Review (71007)
17
Preliminary Evaluation of Soil Moisture Simulated by the Noah
Land Surface Model
Georgy Mostovoy
Geographical distribution of SCAN sites
OBJECTIVE Validation of the Noah Land Surface Model (LSM) baseline runsversus SCAN soil moisture observations
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
3
Identified Needs of USDA NRCS
bull Routine analysis soil moisture over the continental needs
watersoilssunweatherclimatevegetationterrain
watersoilssunweatherclimatevegetationterrain
observe model assimilateobserve model assimilate
NASA Review (71007)
4
Soil Moisture Data Sources in this RPC Experiment
bull In-situ observationsndash USDA Soil Climate Analysis Network (SCAN)
bull Remotely sensed and estimatedndash NASA and JAXA Aqua Advanced Scanning
Microwave Radiometer ndash EOS (AMSR-E)
bull Numerical Modelsndash The Noah model in the NASA Land Information
System
NASA Review (71007)
5
USDA NRCS SCAN
NASA Review (71007)
6
Anticipated Societal Benefits
1 provides critical information to support drought monitoring and mitigation
2 provides essential information for predicting droughts based on weather and climate predictions
3 supports irrigation water management4 supports fire risk assessment5 supports water supply forecasting and NWS flood forecasting6 supplies a critical missing component to assist with snow climate
and associated hydrometeorological data analysis7 supports climate change assessment8 enables water quality monitoring9 supports a wide variety of natural resource management amp research
activities such as NASA remote sensing activities of soil moisture and ARS watershed studies
NASA Review (71007)
7
An Integrated Framework forLand Data Assimilation System
ApplicationsInputs OutputsPhysics
TopographySoils
WaterSupply ampDemand
AgricultureHydro-ElectricPower
EcologicalForecasting
Water Quality
ImprovedShort Term
ampLong TermPredictions
Land Cover and Vegetation (MODIS AMSRTRMM SRTM)
Meteorology Modeled amp
Observed (TRMM GOES Station)
Observed Land States(Snow ET Soil Moisture Water
Carbon etc)
Land Surface Models (LSM)Physical Process Models
Noah CLM VIC SiB2 Mosaic Catchment etc
Data Assimilation Modules(EnKF EKF)Rule-based
Water Fluxes Runoff
Surface States
Moisture Carbon Ts
Energy FluxesLe amp H
Biogeo-chemistry
Carbon Nitrogen etc
(Peters-Lidard Houser Kumar Tian Geiger)
NASA Review (71007)
8
LIS Evaluations Purpose and Activities
NASA Review (71007)
9
Purpose of RPC Evaluations hellip
bull Primaryndash Evaluate LIS capabilities and NASA data to enhance
and extend USDA-NRCS SCANbull Approach
ndash Evaluate LIS performancendash Assimilate SCAN and AMSR-E observations and
evaluate LIS capabilities to enhance SCAN by means of Observation Sensitivity Experiments (OSE)
ndash Derive physically consistent soil moisture maps at a range of spatial resolutions from 25x25 km2 to 1x1 km2
ndash Quantify uncertainties at all scales
NASA Review (71007)
10
Team Activity
bull MsState Project Management RPC Integration Control Run MODIS-VF [SSURGO]
bull NASA GSFC LIS Support AMSR-E data assimilation science expertise
bull GMU CREW SCAN data assimilation science expertise
NASA Review (71007)
11
Data Assimilation and Observation Sensitivity Experiments
bull Evaluation of data assimilation techniquesndash EKF EnKF
bull Data assimilation (land state)ndash Soil moisture
bull Soil moisture stationsbull AMSR-E
ndash Temperaturebull MODIS LST []
bull Sensitivity studiesbull Expected Outcomes high resolution soil moisture
analysis product uncertainty characterization
NASA Review (71007)
12
Status of Current Activities
bull Preliminary evaluation of simulated soil moisture data ndash Georgy Mostovoy
bull Quality Assessment of soil moisture measurements AMSR-E and SCAN - Anish Turlapaty
NASA Review (71007)
13
Future Directions
bull Assimilate AMSR-E soil moisture datandash Evaluate AMSR-E impacts
bull Incorporate MODIS Vegetation Fraction (VF) and compare with control runndash Evaluate MODIS VF impacts
bull Assimilate SCAN soil moisture datandash Evaluate SCAN impacts
NASA Review (71007)
14
ASMR-E Soil Moisture Data Assimilation and Evaluation
Noah Land Surface Model of NASA Land Information System
Soil Moisture Data
Soil Climate Analysis Network
AMSR-Eon NASA
AQUA Satellite
Evaluation Study
Soil Moisture Data
Soil Moisture Data
Soil Moisture Data
No D
A
EnKF DA
NASA Review (71007)
15
Future plansAssimilation of AMSR-E soil moisture data
12 hour time step 3 hourly output and 5 ensemble members
00Z 03Z 06Z 09Z 12Z 15Z 18Z 21Z 00Z
12 hr forecast+obs 12 hr forecast+obs
Data assimilation frequency will be twice daily at 06Z and 18Z DADA will will not be ldquoturned onrdquo until observation is available not be ldquoturned onrdquo until observation is available We plan to take the ensemble mean as first guess for next time step initial conditions
NASA Review (71007)
16
Noah LSM RUN AMSR-E SM EnKF Assimilation(TEST2)
Scaled AMSR-E SM
Expected Result [Example Only]EnKF Assimilation of AMSR-E SM Retrievals
Noah LSM RUN
EnKF Assimilation of Scaled AMSR-E SM RetrievalsEnKF Assimilation (TEST2)
Example
Only
NASA Review (71007)
17
Preliminary Evaluation of Soil Moisture Simulated by the Noah
Land Surface Model
Georgy Mostovoy
Geographical distribution of SCAN sites
OBJECTIVE Validation of the Noah Land Surface Model (LSM) baseline runsversus SCAN soil moisture observations
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
4
Soil Moisture Data Sources in this RPC Experiment
bull In-situ observationsndash USDA Soil Climate Analysis Network (SCAN)
bull Remotely sensed and estimatedndash NASA and JAXA Aqua Advanced Scanning
Microwave Radiometer ndash EOS (AMSR-E)
bull Numerical Modelsndash The Noah model in the NASA Land Information
System
NASA Review (71007)
5
USDA NRCS SCAN
NASA Review (71007)
6
Anticipated Societal Benefits
1 provides critical information to support drought monitoring and mitigation
2 provides essential information for predicting droughts based on weather and climate predictions
3 supports irrigation water management4 supports fire risk assessment5 supports water supply forecasting and NWS flood forecasting6 supplies a critical missing component to assist with snow climate
and associated hydrometeorological data analysis7 supports climate change assessment8 enables water quality monitoring9 supports a wide variety of natural resource management amp research
activities such as NASA remote sensing activities of soil moisture and ARS watershed studies
NASA Review (71007)
7
An Integrated Framework forLand Data Assimilation System
ApplicationsInputs OutputsPhysics
TopographySoils
WaterSupply ampDemand
AgricultureHydro-ElectricPower
EcologicalForecasting
Water Quality
ImprovedShort Term
ampLong TermPredictions
Land Cover and Vegetation (MODIS AMSRTRMM SRTM)
Meteorology Modeled amp
Observed (TRMM GOES Station)
Observed Land States(Snow ET Soil Moisture Water
Carbon etc)
Land Surface Models (LSM)Physical Process Models
Noah CLM VIC SiB2 Mosaic Catchment etc
Data Assimilation Modules(EnKF EKF)Rule-based
Water Fluxes Runoff
Surface States
Moisture Carbon Ts
Energy FluxesLe amp H
Biogeo-chemistry
Carbon Nitrogen etc
(Peters-Lidard Houser Kumar Tian Geiger)
NASA Review (71007)
8
LIS Evaluations Purpose and Activities
NASA Review (71007)
9
Purpose of RPC Evaluations hellip
bull Primaryndash Evaluate LIS capabilities and NASA data to enhance
and extend USDA-NRCS SCANbull Approach
ndash Evaluate LIS performancendash Assimilate SCAN and AMSR-E observations and
evaluate LIS capabilities to enhance SCAN by means of Observation Sensitivity Experiments (OSE)
ndash Derive physically consistent soil moisture maps at a range of spatial resolutions from 25x25 km2 to 1x1 km2
ndash Quantify uncertainties at all scales
NASA Review (71007)
10
Team Activity
bull MsState Project Management RPC Integration Control Run MODIS-VF [SSURGO]
bull NASA GSFC LIS Support AMSR-E data assimilation science expertise
bull GMU CREW SCAN data assimilation science expertise
NASA Review (71007)
11
Data Assimilation and Observation Sensitivity Experiments
bull Evaluation of data assimilation techniquesndash EKF EnKF
bull Data assimilation (land state)ndash Soil moisture
bull Soil moisture stationsbull AMSR-E
ndash Temperaturebull MODIS LST []
bull Sensitivity studiesbull Expected Outcomes high resolution soil moisture
analysis product uncertainty characterization
NASA Review (71007)
12
Status of Current Activities
bull Preliminary evaluation of simulated soil moisture data ndash Georgy Mostovoy
bull Quality Assessment of soil moisture measurements AMSR-E and SCAN - Anish Turlapaty
NASA Review (71007)
13
Future Directions
bull Assimilate AMSR-E soil moisture datandash Evaluate AMSR-E impacts
bull Incorporate MODIS Vegetation Fraction (VF) and compare with control runndash Evaluate MODIS VF impacts
bull Assimilate SCAN soil moisture datandash Evaluate SCAN impacts
NASA Review (71007)
14
ASMR-E Soil Moisture Data Assimilation and Evaluation
Noah Land Surface Model of NASA Land Information System
Soil Moisture Data
Soil Climate Analysis Network
AMSR-Eon NASA
AQUA Satellite
Evaluation Study
Soil Moisture Data
Soil Moisture Data
Soil Moisture Data
No D
A
EnKF DA
NASA Review (71007)
15
Future plansAssimilation of AMSR-E soil moisture data
12 hour time step 3 hourly output and 5 ensemble members
00Z 03Z 06Z 09Z 12Z 15Z 18Z 21Z 00Z
12 hr forecast+obs 12 hr forecast+obs
Data assimilation frequency will be twice daily at 06Z and 18Z DADA will will not be ldquoturned onrdquo until observation is available not be ldquoturned onrdquo until observation is available We plan to take the ensemble mean as first guess for next time step initial conditions
NASA Review (71007)
16
Noah LSM RUN AMSR-E SM EnKF Assimilation(TEST2)
Scaled AMSR-E SM
Expected Result [Example Only]EnKF Assimilation of AMSR-E SM Retrievals
Noah LSM RUN
EnKF Assimilation of Scaled AMSR-E SM RetrievalsEnKF Assimilation (TEST2)
Example
Only
NASA Review (71007)
17
Preliminary Evaluation of Soil Moisture Simulated by the Noah
Land Surface Model
Georgy Mostovoy
Geographical distribution of SCAN sites
OBJECTIVE Validation of the Noah Land Surface Model (LSM) baseline runsversus SCAN soil moisture observations
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
5
USDA NRCS SCAN
NASA Review (71007)
6
Anticipated Societal Benefits
1 provides critical information to support drought monitoring and mitigation
2 provides essential information for predicting droughts based on weather and climate predictions
3 supports irrigation water management4 supports fire risk assessment5 supports water supply forecasting and NWS flood forecasting6 supplies a critical missing component to assist with snow climate
and associated hydrometeorological data analysis7 supports climate change assessment8 enables water quality monitoring9 supports a wide variety of natural resource management amp research
activities such as NASA remote sensing activities of soil moisture and ARS watershed studies
NASA Review (71007)
7
An Integrated Framework forLand Data Assimilation System
ApplicationsInputs OutputsPhysics
TopographySoils
WaterSupply ampDemand
AgricultureHydro-ElectricPower
EcologicalForecasting
Water Quality
ImprovedShort Term
ampLong TermPredictions
Land Cover and Vegetation (MODIS AMSRTRMM SRTM)
Meteorology Modeled amp
Observed (TRMM GOES Station)
Observed Land States(Snow ET Soil Moisture Water
Carbon etc)
Land Surface Models (LSM)Physical Process Models
Noah CLM VIC SiB2 Mosaic Catchment etc
Data Assimilation Modules(EnKF EKF)Rule-based
Water Fluxes Runoff
Surface States
Moisture Carbon Ts
Energy FluxesLe amp H
Biogeo-chemistry
Carbon Nitrogen etc
(Peters-Lidard Houser Kumar Tian Geiger)
NASA Review (71007)
8
LIS Evaluations Purpose and Activities
NASA Review (71007)
9
Purpose of RPC Evaluations hellip
bull Primaryndash Evaluate LIS capabilities and NASA data to enhance
and extend USDA-NRCS SCANbull Approach
ndash Evaluate LIS performancendash Assimilate SCAN and AMSR-E observations and
evaluate LIS capabilities to enhance SCAN by means of Observation Sensitivity Experiments (OSE)
ndash Derive physically consistent soil moisture maps at a range of spatial resolutions from 25x25 km2 to 1x1 km2
ndash Quantify uncertainties at all scales
NASA Review (71007)
10
Team Activity
bull MsState Project Management RPC Integration Control Run MODIS-VF [SSURGO]
bull NASA GSFC LIS Support AMSR-E data assimilation science expertise
bull GMU CREW SCAN data assimilation science expertise
NASA Review (71007)
11
Data Assimilation and Observation Sensitivity Experiments
bull Evaluation of data assimilation techniquesndash EKF EnKF
bull Data assimilation (land state)ndash Soil moisture
bull Soil moisture stationsbull AMSR-E
ndash Temperaturebull MODIS LST []
bull Sensitivity studiesbull Expected Outcomes high resolution soil moisture
analysis product uncertainty characterization
NASA Review (71007)
12
Status of Current Activities
bull Preliminary evaluation of simulated soil moisture data ndash Georgy Mostovoy
bull Quality Assessment of soil moisture measurements AMSR-E and SCAN - Anish Turlapaty
NASA Review (71007)
13
Future Directions
bull Assimilate AMSR-E soil moisture datandash Evaluate AMSR-E impacts
bull Incorporate MODIS Vegetation Fraction (VF) and compare with control runndash Evaluate MODIS VF impacts
bull Assimilate SCAN soil moisture datandash Evaluate SCAN impacts
NASA Review (71007)
14
ASMR-E Soil Moisture Data Assimilation and Evaluation
Noah Land Surface Model of NASA Land Information System
Soil Moisture Data
Soil Climate Analysis Network
AMSR-Eon NASA
AQUA Satellite
Evaluation Study
Soil Moisture Data
Soil Moisture Data
Soil Moisture Data
No D
A
EnKF DA
NASA Review (71007)
15
Future plansAssimilation of AMSR-E soil moisture data
12 hour time step 3 hourly output and 5 ensemble members
00Z 03Z 06Z 09Z 12Z 15Z 18Z 21Z 00Z
12 hr forecast+obs 12 hr forecast+obs
Data assimilation frequency will be twice daily at 06Z and 18Z DADA will will not be ldquoturned onrdquo until observation is available not be ldquoturned onrdquo until observation is available We plan to take the ensemble mean as first guess for next time step initial conditions
NASA Review (71007)
16
Noah LSM RUN AMSR-E SM EnKF Assimilation(TEST2)
Scaled AMSR-E SM
Expected Result [Example Only]EnKF Assimilation of AMSR-E SM Retrievals
Noah LSM RUN
EnKF Assimilation of Scaled AMSR-E SM RetrievalsEnKF Assimilation (TEST2)
Example
Only
NASA Review (71007)
17
Preliminary Evaluation of Soil Moisture Simulated by the Noah
Land Surface Model
Georgy Mostovoy
Geographical distribution of SCAN sites
OBJECTIVE Validation of the Noah Land Surface Model (LSM) baseline runsversus SCAN soil moisture observations
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
6
Anticipated Societal Benefits
1 provides critical information to support drought monitoring and mitigation
2 provides essential information for predicting droughts based on weather and climate predictions
3 supports irrigation water management4 supports fire risk assessment5 supports water supply forecasting and NWS flood forecasting6 supplies a critical missing component to assist with snow climate
and associated hydrometeorological data analysis7 supports climate change assessment8 enables water quality monitoring9 supports a wide variety of natural resource management amp research
activities such as NASA remote sensing activities of soil moisture and ARS watershed studies
NASA Review (71007)
7
An Integrated Framework forLand Data Assimilation System
ApplicationsInputs OutputsPhysics
TopographySoils
WaterSupply ampDemand
AgricultureHydro-ElectricPower
EcologicalForecasting
Water Quality
ImprovedShort Term
ampLong TermPredictions
Land Cover and Vegetation (MODIS AMSRTRMM SRTM)
Meteorology Modeled amp
Observed (TRMM GOES Station)
Observed Land States(Snow ET Soil Moisture Water
Carbon etc)
Land Surface Models (LSM)Physical Process Models
Noah CLM VIC SiB2 Mosaic Catchment etc
Data Assimilation Modules(EnKF EKF)Rule-based
Water Fluxes Runoff
Surface States
Moisture Carbon Ts
Energy FluxesLe amp H
Biogeo-chemistry
Carbon Nitrogen etc
(Peters-Lidard Houser Kumar Tian Geiger)
NASA Review (71007)
8
LIS Evaluations Purpose and Activities
NASA Review (71007)
9
Purpose of RPC Evaluations hellip
bull Primaryndash Evaluate LIS capabilities and NASA data to enhance
and extend USDA-NRCS SCANbull Approach
ndash Evaluate LIS performancendash Assimilate SCAN and AMSR-E observations and
evaluate LIS capabilities to enhance SCAN by means of Observation Sensitivity Experiments (OSE)
ndash Derive physically consistent soil moisture maps at a range of spatial resolutions from 25x25 km2 to 1x1 km2
ndash Quantify uncertainties at all scales
NASA Review (71007)
10
Team Activity
bull MsState Project Management RPC Integration Control Run MODIS-VF [SSURGO]
bull NASA GSFC LIS Support AMSR-E data assimilation science expertise
bull GMU CREW SCAN data assimilation science expertise
NASA Review (71007)
11
Data Assimilation and Observation Sensitivity Experiments
bull Evaluation of data assimilation techniquesndash EKF EnKF
bull Data assimilation (land state)ndash Soil moisture
bull Soil moisture stationsbull AMSR-E
ndash Temperaturebull MODIS LST []
bull Sensitivity studiesbull Expected Outcomes high resolution soil moisture
analysis product uncertainty characterization
NASA Review (71007)
12
Status of Current Activities
bull Preliminary evaluation of simulated soil moisture data ndash Georgy Mostovoy
bull Quality Assessment of soil moisture measurements AMSR-E and SCAN - Anish Turlapaty
NASA Review (71007)
13
Future Directions
bull Assimilate AMSR-E soil moisture datandash Evaluate AMSR-E impacts
bull Incorporate MODIS Vegetation Fraction (VF) and compare with control runndash Evaluate MODIS VF impacts
bull Assimilate SCAN soil moisture datandash Evaluate SCAN impacts
NASA Review (71007)
14
ASMR-E Soil Moisture Data Assimilation and Evaluation
Noah Land Surface Model of NASA Land Information System
Soil Moisture Data
Soil Climate Analysis Network
AMSR-Eon NASA
AQUA Satellite
Evaluation Study
Soil Moisture Data
Soil Moisture Data
Soil Moisture Data
No D
A
EnKF DA
NASA Review (71007)
15
Future plansAssimilation of AMSR-E soil moisture data
12 hour time step 3 hourly output and 5 ensemble members
00Z 03Z 06Z 09Z 12Z 15Z 18Z 21Z 00Z
12 hr forecast+obs 12 hr forecast+obs
Data assimilation frequency will be twice daily at 06Z and 18Z DADA will will not be ldquoturned onrdquo until observation is available not be ldquoturned onrdquo until observation is available We plan to take the ensemble mean as first guess for next time step initial conditions
NASA Review (71007)
16
Noah LSM RUN AMSR-E SM EnKF Assimilation(TEST2)
Scaled AMSR-E SM
Expected Result [Example Only]EnKF Assimilation of AMSR-E SM Retrievals
Noah LSM RUN
EnKF Assimilation of Scaled AMSR-E SM RetrievalsEnKF Assimilation (TEST2)
Example
Only
NASA Review (71007)
17
Preliminary Evaluation of Soil Moisture Simulated by the Noah
Land Surface Model
Georgy Mostovoy
Geographical distribution of SCAN sites
OBJECTIVE Validation of the Noah Land Surface Model (LSM) baseline runsversus SCAN soil moisture observations
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
7
An Integrated Framework forLand Data Assimilation System
ApplicationsInputs OutputsPhysics
TopographySoils
WaterSupply ampDemand
AgricultureHydro-ElectricPower
EcologicalForecasting
Water Quality
ImprovedShort Term
ampLong TermPredictions
Land Cover and Vegetation (MODIS AMSRTRMM SRTM)
Meteorology Modeled amp
Observed (TRMM GOES Station)
Observed Land States(Snow ET Soil Moisture Water
Carbon etc)
Land Surface Models (LSM)Physical Process Models
Noah CLM VIC SiB2 Mosaic Catchment etc
Data Assimilation Modules(EnKF EKF)Rule-based
Water Fluxes Runoff
Surface States
Moisture Carbon Ts
Energy FluxesLe amp H
Biogeo-chemistry
Carbon Nitrogen etc
(Peters-Lidard Houser Kumar Tian Geiger)
NASA Review (71007)
8
LIS Evaluations Purpose and Activities
NASA Review (71007)
9
Purpose of RPC Evaluations hellip
bull Primaryndash Evaluate LIS capabilities and NASA data to enhance
and extend USDA-NRCS SCANbull Approach
ndash Evaluate LIS performancendash Assimilate SCAN and AMSR-E observations and
evaluate LIS capabilities to enhance SCAN by means of Observation Sensitivity Experiments (OSE)
ndash Derive physically consistent soil moisture maps at a range of spatial resolutions from 25x25 km2 to 1x1 km2
ndash Quantify uncertainties at all scales
NASA Review (71007)
10
Team Activity
bull MsState Project Management RPC Integration Control Run MODIS-VF [SSURGO]
bull NASA GSFC LIS Support AMSR-E data assimilation science expertise
bull GMU CREW SCAN data assimilation science expertise
NASA Review (71007)
11
Data Assimilation and Observation Sensitivity Experiments
bull Evaluation of data assimilation techniquesndash EKF EnKF
bull Data assimilation (land state)ndash Soil moisture
bull Soil moisture stationsbull AMSR-E
ndash Temperaturebull MODIS LST []
bull Sensitivity studiesbull Expected Outcomes high resolution soil moisture
analysis product uncertainty characterization
NASA Review (71007)
12
Status of Current Activities
bull Preliminary evaluation of simulated soil moisture data ndash Georgy Mostovoy
bull Quality Assessment of soil moisture measurements AMSR-E and SCAN - Anish Turlapaty
NASA Review (71007)
13
Future Directions
bull Assimilate AMSR-E soil moisture datandash Evaluate AMSR-E impacts
bull Incorporate MODIS Vegetation Fraction (VF) and compare with control runndash Evaluate MODIS VF impacts
bull Assimilate SCAN soil moisture datandash Evaluate SCAN impacts
NASA Review (71007)
14
ASMR-E Soil Moisture Data Assimilation and Evaluation
Noah Land Surface Model of NASA Land Information System
Soil Moisture Data
Soil Climate Analysis Network
AMSR-Eon NASA
AQUA Satellite
Evaluation Study
Soil Moisture Data
Soil Moisture Data
Soil Moisture Data
No D
A
EnKF DA
NASA Review (71007)
15
Future plansAssimilation of AMSR-E soil moisture data
12 hour time step 3 hourly output and 5 ensemble members
00Z 03Z 06Z 09Z 12Z 15Z 18Z 21Z 00Z
12 hr forecast+obs 12 hr forecast+obs
Data assimilation frequency will be twice daily at 06Z and 18Z DADA will will not be ldquoturned onrdquo until observation is available not be ldquoturned onrdquo until observation is available We plan to take the ensemble mean as first guess for next time step initial conditions
NASA Review (71007)
16
Noah LSM RUN AMSR-E SM EnKF Assimilation(TEST2)
Scaled AMSR-E SM
Expected Result [Example Only]EnKF Assimilation of AMSR-E SM Retrievals
Noah LSM RUN
EnKF Assimilation of Scaled AMSR-E SM RetrievalsEnKF Assimilation (TEST2)
Example
Only
NASA Review (71007)
17
Preliminary Evaluation of Soil Moisture Simulated by the Noah
Land Surface Model
Georgy Mostovoy
Geographical distribution of SCAN sites
OBJECTIVE Validation of the Noah Land Surface Model (LSM) baseline runsversus SCAN soil moisture observations
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
8
LIS Evaluations Purpose and Activities
NASA Review (71007)
9
Purpose of RPC Evaluations hellip
bull Primaryndash Evaluate LIS capabilities and NASA data to enhance
and extend USDA-NRCS SCANbull Approach
ndash Evaluate LIS performancendash Assimilate SCAN and AMSR-E observations and
evaluate LIS capabilities to enhance SCAN by means of Observation Sensitivity Experiments (OSE)
ndash Derive physically consistent soil moisture maps at a range of spatial resolutions from 25x25 km2 to 1x1 km2
ndash Quantify uncertainties at all scales
NASA Review (71007)
10
Team Activity
bull MsState Project Management RPC Integration Control Run MODIS-VF [SSURGO]
bull NASA GSFC LIS Support AMSR-E data assimilation science expertise
bull GMU CREW SCAN data assimilation science expertise
NASA Review (71007)
11
Data Assimilation and Observation Sensitivity Experiments
bull Evaluation of data assimilation techniquesndash EKF EnKF
bull Data assimilation (land state)ndash Soil moisture
bull Soil moisture stationsbull AMSR-E
ndash Temperaturebull MODIS LST []
bull Sensitivity studiesbull Expected Outcomes high resolution soil moisture
analysis product uncertainty characterization
NASA Review (71007)
12
Status of Current Activities
bull Preliminary evaluation of simulated soil moisture data ndash Georgy Mostovoy
bull Quality Assessment of soil moisture measurements AMSR-E and SCAN - Anish Turlapaty
NASA Review (71007)
13
Future Directions
bull Assimilate AMSR-E soil moisture datandash Evaluate AMSR-E impacts
bull Incorporate MODIS Vegetation Fraction (VF) and compare with control runndash Evaluate MODIS VF impacts
bull Assimilate SCAN soil moisture datandash Evaluate SCAN impacts
NASA Review (71007)
14
ASMR-E Soil Moisture Data Assimilation and Evaluation
Noah Land Surface Model of NASA Land Information System
Soil Moisture Data
Soil Climate Analysis Network
AMSR-Eon NASA
AQUA Satellite
Evaluation Study
Soil Moisture Data
Soil Moisture Data
Soil Moisture Data
No D
A
EnKF DA
NASA Review (71007)
15
Future plansAssimilation of AMSR-E soil moisture data
12 hour time step 3 hourly output and 5 ensemble members
00Z 03Z 06Z 09Z 12Z 15Z 18Z 21Z 00Z
12 hr forecast+obs 12 hr forecast+obs
Data assimilation frequency will be twice daily at 06Z and 18Z DADA will will not be ldquoturned onrdquo until observation is available not be ldquoturned onrdquo until observation is available We plan to take the ensemble mean as first guess for next time step initial conditions
NASA Review (71007)
16
Noah LSM RUN AMSR-E SM EnKF Assimilation(TEST2)
Scaled AMSR-E SM
Expected Result [Example Only]EnKF Assimilation of AMSR-E SM Retrievals
Noah LSM RUN
EnKF Assimilation of Scaled AMSR-E SM RetrievalsEnKF Assimilation (TEST2)
Example
Only
NASA Review (71007)
17
Preliminary Evaluation of Soil Moisture Simulated by the Noah
Land Surface Model
Georgy Mostovoy
Geographical distribution of SCAN sites
OBJECTIVE Validation of the Noah Land Surface Model (LSM) baseline runsversus SCAN soil moisture observations
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
9
Purpose of RPC Evaluations hellip
bull Primaryndash Evaluate LIS capabilities and NASA data to enhance
and extend USDA-NRCS SCANbull Approach
ndash Evaluate LIS performancendash Assimilate SCAN and AMSR-E observations and
evaluate LIS capabilities to enhance SCAN by means of Observation Sensitivity Experiments (OSE)
ndash Derive physically consistent soil moisture maps at a range of spatial resolutions from 25x25 km2 to 1x1 km2
ndash Quantify uncertainties at all scales
NASA Review (71007)
10
Team Activity
bull MsState Project Management RPC Integration Control Run MODIS-VF [SSURGO]
bull NASA GSFC LIS Support AMSR-E data assimilation science expertise
bull GMU CREW SCAN data assimilation science expertise
NASA Review (71007)
11
Data Assimilation and Observation Sensitivity Experiments
bull Evaluation of data assimilation techniquesndash EKF EnKF
bull Data assimilation (land state)ndash Soil moisture
bull Soil moisture stationsbull AMSR-E
ndash Temperaturebull MODIS LST []
bull Sensitivity studiesbull Expected Outcomes high resolution soil moisture
analysis product uncertainty characterization
NASA Review (71007)
12
Status of Current Activities
bull Preliminary evaluation of simulated soil moisture data ndash Georgy Mostovoy
bull Quality Assessment of soil moisture measurements AMSR-E and SCAN - Anish Turlapaty
NASA Review (71007)
13
Future Directions
bull Assimilate AMSR-E soil moisture datandash Evaluate AMSR-E impacts
bull Incorporate MODIS Vegetation Fraction (VF) and compare with control runndash Evaluate MODIS VF impacts
bull Assimilate SCAN soil moisture datandash Evaluate SCAN impacts
NASA Review (71007)
14
ASMR-E Soil Moisture Data Assimilation and Evaluation
Noah Land Surface Model of NASA Land Information System
Soil Moisture Data
Soil Climate Analysis Network
AMSR-Eon NASA
AQUA Satellite
Evaluation Study
Soil Moisture Data
Soil Moisture Data
Soil Moisture Data
No D
A
EnKF DA
NASA Review (71007)
15
Future plansAssimilation of AMSR-E soil moisture data
12 hour time step 3 hourly output and 5 ensemble members
00Z 03Z 06Z 09Z 12Z 15Z 18Z 21Z 00Z
12 hr forecast+obs 12 hr forecast+obs
Data assimilation frequency will be twice daily at 06Z and 18Z DADA will will not be ldquoturned onrdquo until observation is available not be ldquoturned onrdquo until observation is available We plan to take the ensemble mean as first guess for next time step initial conditions
NASA Review (71007)
16
Noah LSM RUN AMSR-E SM EnKF Assimilation(TEST2)
Scaled AMSR-E SM
Expected Result [Example Only]EnKF Assimilation of AMSR-E SM Retrievals
Noah LSM RUN
EnKF Assimilation of Scaled AMSR-E SM RetrievalsEnKF Assimilation (TEST2)
Example
Only
NASA Review (71007)
17
Preliminary Evaluation of Soil Moisture Simulated by the Noah
Land Surface Model
Georgy Mostovoy
Geographical distribution of SCAN sites
OBJECTIVE Validation of the Noah Land Surface Model (LSM) baseline runsversus SCAN soil moisture observations
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
10
Team Activity
bull MsState Project Management RPC Integration Control Run MODIS-VF [SSURGO]
bull NASA GSFC LIS Support AMSR-E data assimilation science expertise
bull GMU CREW SCAN data assimilation science expertise
NASA Review (71007)
11
Data Assimilation and Observation Sensitivity Experiments
bull Evaluation of data assimilation techniquesndash EKF EnKF
bull Data assimilation (land state)ndash Soil moisture
bull Soil moisture stationsbull AMSR-E
ndash Temperaturebull MODIS LST []
bull Sensitivity studiesbull Expected Outcomes high resolution soil moisture
analysis product uncertainty characterization
NASA Review (71007)
12
Status of Current Activities
bull Preliminary evaluation of simulated soil moisture data ndash Georgy Mostovoy
bull Quality Assessment of soil moisture measurements AMSR-E and SCAN - Anish Turlapaty
NASA Review (71007)
13
Future Directions
bull Assimilate AMSR-E soil moisture datandash Evaluate AMSR-E impacts
bull Incorporate MODIS Vegetation Fraction (VF) and compare with control runndash Evaluate MODIS VF impacts
bull Assimilate SCAN soil moisture datandash Evaluate SCAN impacts
NASA Review (71007)
14
ASMR-E Soil Moisture Data Assimilation and Evaluation
Noah Land Surface Model of NASA Land Information System
Soil Moisture Data
Soil Climate Analysis Network
AMSR-Eon NASA
AQUA Satellite
Evaluation Study
Soil Moisture Data
Soil Moisture Data
Soil Moisture Data
No D
A
EnKF DA
NASA Review (71007)
15
Future plansAssimilation of AMSR-E soil moisture data
12 hour time step 3 hourly output and 5 ensemble members
00Z 03Z 06Z 09Z 12Z 15Z 18Z 21Z 00Z
12 hr forecast+obs 12 hr forecast+obs
Data assimilation frequency will be twice daily at 06Z and 18Z DADA will will not be ldquoturned onrdquo until observation is available not be ldquoturned onrdquo until observation is available We plan to take the ensemble mean as first guess for next time step initial conditions
NASA Review (71007)
16
Noah LSM RUN AMSR-E SM EnKF Assimilation(TEST2)
Scaled AMSR-E SM
Expected Result [Example Only]EnKF Assimilation of AMSR-E SM Retrievals
Noah LSM RUN
EnKF Assimilation of Scaled AMSR-E SM RetrievalsEnKF Assimilation (TEST2)
Example
Only
NASA Review (71007)
17
Preliminary Evaluation of Soil Moisture Simulated by the Noah
Land Surface Model
Georgy Mostovoy
Geographical distribution of SCAN sites
OBJECTIVE Validation of the Noah Land Surface Model (LSM) baseline runsversus SCAN soil moisture observations
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
11
Data Assimilation and Observation Sensitivity Experiments
bull Evaluation of data assimilation techniquesndash EKF EnKF
bull Data assimilation (land state)ndash Soil moisture
bull Soil moisture stationsbull AMSR-E
ndash Temperaturebull MODIS LST []
bull Sensitivity studiesbull Expected Outcomes high resolution soil moisture
analysis product uncertainty characterization
NASA Review (71007)
12
Status of Current Activities
bull Preliminary evaluation of simulated soil moisture data ndash Georgy Mostovoy
bull Quality Assessment of soil moisture measurements AMSR-E and SCAN - Anish Turlapaty
NASA Review (71007)
13
Future Directions
bull Assimilate AMSR-E soil moisture datandash Evaluate AMSR-E impacts
bull Incorporate MODIS Vegetation Fraction (VF) and compare with control runndash Evaluate MODIS VF impacts
bull Assimilate SCAN soil moisture datandash Evaluate SCAN impacts
NASA Review (71007)
14
ASMR-E Soil Moisture Data Assimilation and Evaluation
Noah Land Surface Model of NASA Land Information System
Soil Moisture Data
Soil Climate Analysis Network
AMSR-Eon NASA
AQUA Satellite
Evaluation Study
Soil Moisture Data
Soil Moisture Data
Soil Moisture Data
No D
A
EnKF DA
NASA Review (71007)
15
Future plansAssimilation of AMSR-E soil moisture data
12 hour time step 3 hourly output and 5 ensemble members
00Z 03Z 06Z 09Z 12Z 15Z 18Z 21Z 00Z
12 hr forecast+obs 12 hr forecast+obs
Data assimilation frequency will be twice daily at 06Z and 18Z DADA will will not be ldquoturned onrdquo until observation is available not be ldquoturned onrdquo until observation is available We plan to take the ensemble mean as first guess for next time step initial conditions
NASA Review (71007)
16
Noah LSM RUN AMSR-E SM EnKF Assimilation(TEST2)
Scaled AMSR-E SM
Expected Result [Example Only]EnKF Assimilation of AMSR-E SM Retrievals
Noah LSM RUN
EnKF Assimilation of Scaled AMSR-E SM RetrievalsEnKF Assimilation (TEST2)
Example
Only
NASA Review (71007)
17
Preliminary Evaluation of Soil Moisture Simulated by the Noah
Land Surface Model
Georgy Mostovoy
Geographical distribution of SCAN sites
OBJECTIVE Validation of the Noah Land Surface Model (LSM) baseline runsversus SCAN soil moisture observations
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
12
Status of Current Activities
bull Preliminary evaluation of simulated soil moisture data ndash Georgy Mostovoy
bull Quality Assessment of soil moisture measurements AMSR-E and SCAN - Anish Turlapaty
NASA Review (71007)
13
Future Directions
bull Assimilate AMSR-E soil moisture datandash Evaluate AMSR-E impacts
bull Incorporate MODIS Vegetation Fraction (VF) and compare with control runndash Evaluate MODIS VF impacts
bull Assimilate SCAN soil moisture datandash Evaluate SCAN impacts
NASA Review (71007)
14
ASMR-E Soil Moisture Data Assimilation and Evaluation
Noah Land Surface Model of NASA Land Information System
Soil Moisture Data
Soil Climate Analysis Network
AMSR-Eon NASA
AQUA Satellite
Evaluation Study
Soil Moisture Data
Soil Moisture Data
Soil Moisture Data
No D
A
EnKF DA
NASA Review (71007)
15
Future plansAssimilation of AMSR-E soil moisture data
12 hour time step 3 hourly output and 5 ensemble members
00Z 03Z 06Z 09Z 12Z 15Z 18Z 21Z 00Z
12 hr forecast+obs 12 hr forecast+obs
Data assimilation frequency will be twice daily at 06Z and 18Z DADA will will not be ldquoturned onrdquo until observation is available not be ldquoturned onrdquo until observation is available We plan to take the ensemble mean as first guess for next time step initial conditions
NASA Review (71007)
16
Noah LSM RUN AMSR-E SM EnKF Assimilation(TEST2)
Scaled AMSR-E SM
Expected Result [Example Only]EnKF Assimilation of AMSR-E SM Retrievals
Noah LSM RUN
EnKF Assimilation of Scaled AMSR-E SM RetrievalsEnKF Assimilation (TEST2)
Example
Only
NASA Review (71007)
17
Preliminary Evaluation of Soil Moisture Simulated by the Noah
Land Surface Model
Georgy Mostovoy
Geographical distribution of SCAN sites
OBJECTIVE Validation of the Noah Land Surface Model (LSM) baseline runsversus SCAN soil moisture observations
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
13
Future Directions
bull Assimilate AMSR-E soil moisture datandash Evaluate AMSR-E impacts
bull Incorporate MODIS Vegetation Fraction (VF) and compare with control runndash Evaluate MODIS VF impacts
bull Assimilate SCAN soil moisture datandash Evaluate SCAN impacts
NASA Review (71007)
14
ASMR-E Soil Moisture Data Assimilation and Evaluation
Noah Land Surface Model of NASA Land Information System
Soil Moisture Data
Soil Climate Analysis Network
AMSR-Eon NASA
AQUA Satellite
Evaluation Study
Soil Moisture Data
Soil Moisture Data
Soil Moisture Data
No D
A
EnKF DA
NASA Review (71007)
15
Future plansAssimilation of AMSR-E soil moisture data
12 hour time step 3 hourly output and 5 ensemble members
00Z 03Z 06Z 09Z 12Z 15Z 18Z 21Z 00Z
12 hr forecast+obs 12 hr forecast+obs
Data assimilation frequency will be twice daily at 06Z and 18Z DADA will will not be ldquoturned onrdquo until observation is available not be ldquoturned onrdquo until observation is available We plan to take the ensemble mean as first guess for next time step initial conditions
NASA Review (71007)
16
Noah LSM RUN AMSR-E SM EnKF Assimilation(TEST2)
Scaled AMSR-E SM
Expected Result [Example Only]EnKF Assimilation of AMSR-E SM Retrievals
Noah LSM RUN
EnKF Assimilation of Scaled AMSR-E SM RetrievalsEnKF Assimilation (TEST2)
Example
Only
NASA Review (71007)
17
Preliminary Evaluation of Soil Moisture Simulated by the Noah
Land Surface Model
Georgy Mostovoy
Geographical distribution of SCAN sites
OBJECTIVE Validation of the Noah Land Surface Model (LSM) baseline runsversus SCAN soil moisture observations
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
14
ASMR-E Soil Moisture Data Assimilation and Evaluation
Noah Land Surface Model of NASA Land Information System
Soil Moisture Data
Soil Climate Analysis Network
AMSR-Eon NASA
AQUA Satellite
Evaluation Study
Soil Moisture Data
Soil Moisture Data
Soil Moisture Data
No D
A
EnKF DA
NASA Review (71007)
15
Future plansAssimilation of AMSR-E soil moisture data
12 hour time step 3 hourly output and 5 ensemble members
00Z 03Z 06Z 09Z 12Z 15Z 18Z 21Z 00Z
12 hr forecast+obs 12 hr forecast+obs
Data assimilation frequency will be twice daily at 06Z and 18Z DADA will will not be ldquoturned onrdquo until observation is available not be ldquoturned onrdquo until observation is available We plan to take the ensemble mean as first guess for next time step initial conditions
NASA Review (71007)
16
Noah LSM RUN AMSR-E SM EnKF Assimilation(TEST2)
Scaled AMSR-E SM
Expected Result [Example Only]EnKF Assimilation of AMSR-E SM Retrievals
Noah LSM RUN
EnKF Assimilation of Scaled AMSR-E SM RetrievalsEnKF Assimilation (TEST2)
Example
Only
NASA Review (71007)
17
Preliminary Evaluation of Soil Moisture Simulated by the Noah
Land Surface Model
Georgy Mostovoy
Geographical distribution of SCAN sites
OBJECTIVE Validation of the Noah Land Surface Model (LSM) baseline runsversus SCAN soil moisture observations
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
15
Future plansAssimilation of AMSR-E soil moisture data
12 hour time step 3 hourly output and 5 ensemble members
00Z 03Z 06Z 09Z 12Z 15Z 18Z 21Z 00Z
12 hr forecast+obs 12 hr forecast+obs
Data assimilation frequency will be twice daily at 06Z and 18Z DADA will will not be ldquoturned onrdquo until observation is available not be ldquoturned onrdquo until observation is available We plan to take the ensemble mean as first guess for next time step initial conditions
NASA Review (71007)
16
Noah LSM RUN AMSR-E SM EnKF Assimilation(TEST2)
Scaled AMSR-E SM
Expected Result [Example Only]EnKF Assimilation of AMSR-E SM Retrievals
Noah LSM RUN
EnKF Assimilation of Scaled AMSR-E SM RetrievalsEnKF Assimilation (TEST2)
Example
Only
NASA Review (71007)
17
Preliminary Evaluation of Soil Moisture Simulated by the Noah
Land Surface Model
Georgy Mostovoy
Geographical distribution of SCAN sites
OBJECTIVE Validation of the Noah Land Surface Model (LSM) baseline runsversus SCAN soil moisture observations
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
16
Noah LSM RUN AMSR-E SM EnKF Assimilation(TEST2)
Scaled AMSR-E SM
Expected Result [Example Only]EnKF Assimilation of AMSR-E SM Retrievals
Noah LSM RUN
EnKF Assimilation of Scaled AMSR-E SM RetrievalsEnKF Assimilation (TEST2)
Example
Only
NASA Review (71007)
17
Preliminary Evaluation of Soil Moisture Simulated by the Noah
Land Surface Model
Georgy Mostovoy
Geographical distribution of SCAN sites
OBJECTIVE Validation of the Noah Land Surface Model (LSM) baseline runsversus SCAN soil moisture observations
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
17
Preliminary Evaluation of Soil Moisture Simulated by the Noah
Land Surface Model
Georgy Mostovoy
Geographical distribution of SCAN sites
OBJECTIVE Validation of the Noah Land Surface Model (LSM) baseline runsversus SCAN soil moisture observations
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
Geographical distribution of SCAN sites
OBJECTIVE Validation of the Noah Land Surface Model (LSM) baseline runsversus SCAN soil moisture observations
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
P ndash dominant (across var seasons) positive bias (high temporal variability of SM drying is rather rapid)N ndash dominant negative bias (SM exhibits low variability sluggish behavior in comparison with other sites)0 ndash zero bias
P
P
P
P
P
P
N
N
N
N
0
0
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
Silver City MS Marianna AR
a flat terrain prevails
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
DPEt
w
E - evaporation rate (calculated based on air 2-m T q P and wind speed from NLDAS)P ndash precipitation rate (prescribed from the NLDAS data)D ndash exchange rate with adjacent soil layers (calculated based on soil type and w)
Total water content (w) within a soil layer of an arbitrary depth (10 cm 1 m or 2 m for example)
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
DtwDtEww ttt )1(11
1 Analogy with AR(1) process or the Markov chain
Considering a drying stage (P = 0)
where 1 twE
and α is evaporation efficiency
)1()( ttR is the autocorrelation functionvalue for the time lag Δt
For an arbitrary time t (t = n Δt n = 01 2 hellip) the autocorrelation function is defined as follows
)()(SMT
tEXPtR
))1(1(ln t
tTSM
is the integral correlation scale which defines the soil moisture ldquomemoryrdquo
Relationship between the correlation scale and evaporation efficiency (Δt = 1 day was used)
Given this simple model the evaporation term controls the soil moisture memory
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
DPEt
w
)(
2 An equation for the soil moisture error δw
An accumulated soil moisture error for the time period T can be written as follows
TTT
T DPEw000
)(
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
Persistency (ldquomemoryrdquo) of soil moisture initial anomalies during Fall 2005
1 Weak memory case an initial positive anomaly between two SCAN sites disappears after about 40 days
2 Strong memory case an initial positive anomaly persists and amplifies during 40-days
bull Both cases suggest that local factors such as soil physical properties water table etc control dynamics of soil moisture anomalies (deviations from a regionally-mean)
bull This also implies a little control of precipitation on the initial soil moisture state and its dynamics over the Mississippi Delta Region(Maximum soil moisture values are bounded by the field capacity = the soil porosity)
bull Sub-monthly time scales are considered (2-3 weeks periods)
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
Observed soil moisture (averaged over 12 SCAN sites) evolution at different depths during Fall 2005
error bars stand for standard deviation (SD)
Low SD
HighSD
Wet state -gt low SD
Dry state -gt high SD
Example of soil moisture comparison (averaged over 12 SCAN sites) between SCAN and simulated
by Noah
SM underestimation
O v e r e s t i m a t i o n(evaporation deficiency of the Noah model)
Precipitation event
Drying out
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
Outline for baseline soil moisture simulations over the MS Delta region (I)
Noah Land Surface Model (v 271) available from LIS (v 431) was used for retrospective runs (years 2005 and 2006)
bull 1-km domain size 256x256 points (255x255 latitude-longitude)
North American Land Data Assimilation System (NLDAS Cosgrove et al 2003)
atmospheric forcing was used (specified at approx 15-km grid)
1-km 5-km and 15-km horizontal grid for the Noah model runs
(no substantial difference in 5-cm soil moisture content between 1-km 5-km and 15-km runs was
observed)
Statsgo Soil Data
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
Outline for baseline soil moisture simulations over the MS Delta region (II)
One year (2004) spin-up period was used for the Noah model
bull Soil moisture observations available from 12 SCAN sites (subjective QC visual inspection of
plots) were used for validation of the baseline simulations (daily-
mean values of SM were compared)
bull Frequency distributions of soil moisture and precipitation
errors were plotted for two-month periods (March-April May-June July-August Sept-Oct)
spanning years 2005 and 2006
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
Gap and scale change in the data
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
May-June 2005
P
P
PP
PP
0
P ndash dominant (across var seasons) positive biasN ndash dominant negative bias0 ndash zero bias
N
N N
0
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
May-June 2006
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
Sept-Oct 2005
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
Sept-Oct 2006
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
March-April 2005
Frequency distribution of soil moisture content error (5-cm daily-mean value simulated by Noah model at 1-km grid minus observed at SCAN site) and difference between NLDAS precipitation forcing
and observed local values at SCAN sites
Numbers in upper right corners stand for total number of precipitation events registered either by observations or by NLDAS data (upper row) within two-month period mean difference or bias and standard deviation between simulated and observed soil moisture (middle and lower rows respectively)
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
No apparent relationship was found between precipitation and 5-cm SM biases (measured as simulated minus observed value) from examination of error distribution histograms
Left frame shows a scatterplot between simulated minus observed maximum SM within each two-month period and a corresponding precipitation error bias The scatterplot suggests no significant linear relationship (a correlation coefficient R = -012 is not significant) between the precipitation and SM errors for maximum values of SM
Error sensitivity of simulated maximum 5-cm SM content to error of the NLDAS
precipitation forcing
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
No apparent association between soil texture (and its variability with depth presented in the left lower frame) and a sign of the SM bias was observed Conversely horizontal distribution patterns of simulated SM is controlled by corresponding patterns of the soil texture map (figures not presented)
Because of relatively high persistence (across different years and months) of a bias sign at a particular scan site it is possible to stratify all scan sites into three category (with significant positive negative and smallzero bias) according to this sign Specifically six sites (N Issaquena Perthshire Farm Tunica Earle Campus and Lonoke Farm) demonstrated rather persistent positive SM bias At four sites such as Beasley Lake Vance Marianna and DeWitt) a negative SM bias was dominant Two sites (Silver City and Good Timber Creek) showed a relatively small SM bias
Both positive and negative significant SM biases occurred mostly during drying stages of soilsrsquo matter This fact suggests that an accurate descriptionspecification of other factors (such as upperlower boundary conditions for the SM atmospheric evaporation etc) in addition to precipitation forcing is critical for reduction of the SM error
Soil texture
Soil texture vertical heterogeneity
(numbers indicate scan sites)
Dominant positive SM bias ndash dotted lines
Dominant negative or ldquozerordquo ndash solid lines
4 ndash Silt Loam 8 ndash Silty Clay Loam 9 ndash Clay Loam11 ndash Silty Clay
Local samples versus Statsgo data
Impact on 5-cm SM bias
Increase of clay content
Decr
ease
of
sand
con
ten
t w
ith d
ep
th
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
Soil texture variations with depth Seasonality of soil moisture biases at 5-cm depth
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
Summary
o The Noah model performance was evaluated against soil moisture observations at 12 sites during the years 2005 and 2006
o The Noah model demonstrates a reasonable skill over the Mississippi Delta Region with a typical bias of 5 The Noah model produces an overestimation during drying out periods This deficiency might be fixed by adding the assimilation capability of soil moisture observations (SCAN and AMSR-E) into the Noah model
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
38
Quality Assessment of AMSR-E Soil Moisture Data
Anish Turlapaty
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
39
PROBLEM DESCRIPTION
AMSR-E
Noah Land Surface Model of
NASA Land Information
System
Soil Moisture Data
Assimilation
Data Validation Before data assimilation AMSR-E soil moisture data product has to be validated
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
40
GENERAL APPROACH
bull Test Data Soil moisture data is collected from AMSR-E for the years 2005 2006 for Mississippi and Arkansas
bull Training Data For validation purposes soil moisture data from 20 locations of SCAN network in Mississippi and Arkansas is used
bull Quality control tool One class support vector machines which provide a quality value for each time series
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
RESULTS Quality Map
SVM method
Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
Quality Maps Contd
QC values are assigned at each pixel (28x23)Invalid data
1
Poor data2
Marginal quality
3
Marginal quality
4
Good quality data
5
Remarks on Quality
Quality Level
Mahalanobis Method
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
43
SUMMARY
Quality maps are developed for Mississippi and Arkansas which show the quality of time series at each pixel on scale of five to one
These results are compared with quality map from Mahalanobis method
Currently we are looking for a conventional quality control tool with which these results can be verified
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
44
Questions
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135
NASA Review (71007)
45
Contact Information
Valentine Anantharajltvalgrimsstateedugt
Tel (662)325-5135