WDC for Glaciology, Boulder – 30 th Anniversary Workshop Boulder, Colorado 25 October 2006...

WDC for Glaciology, Boulder – 30th Anniversary WorkshopBoulder, Colorado 25 October 2006

Multi-Sensor Snow Mapping and Global Trends

R. L. Armstrong, M.J. Brodzik, M. Savoie and K. KnowlesUniversity of Colorado, Boulder, USA


Outline

Legacy Data Sets for Hemispheric-Scale Snow Mapping

Cross-calibration of SMMR and SSM/I

Comparison of Optical and Microwave Snow Cover Time Series

Snow Mapping using Blended Optical and Passive Microwave Trend Analysis


NOAA IMS – Interactive Multisensor Snow and Ice Mapping System – http://www.ssd.noaa.gov/PS/SNOW

Source for visible data: NOAA Weekly/Daily Snow Charts since 1966Longest time-series of any parameter derived from satellite remote sensing


SMMR, SSM/I &AMSR-E Brightness Temperatures 1978-2006

EASE-Grid

(Equal-Area Scalable Earth Grid)

Full Global, North and South Hem. Projections

NOAA/NASA Pathfinder Level 3 Gridded Data

Available from NSIDC: http://nsidc.org

Source for passive microwave data


Northern Hemisphere Satellite-Derived Snow Extent1978 – 2006 Visible (NOAA) Passive Microwave (SMMR & SSM/I)

Analysis of the microwave data is complicated by the change in sensors from SMMR to SSM/I

microwave (green/blue) vs. visible (pink) sensors,


Sensors are similar but exhibit important differences in spatial and temporal coverage that affect the sampling density within the long-term record.

Daily passive microwave 37 GHz, horizontally polarized brightness temperatures,July 19, 1987, showing smaller coverage area of SMMR (left) vs. SSM/I (right).

SMMR SSM/I


SMMR-SSM/I Sensor Cross-Calibration Issues

• The SMMR swath width was about half that of SSM/I, further reducing the probability of overlapping observations.

• Sensor overlap occurred in July and August 1987 (Northern Hemisphere summer, essentially no seasonal snow cover).

• The overlap is about 40 days, effectively 20 days of data, since SMMR was cycled on/off on a daily basis.

• Low latitude sites may require up to 5 or 6 days for one SMMR overpass


Time series of Tibetan Plateau snow-covered area and average snow water equivalent (SWE) derived from passive microwave sensors, 1978-2004.

snow extent

snow waterequivalent

Tibet Plateau


Time series of 36/37 GHz, vertically-polarized “cold pass” brightness temperatures at Dome C (Antarctica) from SMMR (1978-1987, orange), various SSM/Is (1987-2006, pink/brown/green), and AMSR-E (2002-2006, blue). Note cross-sensor calibration issues, indicated by higher mean and compressed annual amplitude in SMMR data, compared to SSM/I and AMSR-E.

Time Series of 36/37 GHz at Dome C Antarctica, SMMR-SSM/I-AMSR-E


Sensor Cross-Calibration: Locating Suitable Calibration Targets over Land

A sub-array ‘footprint’ area was defined as a set of 3 x 3 grid cells (approximately 75 km x 75 km) for a given day, with indices j, centered at a particular grid point

The following statistics were considered:

Regions analyzed consisted of 100 x 100 arrays of 25-km EASE-Grid cells centered at 1.5°S, 21.6°E (Salonga, Zaire) and 75.0°S, 123.3°E (Dome C, Antarctica)

Over land, the spatial homogeneity and temporal stability of brightness temperatures (Tb) over selected regions such as tropical forest and ice sheet regions were examined

mi 1

NTbij

j1

N

; i 1

N 1(Tbij mi )

2

j1

N

m 1

ND

mii1

N D

; 1

ND

ii1

N D

m 1

ND 1(mi m )2

i1

N D

+

+

Spatial mean and standard deviation of Tb within a footprint on day i:

Temporal means (over one year) of mi and i:

Temporal standard deviation of mi over one year:

Select land targets with surface characteristics that represent the cold through warm microwave emission range. The final targets are chosen for temporal and spatial stability. This is done by wayof statistical analysis within a moving 3x3 array of pixels to determine the specific locations withminimal spatial and temporal variability.


Tropical Forest - Salonga, Zaire

Statistics for the year 2003 at Salonga, Zaire, for 10 GHz and 37 GHz vertical polarizations, descending passes: (a) Foot print mean, mi ; (b) Foot print standard deviation, i. Units are in Kelvins.

mi

i

m

m

Method to identify location of max. spatial and temporal stability - Tb statistics,central African tropical forest.

(a) Upper left: (K). (b) Upper right: (K). (c) Lower left: (K). (d) Lower right: Surface topography (meters above sea level).

m

m

Avg Daily Footprint Mean Stddev of Daily Footprint Mean (time)

Avg Daily Footprint Stddev (space) Avg Footprint Elevation


Scatter plots of SSM/I vs. SMMR brightness temperatures (19/18 GHz, left, and 37 GHz, right) at Earth targets selected for spatial stability (Dome C (Antarctic ice sheet), Salonga (African tropical forest), Canada (plains), Summit (Greenland ice sheet)).The large plus sign in each plot represents the typical range (+/- 1 standard deviation) of wintertime brightness temperatures in seasonally snow-covered regions.

Scatter plots of SSM/I vs. SMMR brightness temperature at stable targets


Comparison of NASA Aqua AMSR-E (36 GHz) and DMSP SSM/I (37 GHz) Dome-C Antarctica, June 2002 to December 2005.

SSM/I vs. AMSR-E


(50% or more of the weeks in the particular month over the period of record classified as snow covered).

Visible vs. Microwave Sensors

Monthly snow extent climatology for NOAA and passive microwave data


Mean monthly snow extent and SWE for 1978-2005 from a blend of passive microwave (SMMR & SSM/I) and visible (NOAA) data

Blend of microwave and visible data -


NSIDC Global Monthly EASE-Grid Snow Water Equivalent Climatology


Blended snow prototypes, SWE (from SSM/I (left side) and AMSR-E (right side)) with snow cover from MODIS, Fall & Spring 2003. Top panels in each set of four represent SWE (mm), with additional area indicated as snow by MODIS in red. Lower panels represent passive microwave snow extent in grey, with percent area of additional pixels that MODIS classifies as snow in blues.

SSM/I + MODIS AMSR-E + MODIS

P.M. (grey) +MODIS %

Fall 2003

Winter 2003


Example of improvement in shallow snow SWE estimates by inclusion of 89 GHz data in combination with MODIS snow-covered area, October 24-31, 2003. Top row includes standard SWE algorithm results, with zoomed area in Northern Europe and Western Russia. Middle row includes additional snow-covered area (red), determined from at least 25% of component MODIS CMG pixels indicating snow cover. Bottom row includes shallow SWE derived from additional information available from the 89 GHz channel. (select day w/ max diff 37 – 85 GHz)

AMSR-E Only

AMSR-E + MODIS

AMSR-E + MODIS +89 GHz (Nagler)

Improved Shallow Snow Detection through Application of 89 GHz


analysis includes:

• autoregression, to avoid conclusions with false confidence in significance, and

• number of years required to detect significant trends (Weatherhead, et al. 1998)

• Changes in snow cover over the last several decades are being detected, although the message is mixed: no clear hemispheric trends, some regional trends, dependency on time period involved, consistent decreasing trends in spring/summer months.

Trends


There is no evidence of a significant trend in either series from1978-2005, > 17 more years required.

There is evidence of a significant decreasing trend in the full series (1966-2005) of visible-derived snow cover, -1.7 %/decade

Northern Hemisphere Visible-Derived Snow Extent Standardized Anomalies


For Western 11 U. S. States, there is evidence to suggest significant decreasing trends in both series considered from 1978-2005, - 9 %/decade.

but no evidence of a significant trend in the full series (1966-2005) of visible-derived snow cover.

Western US Visible-Derived Snow Extent Standardized Anomalies


Continuing Work, Multi-Sensor

• Refine cross-calibration of SMMR and SSM/I brightness temperatures.

• Cross-calibrate SSM/I and AMSR-E and merge the historical snow cover times series with NASA EOS data and products.

• Continue work on optimal blending MODIS and AMSR-E data including use of 89 GHz.


Continuing Work, Trends

Run snow extent trend analysis for specific regions of interest, Arctic Basin, Tibet Plateau, sub-regions within USA etc.

Produce trend surface maps for snow cover duration and snow water equivalent to identify regions of greatest change.

Compare results with surface temperature trends.

Latest results to be presented at Fall AGU 2006 (U09)


November

Comparison of Monthly SWE Derived from SSM/I and the CCIN Gridded North American Data Set, August 1987 – June 1997 (Brown, Brasnett and Robinson, 2003)


January

Comparison of Monthly SWE Derived from SSM/I and the CCIN Gridded North American Data Set (Brown, Brasnett and Robinson, 2003)


March

Comparison of Monthly SWE Derived from SSM/I and the CCIN Gridded North American Data Set (Brown, Brasnett and Robinson, 2003)


Bare Dry Soil


Medium Depth Snow (10-25 cm)


Wet Snow


Antarctica - Dome C

AMSR-E 10 GHz vertical polarization Tb statistics for East Antarctica, descending passes, in 2003. (a) Upper left: (K). (b) Upper right: (K). (c) Lower left: (K). (d) Lower right: Surface topography (meters above sea level).

Statistics for the year 2003 at Dome ‘C’, Antarctica, for 10 GHz and 37 GHz vertical polarizations: (a) Footprint mean, mi; (b) Footprint standard deviation, i. Units are in Kelvins.

mi

i

m

m

m

m

Avg Daily Footprint Mean Stddev of Daily Footprint Mean (time)

Avg Daily Footprint Stddev(space) Avg Footprint Elevation


Statistics Definitions for Stable Targets Study

• Footprint mean = the mean of the brightness temperatures of each grid cell in a footprint.

• Footprint standard deviation =spatial variance at an instant in time or how the grid cells change in brightness temperature over the footprint. Needs to be small to have a homogeneous target.

• Average daily footprint mean = time average of the footprint mean, e.g. over one year

• Standard deviation of the daily footprint mean = standard deviation of the footprint mean over time. Sites with a low SDDFM when the time period is long are sites that generally have low seasonal brightness temperature variation.

• Average daily footprint standard deviation = the average of each footprint standard deviation over some time series. Sites that are spatially homogeneous or have low brightness temperature gradients across a footprint will have low values of ADFSD. An area with a large ADFSD would not be a suitable calibration target, such as locations near land/water boundaries.


Blended snow product prototypes, SWE (mm) from AMSR-E with additional snow extent from MODIS in red (October 24-31, 2003, max). Lower image represents AMSR-E snow extent in grey, with percent area of additional pixels that MODIS classifies as snow in blues. (AMSR-E @ 25 km, MODIS CMG @ 5 km)

Fall Season Example


Blended snow product prototypes, SWE (mm) from AMSR-E with additional snow extent from MODIS in red (Feb 26 – Mar 5, 2003, max). Lower image represents AMSR-E snow extent in grey, with percent area of additional pixels that MODIS classifies as snow in blues. (AMSR-E @ 25 km, MODIS CMG @ 5 km)

Winter Season Example


Fall Season Example

AMSR-E Only AMSR-E + MODIS

AMSR-E +MODIS+ Nagler

October 24-31, 2003

Mask area where MODIS detects snowand select day w/ max diff 37 – 89 GHzto minimize atmospheric influence


Passive Microwave Remote Sensing of Snow

• Radiation emitted from the soil is scattered by the snow cover

• Scattering increases in proportion to amount (mass) of snow

• Brightness temperature decrease, negative spectral gradient,

polarization difference

Tb = ε * Tp (brightness temperature equals emissivity x physical temperature)


Northern Hemisphere monthly average snow water equivalent derived from SSM/I (left) and AMSR-E (right), November 2003 – February 2004.

SSM/I AMSR-E


Snow Cover Mapping on the Tibetan Plateau Using Optical and Passive Microwave Satellite Data

R. L. Armstrong, M.J. Brodzik, M. Savoie, K. Knowles University of Colorado and World Data Center for Glaciology, Boulder -- USA

Asian Conference on Permafrost, Lanzhou, China, August 7-9, 2006

Acknowledgements – Research Support: NASA


Vis

ible

Mic

row

ave

01 Nov 2001 06 Dec 2001


Tibetan Plateau, Winter vs. Summer

\Climatology differences, November and July, showing the anomalous snow cover to be a cold-season phenomenon and not due to land cover type.

November July


Tibetan Plateau, Air Temperature and Frozen Soil

Clockwise from upper left: SWE derived from SSM/I, ECMWF average daily surface temperatures, MODIS snow cover with frozen soil derived from SSM/I, and NOAA IMS product. (November 30, 2001)


Possible Solution: Atmospheric Correction

Atmospheric correction to microwave spectral gradient, modelled at altitudes of 1 km and 4 km. James Wang, GSFC

Atmosphere should be considered when validating satellite retrieval algorithmsbased on surfaceand low-alt. aircraftMeasurements (Wang&Manning, 2003)

Conversely, atmosphere needsto be “removed” when applyingalgorithms empirically adjustedto perform through a full atmosphere at very high elevations.


AMSR-E SWE, Before & After Atmospheric Correction

Result of atmospheric correction, November 29, 2003. SWE derived from uncorrected AMSR-E (left) and corrected AMSR-E (right).


Tibet Plateau Region Snow Cover Jan 9-16, 2003

AMSR-E

MODIS

SSM/I

NOAAIMS


Snow Cover Derived from SSM/I January 9-16 2003


Snow Cover Derived from AMSR-E January 9-16 2003

WDC for Glaciology, Boulder – 30 th Anniversary Workshop Boulder, Colorado 25 October 2006...

Documents

Transcript of WDC for Glaciology, Boulder – 30 th Anniversary Workshop Boulder, Colorado 25 October 2006...