VALIDATION AND INTER-COMPARISON OF FOUR GLOBAL … · Criteria of selection QA Criteria Perfor...
Transcript of VALIDATION AND INTER-COMPARISON OF FOUR GLOBAL … · Criteria of selection QA Criteria Perfor...
VALIDATION AND INTER-COMPARISON OF FOUR GLOBAL SURFACE ALBEDO PRODUCTS USING THE SALVAL TOOL.
www.eolab.esContact: Jorge Sánchez-Zapero.
EOLAB CTO & Research Engineer
(+34)963769448 [email protected]
https://www.linkedin.com/in/jorgesanchezzapero/
EOLAB – Earth Observation LaboratoryParc Científic Universitat de València. C/ Catedrático Agustín Escardino, nº 9, 46980 Paterna, Valencia, Spain
ABSTRACT
METHODOLOGY
J. Sánchez-Zapero (1), F. Camacho (1), L. de la Madrid (1), Z. Wang (2), M. O. Roman (3)
(1) EOLAB, Valencia, Spain. ([email protected], [email protected], [email protected])
(2) University of Maryland College Park, College Park, MD, USA. ([email protected])
(3) NASA/GSFC, Greenbelt,MD, USA.([email protected])
ProductPlataform/
SensorPeriod
SpatialResolution
Temp. Frec.
Comp. Period
Projection Spectral albedos
C_GLS V1SPOT/VGT 1999-2014
1km10
days30
daysPlate
carrée
▪VIS [0.4-0.7μm]▪NIR[0.7-4.0μm]▪BB [0.3-4.0μm]PROBA/VGT 2014-Present
MCD43A3 C6TERRA+AQUA/
MODIS2000-present 500m Daily
16 days
Sinusoidal ▪VIS [0.3-0.7μm]▪NIR[0.7-5.0μm]▪BB [0.3-5.0μm]
GLASS
NOAA/AVHRR 1981-1999 ~5km
8 days17
daysSinusoidal
▪BB [0.3-3.0μm]TERRA+AQUA/MODIS
2000-2010 1km
EBF DBF NLFCrop
Shrub
Herb.
BA
LANDVAL 9 6 10 26 21 16 11
Global 7 8 16 17 24 10 14
0
5
10
15
20
25
30
Pe
rcen
tage
NOAM
SOAM
EURO
AFRI ASIAOCE
A
LANDVAL 25 12 14 21 22 5
Global 19 11 9 24 31 6
05
101520253035
Pe
rce
nta
ge
Quality Criteria Validation Metrics
CompletenessGap size distribution (annual maps, temporal variations per biome/continent). Length of gaps.
Spatial Consistency
Visual inspection global maps.
Monthly maps and histograms of residuals.
PDFs & histograms of residuals per biome and continent.
Indicator of spatial homogeneity and correlation (CV, MoranIndex)
Temporal ConsistencyQualitative inspection of temporal variations
Cross-correlation and auto-correlation.Intra-annual Precision Histograms of the smoothness
Inter-annual PrecisionBox-plots per bin and median absolute anomaly of 95th and 5th percentiles for two consecutive years
Stability
Box-plots per bin and median absolute anomaly of 95th and 5th
percentilescomparing each year to the long term average
The slope of the evolution of inter-annual precision
Systematic errors (Bias)
Mean differences (bias)
Major Axis Regression linear fit (slope, offset)
Histogram of differences
Box plots of differences per bin
Total Error (Accuracy)
Pearson’s coefficient (power of the linear fit)
Root Mean Square Deviation (RMSD)
Box-plots of absolute bias per bin
Surface albedo is defined as the fraction of incident solar irradiance reflected by Earth's surface over the whole solar spectrum, and represents an Essential Climate Variable (ECV) as established by the Global Climate
Observing System (GCOS). The uncertainty of satellite-based albedo products must be evaluated over global conditions with the aim to determinate the compliance of user requirements (GCOS) and the usability.
Moreover, the availability of numerous products creates a need to understand the level of consistency between satellite products.
This work presents the validation and inter-comparison of four global surface albedo products: SPOT/VGT V1 of the Copernicus Global Land Service, NASA MODIS MCD43 C6, Global LAnd Surface Satellites (GLASS),
and PROBA-V V1 for the continuity of the Global Land Service. The results are generated using the Surface ALbedo VALidation (SALVAL) tool. SALVAL has two main objectives: (i) to provide transparency and
traceability in the validation procedure, designed to be compliant with the CEOS-LPV (Comitee on Earth Observations Satellites - Land Product Validation) sub-group and QA4EO (Quality Assurance for Earth
Observation) recommendations, and (ii) to provide a tool to benchmark new products or update product validation results as the time serie expands, reaching Validation Stage 4 in the CEOS LPV hierarchy.
designed to represent globally the variability of land surface types over optimal conditions in terms of spatial homogeneity and topography
Evaluated over
2001-2004 period: SPOT/VGT V1, MODIS C6 & GLASS SPOT/VGT V1 vs PROBA-V V1 during the overlapperiod (Nov’13-May’14) for the continuity of theGlobal Land Service
Product Completeness
ConclussionsConclussions
Evaluation of the sampling per biome type and continental region
ParameterThresho
ldPurpose
Distance to open water bodies [km]
5
Avoid open water bodies and their
changing reflectance
behavior with viewing geometry
Minimum fraction of
majority land cover type at 5
km distance
60%Avoid areas with
heterogeneous land cover.
Land Cover Majority at
5km
Exclude 'Water bodies' and 'Urban areas'
Vertical range [m] within a distance of
5km
<300m
Avoid areas with significant terrain
variability close to a site.
Location (Latitude)
60ºS to 80ºN
Exclude sites over extreme latitudes, where Global Land products does not
provide data
Network # of sites
SAVS 1.0521
(including256 BELMANIP2.1)
Calibration Sites
20
Additional sites (from existing
networks)26
Additional sites (Geo-Wiki)
158
Total 725
Criteria of selection
QA
Criteria
Perfor
manceComments
Product
Completeness - Main limitations over Northern latitudes in wintertime and Equatorial areas. Similar results than SPOT/VGT (and MODIS)
Spatial Consistency +Global distributions showed systematic positive bias (PROBA-V > SPOT/VGT) for NIR and BB, and bias 0 for VI.
Global distributions of residuals showed ~36% of cases within the optimal level for VI, and 50% for NI and BB
Temporal
Consistency + Reliable temporal variations for most of the cases compared with satellite reference products and ground observations.
Intra-Annual
Precision + Similar smoothness than both references (SPOT/VGT and MODIS C5), showing slightly higher δ values in NI.
Overall Spatio-
Temporal
Consistency±
PROBA-V vs SPOT/VGT shows high correlation (R>0.93) and low scattering, with almost no mean bias in VI and systematic positive
mean bias of ~5% in NI and BB (except in snow). 39%, 43% and 42% (67%, 73%, and 75%) of pixels showed optimal (target level) for
VI, NI and BB.
Comparison of PROBA-V and SPOT/VGT per biome type showed low bias (<3%, random sign) for VI, and positive bias for NI and BB in
all biome types. The exception was the snow class, with negative bias.
Accuracy
Assessment -PROBA-V:N=274; B=0.032 (22.1%); RMSD=0.042; Snow free conditions. 4% of pixels within GCOS.
Improved results for MODIS C5 using the same sampling: B=0.006 (4.9%); RMSD= 0.029; 18.1% of pixels within GCOS.
Product Completeness
Spatial Consistency
Temporal Consistency
Temporal Realism Accuracy Assessment
Intra-AnnualPrecision
Spatio-Temporal Consistency
Spatio-Temporal Consistency
SPOT/VGT V1 vs MODIS C5 SPOT/VGT V1 vs GLASS GLASS vs MODIS C5
Spatial Consistency
Intra-Annual Precision
Inter-Annual Precision
Temporal Consistency
QA
Criteria
SPOT/
VGT
MODIS
C6GLASS Comments
Product
Completeness - - +SPOT/VGT & MODIS showed main limitations over Northern latitudes in
wintertime and Equatorial areas. GLASS is gap-filled product
Temporal
Consistency ± + +Good temporal agreement between the three satellite products.
SPOT/VGT: Some limitation observed to detect spurious snowfall events,
and slight seasonality over desertic calibration sites.
Intra-Annual
Precision + + + Similar smoothness all products
Inter-Annual
Precision ± + +
2004 vs 2003
SPOT/VGT: RMSD 2,6% over calib. Sites. Median abs. Anomaly: 3%
MODIS C6: RMSD 2,6% over calib. Sites. Median abs. Anomaly: 2,3%
GLASS: RMSD 2,6% over calib. Sites. Median abs. Anomaly: 2,2%
QA
Criteria
SPOT/VGT V1
Vs
MODIS C6
SPOT/VGT V1
Vs
GLASS
GLASS
Vs
MODIS C6
Comments
Spatial
Consistency + ± +
Residuals:
VGT vs MOD: 15% GCOS / 75% Target
VGT vs GLASS: 14% GCOS / 53% Target
GLASS vs MOD: 39% GCOS / 82% Target
Overall
Spatio-
Temporal
Consistency
+ + +
VGT vs MOD
VGT vs GLASS
GLASS vs MOD
R 0,95 0,94 0,97
Bias 1,9% -0,4% 1,7%
RMSD 0,04 (19%)
0,05 (22%)
0,03 (16%)
%GCOS/Target 40/69 42/71 54/77