Analysis of different synergy schemes to improve SMOS soil moisture retrievals (accuracy and spatial resolution) A. Chanzy 1, B. Berthelot 2, S. Cros 1, M. Weiss 2, J.-C. Calvet 3, T. Pellarin 4, J.-P P Wigneron 5 1 INRA Avignon (UMR Climat Sol et Environnement) 2 Noveltis Toulouse 3 CNRM - Météo o France Toulouse 4 CNRS LTHE Grenoble 5 INRA Bordeaux (EPHYSE) Funded by ESA contract n AO/1-4676/04/NL-FF
Soil Moisture Retrieval on continental areas Microwave emission process involve several parameters τ ω model is foreseen for the retrieval algorithm Soil moisture Vegetation optical thickness Effective temperature Vegetation parameters (Cpol and ω) Soil surface roughness Soil texture SMOS low resolution heterogeneous pixel land cover knowledge to feed composite models and masks to eliminate non invertible pixels bare surface low vegetation forest open water, snow, frozen soil Urban areas Topography
Soil Moisture Retrieval on continental areas Need of improving spatial resolution Dis aggregation Additional components that influence microwave emission Litter Intercepted water in vegetation After rainfall Dew (morning orbit) Lack of knowledge Difficult to quantify
Synergy Remote sensing Optical remote sensing (20m km) Vegetation parameter Land cover Thermal infrared Effective temperature Thermal infrared/optical Optical/met data Dis aggregation SAR Dis aggregation (Roughness)
Synergy spatial products Land cover product Met product Rainfall dis-aggregation Temperature effective temperature Soil map Snow map Frozen soil map
DATA Simulation of brightness temperature Surface properties simulated by the ISBA model using a global meteorological forcing, ecoclimap,, Fao soil map Brightness temperature computed for every tiles by LMEB model (τ ω model) or SRS model (account( for orbitography, and a simplified instrument model) Inversion τ ω model considering several fractions SMOS pixel Forest Open Water Bare soil Low vegetation
Three data bases DATA Global data set (one year, every day,, 50 km resolution 4 tiles) (Pellarin et al) SRS test data set (Europe : 1500 by 1500 km) ( several days,, 4 km resolution (dominant tile)) Alpilles-ReSeDA (4 km/km, One year, every day,, 20 m resolution
Synergy using optical data Quantify vegetation development Improve land cover knowledge Snow Bare soil Soil characteristics
Synergy using optical data Quantify vegetation development τ= = f(vegetation index) + error
A priori information on optical thickness SRS software Europe dataset Inversion made on moisture and optical thickness
A priori information on optical thickness SRS software Europe dataset Inversion made on moisture only
Impact of Land cover knowledge in Land cover classes Only low vegetation retrieval Inverted parameters Additional informatio n RMSE (m 3 /m 3 ) % of good pixels θs, τ, Teff - 0.049 54 Low vegetation and Forest Low Veg + Forest + Open Water Low Veg + Forest + Open Water+bare Low Veg + Forest + Open Water+bare θs, τ LV, Teff Forest prescribed θs, τ LV, Teff F. and OW prescribed θs, τ LV, Teff F. and OW prescribed θs, τ LV, Teff F. and OW prescribed - 0.041 68.5-0.039 68 Soil texture Moisture difference between fractions 0.035 75 0.0206 0.0204 92.5 92.7
Impact of Land cover products choice
Impact of Land cover products difference Tb on forest is prescribed. Inversion was made on soil moisture and τ
Impact of Land use products difference
Bare soil fraction Bare soil fractions Permanent But also variable in time Bare soil fractions given by optical products (fcover)
Improvement in Soil moisture error when bare soil is taken into account
Improvement in soil moisture when bare soil is taken into account Pixel : North Tunisia : 52 % bare, 4% forest, 44% low vegetation 0,4 0,35 0,3 0,25 0,2 0,15 Theta obs. Theta P3 Theta P4 0,1 0,05 0 0 50 100 150 200 250 300 350 400 RMSE without bare soil : RMSE With bare soil : 0.019 m3/m3 0.0075 m3/m3
Impact of bare soil fraction on τ and Teff retrieval 0,14 0,12 0,1 0,08 0,06 0,04 0,02 0-0,02-0,04-0,06 0 100 200 300 400 305 300 TAUobs TAU sim P4 TAU sim P3 295 290 285 280 T1obs T1sim P4 T1sim P3 275 270 0 100 200 300 400
Impact of information on scene temperature Scheme % of accurate pixel Mean RMSE (m 3 m -3 ) Teff unknown 84.61 0.0314 Teff perfectly known 91.48 0.0210 Teff computed using a 93.82 0.0212 constraint around surface temperature (1K) Teff computed using a 92.08 0.0268 constraint around air temperature (2K) Teff unknown no noise on data 92.54 0.0208
disaggregation schemes Optical/thermal infrared soil moisture through the evaporation fraction Soil texture map/land cover map Rainfall heterogeneities SAR (when( available but difficulties to derive accurate estimation of soil moisture)
Optical/thermal disaggregation scheme NDVI max High Soil Moisture NDVI* Low Soil Moisture NDVI min T min T* T max
Optical/thermal disaggregation scheme NDVI (1km) Ts (1km) ALBEDO (1km) AGGREGATION NDVI (50km) Ts (50km) ALBEDO (50km) Universal triangle coefficients Soil Moisture (8km) Soil Moisture (50km)
Optical/thermal disaggregation scheme Daily based triangle Monthly based triangle
Optical/thermal scheme : results with daily trapezoid
Optical/thermal scheme : results with monthly trapezoid
Conclusions Constrain the inversion (on vegetation parameters and temperature) and add information on surface heterogeneities (bare soil fraction) led to significant improvements. Soil texture is found to be very important parameter arise the question of the accuracy of world soil map (no real alternative) How soil heterogeneity affect the retrieval Disaggregation Optical/thermal scheme SAR optical/thermal infrared needs strong contrast (small, strong?) improvement may be obtaines using more sophisticated models (SVAT models as SEBS) Time revisit Product robustness Take profit of land cover and climatic heterogeneities Working with simulated offers an Optimistic view of the real world.