Scaling Properties of L-band Passive Microwave Soil Moisture: From SMOS to Paddock Scale My work is focused on the scaling properties of L-band retrieval of soil moisture, ranging from SMOS to paddock scale Rocco Panciera1, Jeffrey Walker1, Olivier Merlin1, Jetse Kalma2 and Edward J. Kim3 1 Department of Civil and Environmental Engineering, University of Melbourne, Australia 2 School of Engineering, University of Newcastle, Australia 3 NASA Goddard Space Flight Center, Greenbelt, USA
Statement of the Problem Large footprint of passive microwave observations (30-50km) Soil moisture retrieval algorithms developed with tower based studies (~10m): - Homogeneous conditions - Algorithms are non linear Soil moisture retrieval from space using passive microwave sensors presents crucial limitation: large footprint size Land surface is heterogeneous by default at this scale, including variety of landscapes Instead: Soil moisture retrieval algorithms developed amd parameters estimated at small resolution Algorithms need to be tested with real coarse scale data Sub-pixel heterogeneity effect on operational soil moisture retrieval schemes needs to be assessed 4 December 2018 Rocco Panciera
Statement of the Problem Radiative transfer model High resolution Tb Low resolution Tb Tb = 181.2K 0.28 v/v Retrieved footprint moisture content ….What happens at satellite footprint scale? A simple case Bare soil Uniform soil temperature = 317.5K Uniform soil type = Silty Clay Loam soil moisture 0.36 v/v Mean water content The effect of non linearity when applying same algorithms across scales is clear in a simple example Only variability is soil moisture Smooth soil, silty clay loam The relationship between Tb and soil moisture is NON linear If we use the SAME model at different resolution, do we obtain the same area-averaged soil moisture content???? ? 4 December 2018 Rocco Panciera
Objective Verify applicability of current retrieval algorithms at coarse scale (40km) using real L-band data Our study is focused on SMOS type retrieval approach Algorithm: SMOS L2 algorithm Data: NAFE’05 L-band data 4 December 2018 Rocco Panciera
Approach 1km pixels Ground soil moisture Aggregation ? 40km SMOS L2 40km soil moisture product 1km soil SMOS L2 40km footprint Aggregation ? Described NAFE’05 regional data (Resolution 1km, 4 dates, Coverage 40km) Tb data bi-polarized, daily calibrated, 38.4 degree angle Highlight ground points (say they are now at 2km resolution) Describe process Comparison between the two products should indicate wether further Algorithm development is needed for SMOS L2 for heterogeneity effect\ Coincidently, this study is the first validation of the SMOS L@ algorithm with real data Today I am going to focus on… 40km 4 December 2018 Rocco Panciera
Aggregation to 40km Footprint 16 dates Daily calibrated Nadir-referenced for comparison Resolution 62.5m 250m 500m 1km Mean Tb (K) 257.1 258.4 258.9 258.8 STDEV (K) 9.9 7.8 7.3 5.5 Is linear aggregation legitimate? Aggregation has been verified with Multiresolution data Comment patterns, retained features but smoothed Table is example for one day averaging of TB measurements at high resolution to a lower resolution results in a good agreement with the direct lower resolution measurements. the results suggest that aircraft data can be reliably averaged up to simulate satellite footprint More than 80% of the SEE values are contained within the +/- 3K band, Error standard deviation of 3.2K and 3.4K for Krui and Merriwa respectively. higher SEE are mainly associated with unusually high heterogeneity steep soil moisture gradient provoked by localised showers. +1.3 +1.8 +1.7 Linear aggregation of Tb is reliable 4 December 2018 Rocco Panciera
Simulated L-band emission Model Description SMOS Level-2 algorithm Mixed pixels, 4 surface types Bare soil Grassland Crop Forest Tau-Omega emission model L-MEB for each over type ( Wigneron et al. 2007, Remote Sensing of Environment) Soil moisture L-MEB model (i) Land cover type SOIL TYPE % surface type “i” In pixel Surface type dependent parameters Soil/canopy temperature L-band observation Optimization Simulated L-band emission Within each pixel, the brigthness temperatures from each cover type are simulated with a forward model and then aggregated Parameters driving the forward model are selected and tabulated based on the selected vegetation classes and maps of soil properties The mixed pixel emission is calculated as a linear combination of the Tb of each land cover, weighted by their respective cover fraction Heterogeneity is taken into account, BUT soil moisture is the same across surface cover types) 4 December 2018 Rocco Panciera
Model Description Assumptions Soil moisture uniform across the cover types within each pixel Effective temperature to microwave emission Optical depth Optical depth of forests is a constant No modelling of rainfall interception by plants Teff modelling very recent. Parameterization of w and b undergoing Tau of forest constant (with angle, polarization and time) is related to the maximum LAI over the year – dominance of branch effect with respect to leaf effetcs Only dominant cover type modelled @ 1km resolution Open woodland is modelled as grassland 4 December 2018 Rocco Panciera
Ancillary Data Landcover Landsat derived landcover map (25m) Dominant cover type @ 1km resolution Explain limits of nafe focus areas and approx limits of pixels = 21k aircraft mapping Land cover map verified by ground visual estimation of dominant land cover, Preferred to ECOCLIMAP 1km Dominant cover type that will be used for the radiative transfer Cover Type % Grassland 50.7% Forest 24.3% Open woodland 8.5% Crops 4.5% 4 December 2018 Rocco Panciera
Ancillary Data Vegetation Water Content Experimental relationship VWC vs NDWI Crop Grassland Relationship were established between ground measured VWC and NDWI index, which Was demonstrated to be more efficient then NDVI or LAI for vwc mapping One relationship for crop, one for grass VWC 4 December 2018 Rocco Panciera
Ancillary Data Soil/Canopy Temperature Effective soil temperature to microwave emission 8 stations Ts 2.5cm Ts15cm Soil temperature range @2.5cm Soil temperature range @15cm Oct-31 1.9 2.0 Nov-07 3.1 Nov-14 5.4 3.3 Nov-21 6.1 3.5 Requires an estimation of T soil depth~= T 30cm Tsoil surf~=T @ 2-4cm We have available 12 stations with T2.5cm and T15cm The possibility to use mean values instead of interpolation was explored Given the small spatial variation and uncertainty added by interpolation, average values were taken. Differece in Teff so calculated <1K Assumptions: TSOIL_DEPTH = Average of T15cm TSOIL_SURF = Average of T2.5cm TCANOPY = TSOIL_SURF 4 December 2018 Rocco Panciera
Ancillary Data Surface Type Dependent Parameters All these parameters have been calibrated at tower scale, but literature is in fair agreement for most (that’s why they have been adopted for SMOS L2) Vegetation parameter b for crops – variety of values in literature. The one adopted is in the lower range for L-band Still high level of uncertainty for HR. a number of studies suggested dependence of soil moisture for Crops, grassland with litter and recently bare soil Several runs varying uncertain surface type parameters (b, albedo, HR) to check for sensitivity HR was found crucial and dependence on sm important to reproduce dynamic. Still research topic. Here this formulation assumed from experimental data (Wigneron, J. P. ,Personal communication) HR=1.3 -1 .13 * (SM) (Saleh et al. ,2007 SMOSREX site) 4 December 2018 Rocco Panciera
Validation of 1km Product Results to Date Nov 7 Nov 14 Nov 21 Retrieved soil moisture (v/v) To date, the validation of LMEB at 1km is this. Oct 31 removed due to likelyhood of water interception Comments to patterns: Effect of FAO soil type Effect of soil temperature change in time (deltaTb<7,delta Ts<5) 4 December 2018 Rocco Panciera
Validation of 1km Product Results to Date Soil moisture error (v/v) (retrieved – observed) Nov 7 Nov 14 Nov 21 Bare (14pts) 12.1 6.9 7.5 Grassland (113pts) 11.8 10.1 6.7 Crop (9pts) 19.9 10.2 6.0 Forest (4pts) 14.8 4.9 5.6 No evident patterns, Errors appear bigger and underestimation on sandy soils on wet RMSE reveal we are doing bad on crops variety of crops treated as one, and unique value of parameter b suggest need for better estimation Note difference in number of points in these averages RMSE %v/v 12.4 9.7 6.7 Bias % v/v +1.7 -4.9 +1.5 4 December 2018 Rocco Panciera
Parameter Calibration Estimation HR= a + b *(SM) HR greater impact, and crucial to model soil moisture dynamics Literature suggests is a function of SM for all surface cover types – still research topic Calibrated here @1km resolution Saleh formulation seems accurate for bare, grass and some crops It appears that at very dry values HR discreases again Makes sense if think about “dielectric roughness” effect BARC: low grass cover, no litter ELBARA: moderate grass cover, litter 4 December 2018 Rocco Panciera
Where To Improve? Improve estimation of HR = f(soil moisture) Improve estimation of b for crops Finer scale soil type map Better TEFF estimation from ground data Use MODIS thermal to estimate TCANOPY Improvement in landcover map 4 December 2018 Rocco Panciera
Conclusions Validation of SMOS L2 retrieval scheme with real data undergoing @ 1km resolution Demonstrated reliability of linear Tb aggregation for satellite footprint simulation Insights into effect of crucial surface roughness parameter at aircraft scales NEXT…………. Retrieval from simulated 40km Satellite pixel Assessment of performance of algorithm at different scales 4 December 2018 Rocco Panciera
Thank you 4 December 2018 Rocco Panciera