Towards an Active/Passive Soil Moisture Algorithm for Aquarius/SAC-D Institute of Astronomy and Space Physics (IAFE-CONICET-UBA) Buenos Aires, Argentina C.A. Bruscantini, F.M. Grings, H. Karszenbaum cintiab@iafe.uba.ar

Radiometer algorithm: review of the existing algorithms and an introduction of the bayesian scheme

ω-τ model Aquarius MWR* Land Cover Dependent MODIS Aquarius sm sm sm τ TbH TbV Tb37V b, h, ω Stem Factor NDVI σHH σHV σVV Aquarius- footprint Filter Aquarius- footprint Filter Aquarius- footprint Filter 8σHV σHH+2σHV +σVV Flags Nearest neighbor to SF grid Water Correction Aquarius- footprint Filter Owe et al. 2001 SMAP ATBD RVI Water Correction Tsoil Linear averaging VWC 85% soybean? Soil Texture Linear averaging Non linear aggregation Kim et al. 2012 Ancillary Data ω-τ model VWC This is a flowchart of all the soil moisture retrieval algorithms developed using Aquarius radiometer observations. As you can see in the flowchart, it uses Aquarius brightness temperature and ancillary data, which includes: soil temperature derived from vertical polarization brightness temperature at 37 gigahertz obtained from MWR, which is a radiometer developed by CONAE on board Aquarius, MODIS NDVI to get vegetation water content, and static parameters including soil texture and land cover dependent parameteres. All the retrieval algorithms implement the omega-tao model. *MWR or alternative SCAH SCAV MPDA Bayes sm sm sm τ sm τ

Bayesian Algorithm Repeat for every point of the likelihood domain The bayesian algorithm, instead of using deterministic values of ancillary parameters, treat them as random variables, in particular with gaussian distribution, centered on their estimated values, and with variance related to the their uncertainties. Several samples of this distributions are drawn and entered to the forward model (the RT-0 in this case) together with a fixed sm and tau value. Several pairs of tbh and tbv are generated and through a kernel smoother, the joint density function is obtained, and is finally evaluated on the measured pair TbHm and TbVm. This is repeated for every sm and tau values on the grid domain. This generates the likelihood pdf. Using bayes theorem, the likelihood is multiplied by the prior to finally get the posterior pdf. The prior pdf contains all previous knowledge of the variables to estimate (sm & tau). In our analysis, we used an uniform pdf for sm ranging from 0 to 0.5 volumetric sm, therefore all values are equally likely. For tau, we use a gaussian distribution centered on the estimation of tau derived from MODIS NDVI, and its variance is a function of the difference in the tau values estimated by ndvi and aquarius rvi. Finally, from the posterior distribution, two estimators were implemented, the minimum variance estimator, and the maximum a posteriori, and their respective standard deviation of the estimations were computed.

Interpretation of retrieval algorithms MAP The RT-0 model contour level curves (evaluated at some specific ancillary parameters and Aquarius observations TbH and TbV ) are marked as TbH and TbV curves on the plot. The MPDA, that uses both H- and V- channels, retrieves the values of sm and where both curves intersect. SCAH and SCAV use only TbH and TbV respectively, and tao (derived from NDVI) as inputs to retrieve sm. BRA approach uses RT-0 to compute the likelihood pdf. The likelihood provides a probability for each [sm, tao] pair, because the ancillary parameters were considered to be random variables instead of deterministic values. The posterior distribution is shown in the figure as the colored background image, where reddish colors indicate higher probability values and bluish lower. Given this posterior, the Mean and MAP estimations are computed as the expectation and the mode of the distribution respectively.

Retrieval Algorithms Comparison SM SMOS [m3/m3] SM USDA [m3/m3] SM SCAH [m3/m3] SM SCAV [m3/m3] SM GLDAS [m3/m3] SM GLDAS [m3/m3] SM GLDAS [m3/m3] SM GLDAS [m3/m3] SM MPDA [m3/m3] SM Bayes Mean [m3/m3] SM Bayes Map [m3/m3] SM GLDAS [m3/m3] SM GLDAS [m3/m3] SM GLDAS [m3/m3] [REF]: Bruscantini, C.A.; Grings, F.M.; Barber, M.; Carballo, F.; Perna, P.; “A Bayesian approach for soil moisture and optical depth retrieval: evaluation on Aquarius/SAC-D observations”, Journal of Selected Topics in Applied Earth Observations and Remote Sensing. (undergoing review)

Radar algorithm: New radar-only algorithm

New radar-only algorithm: Conceptual model Sensitivity and Intercept dependence with vegetation (Radar Vegetation Index, RVI) and soil roughness (Radar Roughness Index, RRI). Definition of three limiting cases: Smooth bare soil (1) Rough bare soil (2) Maximum vegetation (3) The radar retrieval intends to capture the nonlinear dependency between backscatter and SM over bare soils and the linear dependency over vegetated areas by proposing a dependency of the form: VV backscatter is equal to sensitivity multiplied by SM elevated to lambda plus an intercept, where lambda goes from 0.3 to 1 as vegetation density increases. Sensitivity is the ratio between sigma [dB] and mv variations, and it varies with vegetation density and soil roughness. Observations of sigma VV over bare soil are expected to have the highest sensitivity to mv, and it decreases with higher vegetation density, because of signal attenuation due to the vegetation layer. Intercept is the sigma [dB] expected over dry soils. Similar to sensitivity, intercept relates to vegetation density and soil roughness. However, unlike sensitivity, intercept increases with vegetation due to backscatter contribution of the vegetation cover. Sensitivity and intercept are based on the definition of three limiting cases, i.e. end-members, (1) smooth bare soil, (2) rough bare soil and (3) maximum vegetation covered soil. (which are the triangles end-points on the Aquarius sensitivities and intercept figures). [REF]: Narvekar, P.S.; Entekhabi, D.; Kim, S.; Njoku, E.G.; , “Soil Moisture retrieval using L-band radar observations”. (undergoing review).

New radar-only algorithm: Sensitivity and Intercept analysis Aquarius global observations Sensitivity Intercept Work in progress. Red points indicate end members.

Vegetation and Roughness Contribution to Backscatter Soil Moisture contribution Vegetation contribution Roughness contribution The radar algorithm allows us to tell apart the vegetation, soil roughness and SM contributions to backscatter. Thus, some sites of interest were selected for time series analysis. It was found that the three mv products capture surprisingly well the precipitation peaks in all focus regions. We found radar retrieval accuracy to be quite promising.

Radiometer/radar fusion: physically-based synergy

Radiometer/Radar fusion: physically based synergy Some previous works have tried to address the issue of relating emissivity and backscatter and how do they depend on soil moisture and biomass. According to Ferrazzoli et al, emissivity versus backscatter plots describe a triangular shape as the one in the figure. When biomass increases so does emissivity and backscatter, whereas when SM increases, emissivity decreases and sigma increases. Keeping this in mind, we wanted to test Aquarius observations. [REF]: P. Ferrazzoli and L. Guerriero, “Synergy of active and passive signatures to decouple soil and vegetation effects”, Microrad 2010.

Aquarius Observed Triangles (global observations) Plotting emissivity derived from Aquarius radiometer and NCEP soil temperature, against Aquarius backscatter at each pixel, yielded this two figures. The one of the left, where color of the datapoints are SM derived from GMAO, being blue points drier and red ones wetter; and the one on the right where colors are Aquarius RVI, related to vegetation. As expected, the emisivitty-backscatter relation arranges in a triangle shape, where the upper triangle end point is where there is highest RVI values, and thus maximum vegetation pixels, and pixels are organized from left to right from wetter to drier conditions.

Triangle Edges Location possible RFI coastal areas deserts Triangle end points location were searched on global maps, and wettest pixels where related to coastal areas, driest pixels to deserts and highest backscatter and emissivities pixels were probably related to RFI areas.

Aquarius Triangles by Land Covers Croplands Triangles were also separeted by land cover classification and these are some examples observed.

Summary More than 3 years have passed since Aquarius/SAC-D mission was launched. On December 2013, Aquarius official soil moisture maps were released worldwide provided by NASA NSIDC webpage. Only Aquarius H-pol radiometer observations are used at this time. More recently, efforts were made towards exploiting all Aquarius available sources of information (radar/radiometer). In this context, active/passive fusion constitutes a new contribution for soil moisture retrieval. This work addresses models and observations related to active/passive fusion. Although Aquarius soil moisture monitoring might be somehow unattractive for final-user applications, its collocated radar and radiometer instruments give a unique opportunity for space- borne active/passive fusion retrieval development, of special interest for forthcoming SMAP (Soil Moisture Active Passive) mission. This presentation shows the latest collaborative efforts (IAFE/MIT/Tor Vergata) towards an active/passive algorithm.