A Performance Analysis on Soil Dielectric Models Over Organic Soils in Alaska for Passive Microwave Remote Sensing of Soil Moisture

Passive microwave remote sensing of soil moisture (SM) requires a physical model that describes the impact of SM on the soil dielectric constant and subsequently on the observed brightness temperatures. Mironov 2009, the semi-empirical dielectric mixing model that is adopted in the operating algorithm of the NASA SMAP and the ESA SMOS missions, however, exhibits a degraded performance over organic soils that were not considered in this model development. To this end, we present a comparative analysis of the performances of nine advanced soil dielectric models in Alaska where soils are characterized by a significant number of organic compounds. Notably, four out of these nine models account for the effect of soil organic matter on the dielectric constant of the soil-water-air system. SM retrievals related to different dielectric models were derived via an iterative optimization scheme presently employed in the SMAP SCA-V. Statistical metrics computed by comparing these SM retrievals against in-situ measurements over sites in a variety of organic matter were utilized to reflect the practical quality of each model at a 36 km spatial scale. As a result, there is a systematic discrepancy between the SM retrievals obtained by the mineral- and organic-soil-based models evidently when the soil organic matter exceeds 15% and SM is more than 0.1 m3/m3. Based on the identical SMAP-observed brightness temperatures, SM retrievals from the models that consider organic content tend to overestimate while those from the remaining models display dry biases compared to in-situ benchmarks. During a seven-year studying period, Mironov 2019 exhibits a mean ubRMSE of 0.0507 m3/m3 and a mean R of 0.465, showing a moderately superior performance to other models. Globally, SM estimations from Mironov 2019 are at least 0.05 m3/m3 higher than those from Mironov 2009 in organic soils but are occasionally smaller than Mironov 2009 estimations in mineral soils especially when SM is smaller than 0.1 m3/m3. In an integrated consideration of the model inputs, the physical reasonability, and the practical accuracy based on the results hereon and previous reports, the separate use of Mironov 2009 and Mironov 2019 in the SMAP retrieval algorithm for mineral and organic soils would be a preferred option.