Abstract
An algorithm is developed for retrieving soil moisture at plateau scale by combined usage of Aquarius active and passive L-band observations. In this algorithm, Look-Up-Tables (LUTs) are established for bare soil and vegetated areas by using the physical based Tor Vergata discrete electromagnetic model (hereafter, TV-DEM). In the case of vegetated area, the LUT is built based on simulations with varying soil moisture and Leaf Area Index (LAI). Only soil moisture is variable for the bare soil case, and values calibrated in previous works are adopted for the other TV-DEM parameters. Soil moisture is then retrieved by minimizing a squared difference object function based on the emissivity and backscatter coefficient observed by Aquarius and the corresponding TV-DEM simulations included in the LUT. The soil moisture retrievals are assessed at Aquarius footprint scale using in-situ measurements collected at three regional scale networks spread across the Tibetan Plateau. The unbiased root mean squared differences (ubRMSDs) from the three networks vary from 0.016 to 0.050 m(3) m(-3) and coefficients of determination (R-2) are from 0.274 to 0.499 (-). This ubRMSD and R-2 is in the same order of the passive-only official Aquarius product, Metop-A Advanced SCATterometer (ASCAT) L2 soil moisture product (hereafter ASCAT) as well as reanalysis data generated by European Centre for Medium-Range Weather Forecasts (hereafter, ERA-Interim). At Plateau-scale, all four soil moisture products capture the seasonal trend, whereby the dynamic range during the monsoon season captured by the ERA-Interim product is relatively small. Further, the northwest-southeast dry-wet gradient due to precipitation and evapotranspiration is well captured by the soil moisture spatial distributions produced by TV-DEM Aquarius, official Aquarius and ASCAT products, but less pronounced in the ERA-Interim product. This study demonstrates that the TV-DEM based algorithm can be used to retrieve soil moisture over a Plateau scale with robust results in terms of error statistics (e.g. bias, R-2 and ubRMSD) and can generate realistic spatial patterns for soil moisture at Plateau-scale.
NSTL
Elsevier