Performance of Four Passive Microwave Soil Moisture Products in Maize Cultivation Areas of Northeast China

Passive microwave remote sensing is an effective way to obtain global soil moisture (SM) measurements, and many studies have explored the uncertainty inherent in microwave-based SM products. However, SM product accuracy has not been evaluated in northeast China, a national and global production base for commodity grain. In this study, a ground-based wireless sensor network with 28 observation nodes that were spatially distributed within 36 x 36 km was established to achieve satellite-scale "true" SM values through sensor calibration for specific soil types, senor consistency testing, and spatial scale transformation. The uncertainties of four passive microwave SM products (SMAP L3, SMOS L3, the Japan Aerospace Exploration Agency/JAXA AMSR2, and FY3C) were investigated and the following conclusions were obtained: 1) SMAP SM accuracy was very close to the expected application accuracy of 0.04 cm(3)/cm(3), followed by SMOS, FY3C, and AMSR2; 2) for SMOS and SMAP, there were no significant temporal changes in SM errors, except for the larger error of descending SMOS SM and June SM values for descending SMAP and ascending SMOS. AMSR2 SM generally underestimated field SM, while FY3C SM values under low vegetation conditions were more consistent with field data, with an error of about 0.06 cm(3)/cm(3); 3) agricultural activities and rainfall caused the soil surface roughness to increase or decrease within a growing season, which may have been an important source of satellite-scale SM error indicated by high bias values in July for both SMAP and SMOS; and 4) the standard deviation of field SM (0.06 cm(3)/cm(3)) produced a SMAP SM error of about 0.06 cm(3)/cm(3) in low vegetation water content conditions, indicating that SM spatial heterogeneity cannot be ignored in the retrieval algorithm. This article investigated the accuracy and error sources of four satellite SM products in the farmland area of northeast China, and identified future research directions for further improving SM algorithms