Validation of SMAP Soil Moisture at Terrestrial National Ecological Observatory Network (NEON) Sites Show Potential for Soil Moisture Retrieval in Forested Areas
preprintposted on 28.05.2021, 08:02 by Edward AyresEdward Ayres, Andreas Colliander, Michael CoshMichael Cosh, Joshua A. Roberti, Sam Simkin, Melissa A. Genazzio
Soil moisture influences forest health, fire occurrence and extent, and insect and pathogen impacts, creating a need for regular, globally extensive soil moisture measurements that can only be achieved by satellite-based sensors, such as NASA’s Soil Moisture Active Passive (SMAP). However, SMAP data for forested regions, which account for ~20% of land cover globally, are flagged as unreliable due to interference from vegetation water content, and forests were underrepresented in previous validation efforts, preventing an assessment of measurement accuracy in these biomes. Here we compare over twelve thousand SMAP soil moisture measurements, representing 88 site-years, to in-situ soil moisture measurements from forty National Ecological Observatory Network (NEON) sites throughout the US, half of which are forested. At unforested NEON sites, agreement with SMAP soil moisture (unbiased RMSD: 0.046 m3 m-3) was similar to previous sparse network validations (which include inflation of the metric due to spatial representativeness errors). For the forested sites, SMAP achieved a reasonable level of accuracy (unbiased RMSD: 0.06 m3 m-3 or 0.053 m3 m-3 after accounting for random representativeness errors) indicating SMAP is sensitive to changes in soil moisture in forest ecosystems. Moreover, we identified that both an index of vegetation water content and canopy height were related to mean difference, which incorporates measurement bias and representativeness bias, and suggests a potential approach to improve SMAP algorithm parameterization for forested regions. In addition, expanding the number and extent of soil moisture measurements at forested validation sites would likely further reduce mean difference by minimizing representativeness errors.