AUTHOR=Shi Xiao , Yang Ruyi , Ullah Waheed 

TITLE=Evaluation of ERA5 reanalysis and ECV satellite soil moisture products based on in situ observations over Jiangsu, China

JOURNAL=Frontiers in Environmental Science

VOLUME=Volume 13 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/environmental-science/articles/10.3389/fenvs.2025.1672787

DOI=10.3389/fenvs.2025.1672787

ISSN=2296-665X

ABSTRACT=Accurate and spatially continuous soil moisture data are essential for applications including numerical weather prediction, hydrological forecasting, and data assimilation. This study evaluates the global ERA5 reanalysis soil moisture (SMERA5) and Essential Climate Variable (ECV) satellite-derived soil moisture (SMECV) against in situ measurements from 2013 to 2015 in Jiangsu Province, China. Five evaluation indices accuracy metrics and Triple collocation methods are used in this study. Taking SMin-situ as the reference, the SMERA5 outperforms the SMECV in terms of correlation coefficient (0.56 for SMERA5 and 0.42 for SMECV) and Triple Collocation (TC) errors (0.01 m3 m-3 for SMERA5 and 0.025 m3 m-3 for SMECV). However, the SMECV can better characterize the soil moisture with smaller random differences (ubRMSD = 0.045 m3 m-3 for SMECV and 0.052 m3 m-3 for SMERA5relative to the SMin-situ data. Both SMECV and SMERA5 exhibit consistent spatial patterns across seasons, although with notable magnitude differences. These two products effectively capture in situ soil moisture (SMin-situ) temporal dynamics in the northern region, while larger discrepancies occur in the southern region. In addition, we evaluate these products from the perspective of soil moisture sensitivity to precipitation. Results show that SMERA5 data more effectively capture soil moisture response to heavy precipitation events than SMECV. Overall, SMERA5 demonstrates superior performance in temporal correlation and precipitation sensitivity, whereas SMECV excels in minimizing random errors. Both datasets exhibit uncertainties linked to sensor limitations and model parameterization, suggesting targeted improvements (e.g., multi-sensor fusion, bias correction) could enhance their reliability.