Comment on hess-2022-76

Abstract

Soil moisture (SM) is a critical component of the water cycle and a key ecological process connecting the soil-vegetation-atmosphere system. Spatiotemporally continuous SM is increasingly demanded for ecological and hydrological research fields. Satellite remote sensing has opened opportunities for mapping SM. Nevertheless, the continuity of SM imagery is hampered by data gaps resulting from inadequate satellite coverage and radio frequency interference. In light of this, we propose a new gap-filling approach to reconstruct daily SM time series using the European Space Agency’s Climate Change Initiative (ESA CCI). The developed approach integrates satellite observations, model-driven knowledge and machine learning algorithm that leverages both spatial and temporal domains. Taking SM in China as an example, we show high accuracy of the reconstructed SM when validated with multiple sets of in situ measurements, with a root mean square error (RMSE) and mean absolute error (MAE) of 0.09–0.14 and 0.07–0.13 cm3/cm3, respectively. Further evaluation with a 10 fold cross validation reveals a median value of the coefficient of determination (R2), RMSE, and MAE of 0.56, 0.025 cm3/cm3 and 0.019 cm3/cm3, respectively. The reconstructive performance is noticeably reduced when excluding an explanatory variable while the rest remains unchanged, as well as when removing the spatiotemporal domain strategy and the residual calibration procedure, respectively. Compared to that using satellite-derived diurnal temperature range (DTR), reconstructed SMs using bias-corrected model-derived DTRs exhibit acceptable accuracies and higher spatial coverage. Applying our gap-filling approach to long-term SM data sets (2005–2015), we show a promising result with a R2 of 0.72. A more accurate trend is achieved relative to that of the original CCI SM when assessed with in situ measurements (0.45 versus 0.23 in terms of R2). Our findings indicate the feasibility of integrating satellite observations, model-driven knowledge, and spatiotemporal machine learning for filling gaps in SM time series over short and long time scales, providing a potential avenue for applications to similar studies.