Predicting root zone soil moisture using observations at 2121 sites across China

Root zone soil moisture (RZSM) is particularly useful for understanding hydrological processes, plant-land-atmosphere exchanges, and agriculture- and climate-related research. This study aims to estimate RZSM across China by using a one-parameter (T) exponential filter method (EF method) together with a randomforest (RF) regionalization approach and by using a large dataset containing in situ observations collected at 2121 sites across China. First, at each site, T is optimized at each of four soil layers (10-20 cm, 20-30 cm, 30-40 cm and 40-50 cm) by using 0-10-cm soil layer observations and the corresponding calibration layers. Second, an RF classifier is built for each layer according to the calibrated T values and 14 soil, climate and vegetation parameters across 2121 sites. Third, the calibrated T at each soil layer is regionalized with an established RF classifier. Spatial T maps are given for each soil layer across China. Our results show that the EF method performs reasonably well in predicting RZSM at the 10-20-cm, 20-30-cm, 30-40-cm and 40-50-cm layers, with Nash-Sutcliffe efficiency (NSE) medians of 0.73, 0.52, 0.38 and 0.27, respectively, between the observations and estimations. The T parameter shows a spatial pattern in each soil layer and is largely controlled by climate regimes. This study offers an improved RZSM estimation method using a large dataset containing in situ observations; the proposed method also has the potential to be used in other parts of the world.

Automated summary

This study estimates the root zone soil moisture (RZSM) across China using an exponential filter method and a randomforest regionalization approach. The results show that the method performs well in predicting RZSM in different soil layers, and the T parameter is controlled by climate regimes.