Root zone soil moisture estimation with Random Forest

Accurate estimates of root zone soil moisture (RZSM) at relevant spatio-temporal scales are essential for many agricultural and hydrological applications. Applications machine learning (ML) techniques to estimate limited compared commonly used process-based models based on flow transport equations in the vadose zone. However, data-driven ML present unique opportunities develop quantitative without having assumptions processes operating within system being investigated. In this study, Random Forest (RF) ensemble algorithm, is tested demonstrate capabilities advantages RZSM estimation. Interpolation extrapolation a daily timescale was carried out using RF over small catchment from 2016 2018 situ measurements. Results show that predictions have slightly higher accuracy interpolation similar comparison with simulated model combined data assimilation. extreme wet dry conditions were, however, less accurate. This inferred be due infrequent sampling such led poor trained incomplete representation subsurface study sites covariates. Since does not depend parameters required water flow, it more advantageous than data-poor regions where hydraulic or missing, especially when primary goal only estimation states.