Estimating multi-depth daily soil moisture at 10 m resolution using SMAP SSM and Sentinel-1/2 data based on random forest regression algorithm

Soil moisture (SM) is a critical variable within the hydrological cycle, closely linked to weather patterns and climate change. However, the limited availability of high-resolution SM datasets, constrained by the resolution of satellite sensors, poses significant challenges for field-scale research, which is essential for agricultural management and precision irrigation planning. This study addresses this limitation by estimating daily SM across surface (0-5 cm), active rootzone (0-30 cm), and unsaturated zone (0-120 cm) depth at a spatial resolution of 10 m. A random forest (RF) regression model was developed using in-situ SM measurements as the training dataset. The predictor variables included meteorological parameters, topographic features, soil texture properties, Sentinel-1 synthetic aperture radar (SAR) signals, groundwater depth, and vegetation indices derived from Sentinel-2 imagery.

Model predictions were validated in the Twente and Raam regions of the Netherlands over a one-year period (2023-05-18 to 2024-05-18), using independent observations from six in-situ SM stations that were excluded from both the training and testing phases. The results indicated strong model performance, with unbiased root mean square error (ubRMSE) values ranging from 0.03 to 0.08 cm³/cm³ and Pearson correlation coefficients (R) from 0.71 to 0.90. Comparisons with the European Space Agency Climate Change Initiative (ESA CCI) SM product further corroborated the model’s accuracy. While the model effectively captured daily SM dynamics, particularly during winter months, some discrepancies, such as over- or under-estimations, were noted during the summer. These high-resolution SM estimates provide valuable insights for precision agriculture and hydrological research, enhancing decision-making processes in these fields.