This study aims to estimate multi-layer soil moisture (SM) in 30-cm increments up to 180 cm by integrating multimodal sub-field and remote sensing data and applying machine learning algorithms. PlanetScope (optical sensor) data, climatic variables, and in situ soil properties were used as input features to estimate SM at 3-m spatial resolution. The extreme gradient boosting model was trained and tested using two approaches: (1) using only input features and (2) by adding predicted SM from the adjacent upper layer to (1). Raw remote sensing bands (blue and near-infrared), vegetation indices (normalized difference vegetation index green band and visible atmospheric resistant index), climatic variables (precipitation reference evapotranspiration), and soil temperature were more influential in surface SM estimation. In contrast, soil properties (texture, organic matter, and soil carbon) were key predictors for deeper layer SM. The R-2, root mean squared error (RMSE), and ratio of RMSE to observed mean (RRMSE) values for SM predictions across six depths varied from 0.78 to 0.89, 0.021 to 0.028 cm(3)/cm(3), and 11.7% to 14.6%, respectively. Incorporating adjacent upper-layer SM information increased R-2 by 8%-24% and reduced RMSE by 10%-27% at various depths. Importantly, our results show that using remote sensing data and easily obtained topographic and climate data alone (excluding soil property input features) captured a considerable fraction of root zone soil moisture (RZSM) variability with depth (R-2 of 0.66-0.77), a practical alternative in RZSM analysis applications with limited or no soil property data. Additionally, model performance enhanced by adding adjacent upper-layer SM input features can generate high-resolution (3 m), multi-layer (to 180 cm) SM maps, offering a new tool for precision agricultural water management and other applications requiring high spatiotemporal resolution SM data.