Speaker
Description
The difficulty of installing soil moisture sensors in the deeper layers of the root zone represents a major limitation for both hydrological modeling and irrigation management. In this study, we propose a predictor–corrector Physics-Informed Neural Network (PINN) framework to estimate soil water content at 60 cm depth using solely measurements collected at 30 cm. The predictor–corrector architecture enables a physically consistent data augmentation strategy, whereby additional informative samples are generated over a depth domain spanning from 0 to 60 cm, allowing the transition from Neumann to Dirichlet boundary conditions and the progressive refinement of the predictions through successive training stages. The results show that the reconstructed soil moisture at 60 cm closely reproduces the observed dynamics, highlighting the capability of the proposed approach to recover soil moisture profiles even in the absence of direct measurements. This methodology therefore provides a practical solution for extending observational datasets in real-world applications, particularly when in situ sensors or satellite products are limited to shallow soil layers.