Rainfall infiltration induces transient seepage responses in slopes, but continuous reconstruction of internal hydraulic fields remains difficult under nonlinear unsaturated flow, complex boundaries, and sparse observations. This study develops a physics-informed neural network (PINN) framework for reconstructing transient total-head fields in slopes under rainfall infiltration. SEEP/W benchmark simulations based on the Richards equation and the van Genuchten–Mualem model were used to generate reference fields and synthetic sparse observations. The effects of monitoring layout were first examined in a non-layered benchmark slope by comparing installation-depth layouts and vertical monitoring-array layouts. The recommended layout was then applied to layered heterogeneous slopes to evaluate its applicability under different hydraulic conductivity structures. The results show that the PINN framework can recover the dominant transient seepage pattern from sparse observations. Shallow observations provide the strongest constraint on rainfall-induced near-surface responses, middle-depth observations supplement internal hydraulic transmission, and deep observations mainly constrain the lower background field. Two vertical monitoring arrays provide a practical balance between reconstruction stability and monitoring economy, while an additional array mainly improves local details. In layered heterogeneous slopes, the recommended layout remains effective for reconstructing the main total-head pattern, although local errors increase near layer interfaces, slope-shoulder transitions, slope toes, and preferential-flow zones. These findings provide a physically interpretable basis for sparse monitoring layout design and seepage-field reconstruction in rainfall-induced slopes.
Yang et al. (Mon,) studied this question.