Answers are grounded in this paper and cite their sources.
Forecasting soil moisture (SM) is challenging because of its spatial and temporal variability and its sensitivity to changing environmental and climatic conditions. Accurate SM forecasting is important for water-resource management and agricultural sustainability. In this study, we present a graph-based approach to forecast SM using graph neural networks (GNNs) for Karnataka, India, using multi-source data from April 2015 to December 2021. We developed static and dynamic temporal SM GNN (T-SM-GNN) variants by integrating graph convolutional network (GCN), graph attention network (GAT), combined graph convolutional and attention network (GCAT, and GACN) architectures with long short-term memory (LSTM) networks for spatiotemporal SM forecasting. The proposed models were evaluated against six baseline models from the literature. Among the evaluated models, GAT-based T-SM-GNN achieved the best overall performance, with the dynamic variant attaining an RMSE of 0.022 ± 0.000 m³ / m³ and R² of 0.919 ± 0.003 , corresponding to a 4.34% reduction in RMSE compared with the strongest baseline, convolutional long short-term memory (ConvLSTM). These results highlight the potential of graph-based spatiotemporal modelling for regional-scale SM forecasting.
No takes yet. Share an insight, caveat, or question.
Sudhakara et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: