Soil Moisture Prediction Using Multi-Source Datasets and Graph Neural Network

In this study, we introduce a novel method for predicting soil moisture (SM) in the Bidar district of Karnataka by leveraging a combination of advanced Machine Learning (ML) techniques and remote sensing data. The study utilizes multi-source inputs, such as MODIS satellite data, soil properties, elevation, and precipitation, to predict the SM. Our methodology implements state-of-the-art graph neural networks (GNNs), particularly Graph Convolutional Networks (GCNs), in combination with Long Short-Term Memory (LSTM) networks to capture both spatio-temporal dependencies in the data. The proposed model is compared with LSTM and CNN (Convolutional Neural Network)-LSTM models and experimental results demonstrate that the GCN-LSTM model outperforms other approaches, achieving an $R^{2}$ value of 0.9152, a mean absolute error (MAE) of $0.0240 \text{m}^{3} / \text{m}^{3}$, and a root mean square error (RMSE) of $0.0322 \text{m}^{3} / \text{m}^{3}$. These findings highlight the potential of graph neural networks in enhancing the accuracy of SM predictions, providing valuable insights for agricultural and water resource management in the region.