Enhancing soil moisture estimation in alfalfa root-zone using UAV-based multimodal remote sensing and deep learning

Accurate estimation of soil moisture content (SMC) is essential for optimizing irrigation schedules and identifying drought-tolerant varieties. The integration of unmanned aerial vehicles (UAVs) with advanced sensors provides a novel method for monitoring SMC with high flexibility, resolution, and performance. This study utilized UAVs to capture RGB, multispectral, and thermal imagery of alfalfa (Medicago sativa L.) at the Linze Grassland Agricultural Experiment Station, Lanzhou University, and to evaluate the potential of fusing multimodal UAV data for SMC estimation in the root zone of densely and uniformly distributed leafy plants, using alfalfa as a case study, within a deep learning framework. Results showed that combining multimodal data—encompassing canopy spectral, structural, thermal, and textural information—significantly improved SMC estimation accuracy. Among the four regression models evaluated—partial least squares (PLSR), support vector machine (SVM), random forest (RF), and deep neural network (DNN)—the DNN model achieved the highest accuracy in overall multimodal data fusion, with a coefficient of determination (R²) of 0.72 and a root mean square error (RMSE) of 4.98%. It demonstrated good predictive performance for both full and deficit irrigation scenarios, with R² values of 0.74 and 0.75, respectively. The DNN model also provided reliable SMC estimates across the three alfalfa canopy types, with R² values of 0.72, 0.74, and 0.58, respectively. Moreover, it exhibited superior accuracy under both irrigation regimes and demonstrated strong spatial adaptability, characterized by low spatial dependence and autocorrelation. In conclusion, the DNN model based on UAV-derived multimodal data fusion offers a reliable and robust approach for SMC estimation, providing valuable insights for irrigation management at farmland-scale.