Speaker
Description
Soil moisture is an essential climatic variable, significantly influencing ecological and agricultural processes. Its measurement at various geographic scales, from regional to local, has been extensively studied, particularly through the estimation of soil water content. Traditionally, this estimation relies on thermo-hydrodynamic transfer equations, which require specific data on soil properties. However, such data is costly to acquire, and their application is often limited by the geometry of the medium, being most effective at the scale of a soil column. Extending this approach to larger scales, such as agricultural plots, becomes complex due to the heterogeneity of soil properties. This communication explores the use of four deep learning approaches to estimate the volumetric water content (θ) at different depths (10, 20, 30, and 40 cm) on the experimental site of Saaba. The main objective is to move beyond the traditional physical models based on partial differential equations by using machine learning models, particularly deep learning. We compare the performance of the deep learning architectures and provide a ranking of the best predictors. The data used for this study include weather measurements and soil temperature at four depths measured directly in the field. This study highlights the potential of deep learning techniques to overcome the limitations of traditional methods and provide more accurate estimates of soil water content, with practical applications in water resource management and precision agriculture.