Development of a Neural Network System for Predicting Topsoil Moisture Using Remote Sensing and Rainfall Forecast Data

With only 1% of the planet's water resources accessible in rivers and lakes, and with droughts on the rise, the scarcity of fresh water is becoming an increasingly serious problem for the world's population. Of this 1%, 70% is used by the agricultural sector, which tends to consume more of this scarce resource due to population growth and the resulting pressure to increase agricultural production. One of the fundamental variables in this context is soil surface moisture, which is particularly important for the management of water resources and has a direct impact on the interaction between the soil and the atmosphere. Therefore, this work discusses the current advances made in the measurement of this variable, which is fundamental for precision agriculture, and creates a model capable of predicting topsoil moisture in a study basin located in the southeast of the Iberian Peninsula, characterised by its predominantly agricultural. To develop the model, we used topsoil moisture data from the Sentinel-1 satellite, probes distributed over the study area, precipitation measurements from meteorological stations and ensemble forecasts from meteorological models. The final model predicted surface soil moisture (SSM) with an average Root Mean Squared Error (RMSE) of 0.15 and a maximum difference between observed and predicted data of 1.37%, demonstrating the ability to predict from one to ten days in advance.