Aplicación de ‘aprendizaje profundo’ para el pronóstico de precipitación a partir de datos de reflectividad de radar meteorológico

Abstract

The studies published on rain forecasting in the South American region using Deep Learning techniques are very scarce. The available hydrometeorological monitoring networks, usually rain gauge networks, have not provided sufficient data to achieve satisfactory results in the prediction of these patterns (Bendix et al., 2017). In this work, Deep Learning techniques are applied to face this problem. Using information collected from a network of meteorological radars that exist in Ecuador, RadarNet-Sur, this work applies deep learning techniques to the mentioned information and proposes a methodology for rain forecasting. The methodology presented consists of three steps, radar image prediction with Deep Learning techniques, a transformation of the former’s output to precipitation terms, and, interpretation of the results obtained. The results lend themselves to a discussion of how to improve the quality of the prediction obtained. This, despite working with a limited data set, so that there is a possibility of improving the model if working with a representative set. However, the model is not immediately applicable because it does not learn all the existing relationships and patterns for the test set. This is why some solutions to be carried out in future works are discussed that could significantly improve the performance of the model.