Disseny d'un model basat en tècniques d'aprenentatge automàtic per predir el cabal del riu Ter

Abstract

[en] It is obvious that one of the biggest problems of this 21st century is climate change, with floods, droughts and violent storms becoming more common. In addition, in Catalonia, the Mediterranean climate is one of the most affected areas of this global crisis. This and the constant and growing need to send water to the metropolitan area of ​Barcelona due to high consumption, mean that the Generalitat has to manage and regulate the flows of different rivers in Catalonia. In our case, we focus on the Ter river, a river that is born in Ulldeter (Pyrenees) and flows into Estartit (Mediterranean), and is constantly in the spotlight for various reasons. Among other things, floods and floods have accompanied the recent history of the river. However, we are aware that the river is undergoing very meticulous management by public administrations. Inspired by the Data4Good movement, which brings together projects that want to solve real problems from open data, with the common benefit and social commitment as the ultimate goal. In this work we provide an analysis of the river and a methodology based on Machine Learning to predict its flow in 24 hours. During this project, open data from the Catalan Water Agency and the Catalan Meteorological Service regarding the Ter river basin are analyzed in detail and processed. This data is cross-referenced and used to train different machine learning models. These models attempt to predict the flow of the river Ter at some strategic points in the river. By studying them we extract the relationship that a weather station has on the impact on the flow, locating the most influential areas. Precipitation plays a direct role with river flows, which is why it is so important to be able to use meteorological data in models. All this allows us to extract valuable new information about the behavior of the river, which helps us to understand it better, so that, we can manage it more efficiently. This work lays the foundation stone for the use of machine learning techniques to support the management of the river Ter. It also points out lines of work of interest to continue to improve this type of model and export it to other rivers.