| dc.description.abstract | Hydrological forecasting has been an ongoing area of research due to its importance to improve decision making on water resource management, flood management, and climate change mitigation. With the increasing availability of hydrological data, Machine Learning (ML) techniques have started to play an important role, enabling us to better understand and predict complex hydrological events. However, some challenges remain. Hydrological processes have spatial and temporal dependencies that are not always easy to capture with traditional ML models, and a thorough understanding of these dependencies is essential when developing accurate predictive models. This thesis explores the use of ML techniques in hydrological forecasting and consists of an introduction, two papers, and an application developed alongside the case study. The motivation for this research is to enhance our understanding of the spatial and temporal dependencies in hydrological processes and to explore how ML techniques, particularly those incorporating attention mechanisms, can aid in hydrological forecasting. The first paper is a chronological literature review that explores the development of data-driven forecasting in hydrology, and highlighting the potential application of attention mechanisms in hydrological forecasting. These attention mechanisms have proven to be successful in various domains, allowing models to focus on the most relevant parts of the input for making predictions, which is particularly useful when dealing with spatial and temporal data. The second paper is a case study of a specific ML model incorporating these attention mechanisms. The focus is to illustrate the influence of spatial and temporal dependencies in a real-world hydrological forecasting scenario, thereby showcasing the practical application of these techniques. In parallel with the case study, an application has been developed, employing the principles and techniques discovered throughout the course of this research. The application aims to provide a practical demonstration of the concepts explored in the thesis, contributing to the field of hydrological forecasting by introducing a tool for hydropower suppliers. | |
