Abstract
In view of the problems related to flood forecasting and the different types of neural networks that can be applied, this study aims to analyse which neural networks and statistical metrics have been most used between the years 2015 to 2021 to prediction of flooding in river systems. A bibliographical survey was carried out in the Web of Science database and the articles were separated according to taxonomy: feedforward, recurrent, hybrid and others, as well as the most common error metrics in each article and the countries of each study. A total of 118 articles were obtained. Feedforward networks were applied in 87% of the articles, recurrent networks 20%, hybrid 12% and others 3%. It was found that 75% of the total number of articles applied Root Mean Square Error (RMSE) in their analyses, the Nash-Sutcliff Coefficient (NSE) 56%, the Coefficient of Determination (R²) 41%, mean absolute error (MAE), correlation coefficient (r) and Mean Square Error (MSE) obtained a total of 36%, 28% and 13% respectively. Through this study it is possible to expand and disseminate the main studies that deal with this approach in a more targeted way, seeking to update and improve research and guide public policies for urban planning and management as a way of reducing economic and social losses. and medium and long-term environmental impacts with precision and effectiveness. Overall, ANNs offer a powerful tool for authorities to enhance operational capabilities and strategic planning for flood predictions, ultimately improving resilience and reducing the impact of flooding on communities and infrastructure.
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call