A COMPARISION USING STATISTICAL AND MACHINE LEARNING METHODS FOR STREAMFLOW TIME SERIES

• Autor(a) principal: Villavicencio, Lourdes Milagros Mendoza
• Data de Publicação: 2020
• Outros Autores: Mendes, David, Monteiro, Felipe Ferreira, Andrade, Lara de Melo Barbosa, Silva, Cássia Monalisa dos Santos
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Revista Brasileira de Climatologia (Online)
• Texto Completo: https://revistas.ufpr.br/revistaabclima/article/view/70245
• topic: Time-series analysis; Streamflow Forecasting; Neural Networks.

This study was carried out in the Sibinacocha lake watershed in the Peruvian Andes. In this region the long-term meteorological data are scarce and there are few studies of flow forecasts. Based on this evidence, in this study we present the monthly flow simulation, using statistical models and data-oriented model, with the purpose of evaluating the performance of these methodologies. The results of the comparative statistical analyses indicated that the data-oriented models, specifically the Recurrent Neural Networks, provided great improvements over the other models applied, specifically the ability to capture the minimum and maximum monthly flow, resulting in excellent statistical values (R2=0.85, d=0.96), thus suggesting this methodology as a possible application for flow forecasts.