Author(s): Diego Silva Piedra; Monica Fossati; Pablo Ezzatti

Linked Author(s): Mónica Fossati

Keywords: Forecasting; Machine learning; LightGBM; Rio de la Plata

Abstract: PronUy_RPFM is an operational tool designed by researchers from the Institutes of Fluid Mechanics and Environmental Engineering (IMFIA) and Computing (INCO) at the Facultad de Ingeniería (FIng), Universidad de la República (UdelaR) in Uruguay, to predict tide levels in the Río de la Plata and its continental shelf. The tool provides hourly sea level forecasts for the next three days, which are publicly available on its dedicated website. The forecast is originally based on the TELEMAC2D finite element numerical model, designed to resolve 2D shallow water equations. This study, in turn, presents the advances in the application of machine learning (ML) methods within the forecasting stage of PronUy_RPFM. Specifically, the Light Gradient Boosting Machine (LightGBM) framework was implemented for tide level forecasting, resulting in the new models: PronUy_LightGBM1 and PronUy_LightGBM2. We detail the structure and present the results of these new models, as well as the data used for their training. The effectiveness of both models is evaluated by comparing their predictions with measured data from tide gauges located along the Uruguayan and Argentine coasts.

DOI: https://doi.org/10.64697/iahr.proc.hic2026.162

Year: 2026