An alert system for flood forecasting based on multiple seasonal holt-winters models: a case study of southeast Brazil

Floods are natural phenomena that can cause social and economic damage. Predicting these events is crucial for preventing significant losses. The city of Volta Redonda in the state of Rio de Janeiro is particularly vulnerable to floods because it is traversed by the Para & iacute;ba do Sul River. Therefore, this work proposes the development of a flood alert system for a city based on time series analysis using hourly flow data. Two seasonal forecasting models were analyzed to construct the alert system: the SARIMA and Holt-Winters-Taylor (HWT) models with dual seasonality (daily and weekly), with the latter demonstrating a better fit, with a mean absolute percentage error of 1.69%. The key curve methodology contributes to determining river levels from forecasted flows. Using such methods and tools, a flood alert system with level intervals was developed to signal a river's flood behavior. The program has proven reliable in timely alerting individuals and relevant institutions, thereby preventing further harm to the population.