A Tarso model for studying the level of the Cuiabá river

Abstract

The Cuiabá River registers cyclical occurrences of floods and droughts over the years, according to the hydrological periods, thus characterizing a non-linear behavior. The prediction of the level of the Cuiabá river is important to help institutions such as the Civil Defense of the state of Mato Grosso and many other institutions that are concerned with the prevention and mitigation of natural disasters. Thus, this study considered the nonlinear Threshold Autoregressive Self-Excking Open-loop (TARSO) model with 2 regimes, with a Bayesian approach. We tested models to which values of the linimetric quota (riverwater level in millimeters) with and without rainfall (mm) were associated. All models were compared using the lowest DIC, MAPE and MSE criterion, and the TARSO (2; 1, 0, 3, 1, 1) model performed best according to these criteria. Finally, the selected model was shown to produce reliable predictions.