Integración de un sistema de alerta temprana mediante modelación hidrodinámica y predicción de flujos con redes neuronales. Caso de estudio: río Tomebamba

Abstract

The urban area of the city of Cuenca has been affected by the flooding of its rivers causing flooding in residential areas and with it, material losses or human lives. Therefore, knowing in advance the flow that will cause damage is important. For that reason, the objective of the study is to integrate an early warning system to forecast daily flows and assess the vulnerability to floods of the Tomebamba River. First, a multilayer perceptron back-propagation neural network was constructed with Matlab to forecast flows up to three days using rainfall and flow data. From a qualitative (graphical comparison) and quantitative (MSE, NS, R and PBIAS) evaluation, a model capable of predicting flows up to one day was obtained. Nonetheless, there were overestimations of low flows and underestimations of high flows in all models. Secondly, a river modeling was performed with HEC-RAS for calibration and with Iber to model six flood hydrographs, which resulted in the location of overflow points, maximum areas, alerts and flood maps of the area. Hence, the two models together constitute an integral system for flood forecasting on the Tomebamba River. Wherein, the predicted flow is compared with established thresholds to have an approximation of the damages that could cause. By this means, establish the necessary actions for its dissemination and alert by the body that manages the system.