Pronóstico de alturas en cursos de llanura mediante el uso de un modelo de caja negra

Abstract

Floods are the most common disaster in our country, producing the largest number of affected and damaging infrastructure and private property. In this paper, a black box model called functional networks is presented. This model was used to forecast water levels in flatland courses and was applied in the Gran Rosario basins. The input variables are rainfall and water level linked to a time t0, while the output is given by predicted water levels associated with different time horizons tpi. From the observed events, on average 10 for each gage, all combinations are calculated to form two groups: learning and validation. Model evaluation is done through various statistical index, including: relative and absolute maximum difference in peak level, coefficient of efficiency of Nash-Sutcliffe, root mean square error and coefficients of the regression line. For the results presented in this paper mean values in difference peak level for 6 hours forecasting was 0.27 m in learning and 0.33 m in validation. The potential of the model is that it can be applied in any basin with precipitation data and levels.