Atmospheric Model Precipitation Forecast Analysis to Support Reservoir Systems Operation

Abstract Global and regional atmospheric models are important tools for water resources management. Aiming to provide river flow forecast and to support water supply systems operation, this study evaluated the precipitation forecasts from three atmospheric models. Forecasts from ETA (horizontal resolution of 40 km, 10 days range), BAM (horizontal resolution of 20 km, 10 days range) and WRF (horizontal resolution of 5 km, 3 days range) models were analyzed for the Cantareira system basins, on Brazil’s southwestern region. The forecasts were compared to rain gauge and radar observations. Different performance indices were used: Pearson’s correlation coefficient, modified Kling-Gupta efficiency, slope for the linear regression between forecasts and observations, RMSE and PBIAS, in addition to categorical indices fBIAS, POD e FAR. It was found that the correlations were stronger for the first days of the forecast. The strongest correlations were found when comparing the accumulated precipitations for the whole forecast period. Bias analysis indicated ETA strongly underestimated observations, while BAM strongly overestimated, and WRF slightly underestimated. Global and regional atmospheric models rainfall forecasts are an important input for reservoir operation. It’s essential, though, to understand its behavior to minimize inaccuracies and maximize its value in decision making.