Extracting value from seasonal forecasts for droughts: an approach based on persistence rules.

Long-term planning is part of good water resource management. This management takes place according to the needs of society, such as optimizing energy production, providing water for agriculture, supplying industry and guaranteeing urban water supply. The quantification of water resources and their long-term forecast are tools that help with this planning. In this work, we evaluated the ability of seasonal streamflow forecasts to detect droughts periods based on forecast evaluation rules of a Hydrologic Ensemble Prediction System (H-EPS), and compare them with the benchmark Ensemble Streamflow Prediction (ESP). Seasonal forecasts of seven months horizon were generated using the MGB-AS model forced with ECMWF seasonal precipitation forecasts (SEAS5), for basins larger than 1000 square kilometers. The focus of the study were the 153 hydroelectric plants of the Brazilian National Interconnected System (SIN), which affluent flows were studied for the period of 2007 to 2016.  For each plant, drought forecasts were analyzed using the Area Under the Receiver Operating Characteristics curve (AUROC). The drought threshold was defined as the 90th percentile of the flow, using data from 1979 to 2006. Exceedance diagrams were made, where each forecast horizon was represented with the percentage of members that indicated the occurrence of a dry month. Then, contingency tables were set up, considering the drought detection criterion as at least one member indicating drought in the seven horizons and six months with at least 20% of the members indicating drought. The rule was elaborated based on data from the Itaipu power plant (Paraná River) and applied to all plants. Through the results, it was observed that the H-EPS forecasts were more accurate in detecting droughts than the benchmark. In the regional analysis of the results, the rule chosen for Itaipu was also suited for the plants in the southeast region. This may have occurred because these hydroelectric plants rivers present similar hydrological behavior, related to the type of soil, evapotranspiration rates, precipitation, climate and similar relief. Our next steps involve the creation, testing and analysis of more locally and temporally optimized rules, including some that consider only the first months predicted in the analysis.