The paper addresses the use of an adaptive, recursive fuzzy modeling based on the notion of level set to forecast monthly streamflows of a major hydroelectric power plant reservoir at the northeast of Brazil. Streamflows are highly complex nonstationary time series with high variability during the year, a feature that turns modeling and forecasting very hard and challenging. The adaptive level set method is evaluated against periodic autoregressive moving average models, currently adopted by many power industries, and against granular, neural, neural fuzzy, recurrent neural, and data driven level set models. The results show that adaptive level set modeling achieves the best root mean square error performance, surpassing all the models considered herein.