Streamflow Prediction using a Forecast Combining System

In this work, we describe a forecasts combining system for predicting the monthly streamflows of the Guadalupe river in Colombia; the forecast combining system is composed of four individual experts implementing the following models: autoregresive (AR), autoregresive integrated moving averages, artificial neural networks and singular spectral analysis. We test the methods simple average, weighted average, neural networks, and ANFIS for combining expert forecasts. The combining system is used for streamflow prediction with horizons of one, three and six months ahead. We found that the combination of forecasts using ANFIS outperforms the accuracy of each individual model and the composed forecasts obtained using the other combining techniques.