Forecasting municipal water demand based on the Global Ensemble Forecast System

Climatic variables have a great effect on municipal water demand. The objective of this study was to improve water demand forecasts using retrospective forecast (reforecast) analogs of several climatic variables including weekly total rainfall, number of rainy days, number of consecutive rainy days, and number of hot days. The reforecast of the Global Ensemble Forecast System (GEFS) was used with both forecast analog and K nearest neighbor (KNN) approaches to generate ensemble forecasts. Ensemble forecasts of water demand were generated using an Auto Regressive model with External Inputs (ARX) water demand model with the input of the analogs or KNN of the climatic variables. The probabilistic forecast skill for both climatic variables and water demand forecasts were evaluated using the rank probability skill score (RPSS) in different months over the Tampa Bay region. The analog approach generally showed higher skill than the KNN approach for forecasting both climatic variables and water demand at all lead days over the Tampa Bay region. The forecast skill for both climatic variables and water demand were generally positive up to lead day 7 throughout the year with the winter months showing longer skillful lead days than the summer months. The temperature-based climate variables showed higher skill than the rainfall-based climate variables.