Short-Term Urban Water Demand Forecast Models in Action: Challenges from Model Development to Implementation to Real-Time Operations

Accurate short-term water demand forecasts are an important component of sound decision making by Tampa Bay Water, Florida's largest wholesale water supplier. Tampa Bay Water and its members serve more than two million people in the Northern Tampa Bay Tri-county area with over 250 million gallons of water per day on average annually. Week-to-week water demand variability follows complex spatial and temporal patterns that are not only dependent on prevailing and/or upcoming weather conditions, but also on weather-induced behavior associated with outside use of water. Auto-Regressive Moving Average exogenous (ARMAX) aggregated models that forecast weekly water supply demands at spatially variable Water Demand Planning Areas (WDPA) were developed. The forecasts from these models further pass through disaggregation algorithms to provide estimates at specific point of connections (POC) within the WDPAs. Relatively complex models were identified which meet the desired criteria during model fitting and validation steps; however, these models did not always produce the best results during real-time operations. This is due to different challenges being faced during real-time operations that were not present during model development including missing data for one or more independent variables and data being "old." In addition, we also found that the ease of implementation plays a major role in selecting the best model out of a suit of models whose performances are nearly identical. Details of the approach and results will be discussed in this paper.