Short-to-medium range forecast of natural gas use in the United States residential buildings

Abstract

Natural gas is the primary fuel for residential heating in the United States and plays a critical role in the transition to zero-carbon energy. In this study, a systematic framework is developed to predict short-to-medium-range (weeks to months) natural gas consumption in U.S. homes based on an interpretable model of gas consumption defined from data at the county level and seasonal temperature forecast ensembles from the European Center of Medium Range Weather Forecast. Linear-plus-plateau models were fitted utilizing pipeline delivery of gas to each county, and outdoor temperatures to predict daily, weekly, and biweekly residential gas use for 1000 individual counties in the U.S. and five larger regions. The model presented highly accurate short-range (1 week) and robust medium-range (3 months) gas consumption forecast capabilities with an average accuracy of 80%. The highest model performances were found in the Midwest and East Coast regions (up to 82% accuracy), where temperature significantly drives gas consumption. This study assesses and quantifies how temperature forecast uncertainties impact gas consumption predictions for county and regional gas consumption forecasts using numerical weather forecast ensembles. Forecasting at a fine resolution over such an extensive spatial range is important for policymakers to design well-informed policies adapted to local conditions and impactful for gas savings at the national scale. The discussion presents the potential of this framework in aiding energy planning, supply chain optimization, and anticipating weather-driven gas demand fluctuations.