Integrating Geospatial Data for Weather Station Selection and Enhanced Electric Load Forecasting

Abstract

Weather is an essential source of data for forecasting electricity demand, with the most suitable weather stations varying across regions. The access to reliable gridded weather data at high spatial resolution worldwide is growing, providing a steady stream of more granular data with which to forecast. However, existing virtual-weather station methods have only accounted for point based data and do not leverage the big data available from geospatial weather forecasts. Strategic use of gridded weather data can potentially bring significant improvements to forecast accuracy. The paper proposes a new data integration framework to leverage gridded weather data to improve forecast accuracy via the creation of geospatial virtual-weather stations. The benefit such data can provide is demonstrated across two system operator planning areas on different continents. In addition to improving forecast performance, gridded weather re-analysis data can be used to help electric load forecasters prioritize weather stations for enhanced temporal resolution and data maturity and to identify locations for new weather stations. In the future, as more utilities and system operators produce probabilistic load forecasts to capture uncertainty, use of this framework with gridded ensemble weather data can help maintain the sustainability of accurate load forecasts.