A Bayesian Spatio-Temporal model to Downscaling State-Level Residential Electricity Data.

Modeling residential electricity consumption is complex due to intertwined environmental and socioeconomic factors, requiring new methods to estimate patterns and provide datasets for diverse stakeholders. This project develops a Bayesian hierarchical Spatio-Temporal model (BHST) to downscale U.S. state-level residential electricity data (2000–2020) to the census tract level, capturing variations and uncertainties driven by climate, land use, building, and household characteristics within local urban scales. By integrating spatial and temporal dependencies, the BHST addresses gaps in high-resolution electricity consumption data quantifying uncertainties in estimates. The model leverages high-resolution datasets, including meteorological climate reanalysis products, land use, and gridded population data, alongside census-derived socioeconomic variables. Using the integrated nested Laplace approximation, the approach efficiently handles computational challenges associated in modeling Bayesian spatial and temporal processes. Model validation involves residual analysis (Moran’s I) and K-fold cross-validation at the state level. At the census tract level, the volume-preserving property is tested using posterior predictive checks to compare that aggregates match when downscaling to census tracts. Initial testing focuses on the Southwest U.S., presenting the theoretical formulation, development, and testing of the BHST downscaling method using data from Arizona, New Mexico, Utah, Colorado, Nevada, and California. This research contributes a robust methodological framework, a detailed analysis of socioeconomic and climatic drivers of electricity uses in the residential sector, and a valuable dataset to advance research, policy, and practice in energy efficiency and climate adaptation.