Using logistic regression-cellular automata to project future sites for commercial wind energy development

Abstract

Demands to decarbonize electricity production and improve energy security will continue to drive wind energy development. A Logistic Regression-Cellular Automata (LRCA) model is presented here to project timing and location of this future development and thus aid efforts to meet energy demands. The model's logistic regression equation is trained and tested using aggregated data from key predictors to correctly classify hexagonal grid cells covering the Conterminous United States (CONUS) as currently containing wind farms. The cellular automata component iteratively applies this equation, plus constraints and neighborhood effects, to project grid cells suited for future wind energy development out to the year 2050, with the model's sensitivity to constraint, neighborhood, and scenario definitions also examined. Projected wind farms are concentrated in high wind speed regions currently populated by wind farms (e.g., Central Plains, Midwest). State-level scale projections reveal local influences on future development, such as critical species habitats and infrastructure. Projected wind farm locations are trustworthy since the model correctly classifies over 85% of current grid cell states at CONUS and state-level scales. Current clusters of wind energy development across the CONUS will continue to expand in these projections, with these clusters growing earlier and faster given a larger neighborhood size and looser constraints. Model projections are less sensitive to scenario definitions, with modifications to predictors affecting when existing wind farm clusters expand rather than where new clusters form. Replacement of the wind farm location dataset would allow this model to project other decentralized land use changes, particularly solar energy development.