Unveiling multiscale drivers of wind speed in Michigan using machine learning

Abstract

The Great Lakes region has a unique climatology due to its large water bodies. Dynamic seasonal wind speeds are an important component in this climate that requires further study. Using 10-m wind data from ERA5-Land, this study employs remote teleconnection indices and local climate features to predict low-level wind speeds using Extreme Gradient Boosting (XGBoost) machine learning. The model for monthly winds achieves high accuracy, with an R² of 0.96 and a Root Mean Square Error (RMSE) of 0.12 m/s⁻¹. The Shapley Additive Values (SHAP) analysis reveals that local climate variables, including the proximity to the nearest Great Lake, surface roughness, and surface temperature, are the most influential predictors and are most important in the model. Teleconnections such as the El Niño-Southern Oscillation and the Arctic Oscillation play minor roles. This study provides a new multiscale perspective on wind speed characteristics, drivers, and insights into the region’s wind energy potential.