Short-term Forecasting of Wind Gusts Across CONUS Using Machine Learning
16 years of wind gust data across the US are used to train and test machine learning models to make short-term predictions of all wind gusts, intense wind gusts and strong wind gusts using machine learning and upper-level predictors drawn from ERA5. Stepwise procedures instruct predictor selection, and resampling is used to quantify model stability. A new technique is presented to objectively identify optimal neural-network complexity. We also advance understanding of which predictors are critical to forecast accuracy of wind gust occurrence and magnitude.
Wind gusts are a major natural hazard. We present and evaluate wind gust predictions derived using machine learning. We show artificial neural networks with 3-5 hidden layers exhibit skill for short-term forecasting of wind gust magnitude across a variety of wind climates, but deep and high-complexity artificial neural networks do not overcome the predictability ceiling for very high-intensity events.
Short-term forecasting of wind gusts, particularly those of higher intensity, is of great societal importance but is challenging due to the presence of multiple driving gust generation mechanisms. Wind gust observations from eight high-passenger-volume airports across the continental United States (CONUS) are summarized and used to develop predictive models of wind gust occurrence and magnitude. These short-term (same-hour) forecast models are built using multiple logistic and linear regression, as well as artificial neural networks (ANNs) of varying complexity. A suite of 19 upper-air predictors drawn from the ERA5 reanalysis and an autoregressive (AR) term is used. Stepwise procedures instruct predictor selection, and resampling is used to quantify model stability. All models are developed separately for the warm (April-September) and cold (October-March) seasons. Results show that ANNs of 3-5 hidden layers (HLs) generally exhibit higher hit rates (HRs) than logistic regression models and also improve skill with respect to wind gust magnitudes. However, deeper networks with more HLs increase false alarm rates (FARs) in occurrence models and mean absolute error (MAE) in magnitude models due to model overfitting. The skill of all models is critically dependent on the inclusion of the AR term, while the majority of the remaining skill derives from wind speeds and lapse rates. A predictive ceiling is also clearly demonstrated, particularly for the societally important strong and damaging gust magnitudes, which appears to be partially due to ERA5 predictor characteristics and the presence of mixed wind climates.