Distinguishing environmental controls on strong vs. extreme wind gusts

Statistical and theoretical models of wind gusts may be dominated by more common strong events, rather than rare but damaging extreme ones. We address this by combining case studies of six extreme gust cases in New South Wales (NSW), Australia, with statistical and machine-learning (random forest) models to identify environmental factors distinguishing “strong” ($\ge 18\mathrm{m/s}$) vs. “extreme” ($\ge 25\mathrm{m/s}$) gust events in a 20-year dataset. The BARRA-SY high-resolution regional reanalysis is used to augment in-situ observations and provide a model gust speed diagnostic for evaluation, as well as environmental prediction metrics. All the extreme wind cases were linked to deep convection, often organized into linear systems. A random forest model achieved 89% accuracy for predicting strong winds generally, with the gust diagnostic and environmental background wind speeds as the top predictors. For distinguishing extreme from strong gusts, the model’s accuracy was 79%, but with a high false alarm rate. Both statistical and machine-learning analyses highlight convective instability metrics — Most Unstable Convective Available Potential Energy (MUCAPE), Derecho Composite Parameter (DCP), and k_index - as key predictors of extreme gusts. The BARRA-SY gust speed diagnostic thus informs about strong wind gusts, but not extremes, which depend on variables it ignores. Instability measures, however, are also imperfect predictors of extreme gusts because they fail to capture storm trigger conditions, seen in some of the case studies. These findings demonstrate that the factors driving extreme wind gusts differ substantially from those driving strong but less extreme gusts. Therefore, statistical analyses or predictive models that consider all strong gusts collectively will likely fail to uncover the environmental factors responsible for the most extreme events with greatest impact.