Skip to main content
U.S. flag

An official website of the United States government

Publication Date
1 May 2022

Wind Gust Forecasting Using Machine Learning

Print / PDF

We present a rigorous investigation of possible skill enhancement in the forecasting of gust occurrence and magnitude from artificial neural networks (ANNs) v. multiple predictor logistic and linear regression. ANNs offer skill enhancement, particularly for strong and damaging wind gusts. However, deeper networks are vulnerable to overfitting and still under-predict high-intensity gusts even when an auto-regressive (AR) term is included.


Wind gusts are a major source of weather-induced societal impacts. We present analyses of causes of wind gusts and mechanisms to enhance short-term predictability using machine learning tools. The method is illustrated using AOS observations of near-surface wind gusts and predictors drawn from the ERA5 reanalysis.


Geophysical predictors from the ERA5 reanalysis are used in conjunction with an autoregressive term in regression and ANN models with different predictors, and varying model complexity. Models are derived and assessed using 16 years of hourly observations for the warm (April–September) and cold (October–March) seasons for three high passenger volume airports in the United States. Model uncertainty is assessed by deriving models for 1000 different randomly selected training (70%) and testing (30%) subsets. Gust prediction fidelity in independent test samples is critically dependent on the inclusion of an autoregressive term. Gust occurrence probabilities derived using five-layer ANNs exhibit consistently higher fidelity than those from regression models and shallower ANNs. Inclusion of the autoregressive term and increasing the number of hidden layers in ANNs from 1 to 5 also improve the model performance for gust magnitudes (lower RMSE, increased correlation, and model standard deviations that more closely approximate observed values). Deeper ANNs (e.g., 20 hidden layers) exhibit higher skill in forecasting strong (17–25.7 ms-1) and damaging (25.7 ms-1) wind gusts. However, such deep networks exhibit evidence of overfitting and still substantially underestimate (by 50%) the frequency of strong and damaging wind gusts at the three airports considered herein.

Point of Contact
Sara C Pryor
Cornell University
Funding Program Area(s)