A New Scheme for Daily Peak Wind Gust Prediction Using Machine Learning

Abstract A major challenge in meteorology is the forecasting of winds owing to their highly chaotic nature. However, wind forecasts, and in particular daily peak wind gust forecasts, provide the public with a general sense of the risks associated with wind on a given day and are useful in decision making. Additionally, such knowledge is critical for wind energy production. Currently, no operational daily peak wind gust product exists. As such, this project will seek to develop a peak wind gust prediction scheme based on output from an operational numerical weather prediction model. Output from the North American Mesoscale (NAM) model will be used in a support vector regression (SVR) algorithm trained to predict daily peak wind gusts for ten cities commonly impacted by hazardous wind gusts (cities in the Midwest and central Plains) and with interests in wind energy. Output from a kernel principal component analysis of the fully three-dimensional atmosphere as characterized by the NAM forecasts will be used to predict peak wind gusts for each location at 36 hours lead time. Ultimately, this initial product will lead to the development of a more robust prediction scheme that could one day transition into an operational forecast model.

[1]  Soroosh Sorooshian,et al.  Calibration of Probabilistic Quantitative Precipitation Forecasts with an Artificial Neural Network , 2007 .

[2]  Adrian E. Raftery,et al.  Bias Correction and Bayesian Model Averaging for Ensemble Forecasts of Surface Wind Direction , 2010 .

[3]  Nello Cristianini,et al.  An Introduction to Support Vector Machines and Other Kernel-based Learning Methods , 2000 .

[4]  Adrian E. Raftery,et al.  Weather Forecasting with Ensemble Methods , 2005, Science.

[5]  Paul J. Roebber.  Seeking Consensus: A New Approach , 2010 .

[6]  Pierre Pinson,et al.  Adaptive calibration of (u,v)‐wind ensemble forecasts , 2012 .

[7]  William H. Klein,et al.  forecasting local weather by means of model output statistics , 1974 .

[8]  Pierre Pinson,et al.  Non‐parametric probabilistic forecasts of wind power: required properties and evaluation , 2007 .

[9]  Henrik Madsen,et al.  Using quantile regression to extend an existing wind power forecasting system with probabilistic forecasts , 2006 .

[10]  Andrew E. Mercer,et al.  Identification of severe weather outbreaks using kernel principal component analysis , 2011, Complex Adaptive Systems.

[11]  H. Redkey,et al.  A new approach. , 1967, Rehabilitation record.

[12]  Michael C. Morgan,et al.  A Two-Season Impact Study of the WindSat Surface Wind Retrievals in the NCEP Global Data Assimilation System , 2010 .

[13]  Andrew E. Mercer,et al.  Statistical Modeling of Downslope Windstorms in Boulder, Colorado , 2006 .