Probabilistic wind speed forecasting on a grid based on ensemble model output statistics

TL;DR

This paper introduces a statistical framework for generating locally calibrated probabilistic wind speed forecasts on a grid, utilizing ensemble model output statistics and various distributional approaches, validated over Germany.

Contribution

It proposes three new post-processing methods using different distributions and evaluates interpolation schemes for grid-based probabilistic wind speed forecasting.

Findings

The methods improve local wind speed forecast calibration.

Interpolation schemes effectively produce grid-wide probabilistic forecasts.

Validation over Germany shows enhanced forecast accuracy.

Abstract

Probabilistic forecasts of wind speed are important for a wide range of applications, ranging from operational decision making in connection with wind power generation to storm warnings, ship routing and aviation. We present a statistical method that provides locally calibrated, probabilistic wind speed forecasts at any desired place within the forecast domain based on the output of a numerical weather prediction (NWP) model. Three approaches for wind speed post-processing are proposed, which use either truncated normal, gamma or truncated logistic distributions to make probabilistic predictions about future observations conditional on the forecasts of an ensemble prediction system (EPS). In order to provide probabilistic forecasts on a grid, predictive distributions that were calibrated with local wind speed observations need to be interpolated. We study several interpolation schemes that combine geostatistical methods with local information on annual mean wind speeds, and evaluate the proposed methodology with surface wind speed forecasts over Germany from the COSMO-DE (Consortium for Small-scale Modelling) ensemble prediction system.