Highly resolved modeling of extreme wind speed in North America and Europe

Abstract

High wind speed (U) is one of the most dangerous natural hazards in North America and Europe. As a result, spatially explicit, statistical estimation of extreme U is of particular relevance for many sectors. However, the most common sources of wind speed data such as reanalysis data and in situ measurements are limited for this purpose due to their coarse spatial resolution and low representativeness. Thus, the main goal was to develop a high spatial resolution (250 m × 250 m) model (GloWiSMo‐X) for monthly mapping of the maximum hourly U for a 10‐year return period (U10yr) in North America and Europe. The multistep development of GloWiSMo‐X is based on 2544 hourly U time series available from the integrated surface global hourly meteorological data set (UNCEI), U time series from ERA5 (UERA5), and mean wind speed from the Global Wind Speed Model (Utrue¯GloWiSMo). Firstly, the block maxima method was applied to estimate monthly wind speed for a 10‐year return period for both UNCEI (U10yr,NCEI) and UERA5 (U10yr,ERA5). Secondly, the least squares boosting approach was used to predict the target variable U10yr,NCEI yielding the predictions Utruê10yr. The predictor variables U10yr,ERA5, Utrue¯GloWiSMo, continent, and month were used as input. It was found that the highest monthly continental means of Utruê10yr (Utrue¯10yr) in January are 16.4 m/s in North America and 16.3 m/s in Europe. Utrue¯10yr dropped to 13.4 m/s and 12.5 m/s in August. The annual cycle of Utrue¯10yr is more pronounced in Europe than in North America. The central parts of the USA and Western Europe were identified as intracontinental regions with the highest Utrue¯10yr. GloWiSMo‐X proves to be very broadly applicable as it covers two different continents and all months. The model validation by the mean squared error (MSE) demonstrates its improved predictive power compared to ERA5.

Downscaling ERA5 wind speed data: a machine learning approach considering topographic influences

“…For instance, Monahan [18] downscaled monthly wind speed time series at the buoy locations by using multiple linear regression. Jung et al [19] employed a least square boosting approach to downscale the monthly extreme wind speed value for North America and Europe. Winstral et al [20] developed an optimization scheme to downscale the wind speed time series in Switzerland taking the local terrain structure into consideration.…”