Satellite remote sensed data to improve the accuracy of statistical models for wind resource assessment - (PPT)

摘要

Long-term wind resource assessment (WRA) using mesoscale models is useful in order to identify locations for wind energy projects. In the last decades, (Geo)Statistical methods have been applied for WRA showing lower accuracy compared to numerical methods but, with the advantage of requiring less computing resources. Remote sensing provides environmental and climate data that can be used as covariates in geostatistical algorithms to increase the accuracy of predictions. In this study, we test the impact of remote sensing data on the accuracy in WRA in complex terrain using a geostatistical machine learning approach and we compare it with previous study. The wind resource in Switzerland is mapped at 1km resolution using observations of weather stations at 10m high collected over 5 years (2009-2014). A database of 8'600 climate and environmental covariates is used and 160 parameters are selected by the algorithm based on their spatial correlation with wind observations. The method predicts the spatial distribution of the Weibull parameters, wind speed and wind direction with the corresponding interval of confidence for each of the 18 sectors in which the wind rose is divided. A 5-folds validation process repeated 100 times results in a bias of 0.06m/s, a mean absolute error (MAE) of 0.46m/s and a root mean square error of 0.66m/s. An external validation is carried out using the measurements of 8 weather stations at heights between 22m and 162m not. The bias and the MAE result 0.64m/s and 0.8m/s including forested sites and 0.09m/s and 0.32m/s excluding forested sites.