Abstract

The estimation of hub-height wind speed is critical to a comprehensive wind resource assessment, particularly for the evaluation of future energy mix scenarios. However, gridded datasets of wind speeds are often limited to near-surface winds, especially when it comes to climate model projections, which is a real limitation for using climate models. This study develops a transfer method to calculate 100 m wind speed using three machine learning methods, including the Least Absolute Shrinkage Selector Operator, Random Forest (RF) and extreme Gradient Boost (XGBoost). Compared with the traditional algorithm, based on empirical formulae, the tested machine learning-based algorithms allow much more accurate estimates of 100 m wind speeds. RF and XGBoost have good performance on the hourly scale, and correct the major biases of the classical, simplified algorithms, especially in the diurnal cycle of hub-height wind speeds. RF appears to be the best algorithm when compared with the reanalysis data. In addition, the machine learning transfer model is applied to 19 regional climate projections. Results show that the 100 m wind speed has decreased in most of Europe during 1979–2019, which is consistent with the observed stilling of surface winds in recent years. This trend is projected to increase in the future, under an uncurbed greenhouse gas emission scenario, which indicates adverse effects for the development of wind power generation in Europe. The approach established in this study can be applied to obtain numerical climate model outputs accurately, which is critical to the estimation of the long-term changes of global renewable energy resources.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call