*3.6. Summary*

The preceding subsections presented and analyzed the retrieval performance of several machine learning methods in different wind speed intervals. This subsection summarizes and analyzes their performance gaps. Figure 11 shows the Rs and RMSEs of the models using machine learning methods. It is obvious from Figure 11 that the RMSE of LGBM is smaller than those of other models in a low wind speed interval, while the RMSE of ET is smaller than those of other models in a high wind speed interval. The R values are usually larger when the RMSE values are smaller. The performance of LGBM, ET, ANN and XGB are significantly better than that of SVM, BT and SLR, which means that they are more suitable for wind speed retrieval.

**Figure 11.** Rs and RMSEs of machine learning models.
