*3.2. Exploring the Impact of Date Selection on Wheat Yield Prediction Using VIs Derived from Sentinel-2*

In this study, the effect of adding different VIs derived from S2 corresponding to the three dates and its combination on the prediction of wheat grain yield was investigated using four different algorithms: CatBoost, SVM, RF, and MLR. All results (RMSE and R2) (Figure 4) were obtained from the testing dataset. A consistent pattern was observed for all dates, with the best results obtained using CatBoost and the worst using MLR. When using the data from a single day, the R2 and RMSE values varied greatly depending on the date. The worst results were always obtained when using VIs from Day 1. Thus, RMSE oscillated between 1.20 for CatBoost and 1.45 for MLR while R2 ranged between 0.45 and 0.33. In contrast, the best results for a single day were obtained with Day 2 and CatBoost, reducing the RMSE to 0.56 and increasing the R2 to 0.74.

When considering the predictive ability of the model using two different dates, the performance was better than when using each day separately. The R2 of CatBoost ranged between 0.81 for the Day 1–2 dataset and 0.82 for the Day 2–3 dataset (Figure 4), while the R2 value of MLR ranged between 0.65 for the Day 1–2 dataset and 0.69 for the Day 2–3 dataset. This result suggests that the best predictions were obtained with the dates corresponding to GS39-49 and GS69-75.

Nonetheless, the results indicate that all algorithms obtained the best results when they were trained with a dataset composed of the three dates (corresponding to GS30, GS39-49, and GS69-75 phenological stages). The R2 values ranged from 0.859 for the CatBoost algorithm to 0.77 for MLR, while RMSE ranged from 0.32 for CatBoost to 0.50 for MLR.

**Figure 4.** R<sup>2</sup> and RMSE of the four tested algorithms (MLR, Multiple Linear Model; RF, Random Forest; SVM, Support Vector Machine; CatBoost) when trained with VIs derived from S2 corresponding to different dates. It also shows accuracy metrics of the combination of different days.
