*4.4. Overall Assessment of the Two-Step Merging and Downscaling Method*

Figure 7 presents scatter plots of OIMERG, CIMERG, and the three types of downscaled precipitation data against all of the observation data from the study area. After the first and second processing steps (i.e., OI and GWR), the CC was improved by 10% and 2%, respectively. As shown in Figure 7e, DS\_OIMERG showed slight improvements compared with OIMERG, while it was less accurate than CIMERG and DS\_CIMERG, indicating that direct downscaling based on original satellite precipitation products may have a limited effect in improving the quality of downscaled precipitation estimates.

**Figure 7.** Scatter diagrams of the (**a**) OIMERG, (**b**) CIMERG, (**c**) DS\_Spline, (**d**) DS\_CIMERG, and (**e**) DS\_OIMERG precipitation products against the observed precipitation during the warm seasons from 2014 to 2018. OIMERG, original IMERG precipitation data; CIMERG, OI-corrected IMERG precipitation data; DS\_Spline, obtained by using the spline interpolation method for OIMERG precipitation downscaling; DS\_CIMERG, obtained via GWR downscaling with CIMERG; DS\_OIMERG, obtained via GWR downscaling with OIMERG; OBS, observed precipitation data.

Figure 8 shows boxplots of the evaluation metrics for the five precipitation products. A boxplot divides a dataset into four segments based on the maximum, minimum, median, and two quartiles of the data. The middle horizontal line represents the median, which divides the statistical data into two equal parts. As shown in Figure 8, the distributions of the CC, MAE, and RMSE were all uniform and consistent. Among the five datasets, the best performance was observed for DS\_CIMERG, whose CC values were more concentrated in the upper region, while the MAE and RMSE values were more concentrated in the lower region. In particular, the minimum and maximum MAE and RMSE were the lowest for DS\_CIMERG, followed by CIMERG. The metrics for OIMERG and DS\_Spline were basically the same, while those for DS\_OIMERG were in agreement with the overall assessment results and slightly better than those for OIMERG (Figure 7). In Figure 8, the small rectangular boxes represent the distributions of the metric values around their averages. DS\_CIMERG performed the best among the five products, with CC, MAE, and RMSE values of 0.70, 17.29 mm, and 22.45 mm, respectively. The overall evaluation results showed that the dataset generated using the OI-GWR method proposed

in this study was notably superior to the datasets of the other two downscaling products as well as the initial IMERG precipitation data.

**Figure 8.** Boxplots of the evaluation metrics (**a**, CC; **b**, MAE and **c**, RMSE) for the five precipitation datasets for the warm seasons of 2014–2018. OIMERG, original IMERG precipitation data; CIMERG, OI-corrected IMERG precipitation data; DS\_Spline, obtained by using the spline interpolation method for OIMERG precipitation downscaling; DS\_CIMERG, obtained via GWR downscaling with CIMERG; DS\_OIMERG, obtained via GWR downscaling with OIMERG.
