*4.3. Error Statistics of DS\_Spline, DS\_OIMERG, and DS\_CIMERG*

Based on the improvement in the quality of the satellite precipitation product after OI, the downscaling effect of the GWR method was further validated. Figure 6 shows comparisons among the downscaled data obtained with three different downscaling methods for May to September of 2014–2018. DS\_CIMERG had the highest CC values in 22 of the 24 months, with lower values than those of DS\_Spline only in May and August 2014. On average, the mean CC of DS\_CIMERG was 10% higher than that of DS\_Spline. Between DS\_Spline and DS\_OIMERG, the CC values of DS\_Spline were higher in 8 months and lower in 16 months, suggesting that the quality of satellite precipitation data cannot be effectively improved by downscaling with only the spatial interpolation technique. The RMSE and MAE values of DS\_CIMERG were the lowest

in almost all months. Furthermore, the mean CC, RMSE, and MAE values respectively improved from 0.616, 26.56 mm, and 17.59 mm for CIMERG to 0.635, 25.93 mm, and 17.15 mm for DS\_CIMERG, respectively.

**Figure 6.** (**a**) The correlation coefficient (CC), (**b**) mean absolute error (MAE), and (**c**) root-mean-square error (RMSE) values of downscaled precipitation estimates obtained using GWR based on OIMERG (DS\_OIMERG), GWR based on CIMERG (DS\_CIMERG), and spline interpolation (DS\_Spline) for the Tianshan Mountains between May and September 2014–2018. Independent rain gauge observations were used as benchmarks.
