*5.2. Impact of the Gauge Density on OI-GWR*

To further validate the performance of OI-GWR and its applicability in sparsely gauged regions, we randomly selected 30% of the rain gauges for model training and used the remaining 70% of the rain gauges for model validation. For this analysis, the study period was extended to 2010-2018. Figure 15 shows that even when using only 30% of the stations for training, a positive correction effect could be obtained. However, since fewer rain gauges were used for training, the improvement was less significant than that achieved using more rain gauges. An overall evaluation showed that after the two steps of correction (i.e., OI and GWR), the CC increased from the initial value of 0.533 to values of 0.567 and 0.59, respectively; the RMSE decreased from the initial value of 27.08 mm to values of 26.84 and 26.55 mm, respectively; and the MAE decreased from the initial value of 18.3 mm to values of 17.76 and 17.63 mm, respectively. Therefore, the OI-GWR method proposed in this study can improve the deficiencies of the previous downscaling method. Since OI-GWR works well even when the size of the training dataset is reduced to 30%, the proposed method has great potential for application in sparsely gauged regions.

**Figure 15.** Scatter diagrams of OIMERG, CIMERG, and DS\_CIMERG against the observed precipitation during the warm seasons from 2010 to 2018.
