Flood forecasting in data-scarce regions presents significant challenges due to limited historical records and sparse hydro-climatological observation networks. This issue is further compounded in developing countries such as Indonesia, where global-scale datasets often lack sufficient accuracy. Despite these limitations, effective flood management remains critical, as evidenced by frequent destructive flood events and the high population density in flood-prone areas. This study investigates the potential of integrating limited ground observations with global forecast data to enhance local flood prediction using the RRI Model. Specifically, it evaluates the performance of ECMWF rainfall forecasts in reproducing river discharge by comparing them with gauge-based observations. Hydrograph analyses for 3, 5, 7, and 10 day lead times indicate that shorter lead times yield closer alignment with observations and lower RMSE values, while longer lead times result in greater discharge variability and overestimation. Confusion matrix analysis further confirms that shorter lead times reduce false classifications and improve overall accuracy. Moreover, probabilistic forecasts provide insights into the range of possible future discharge scenarios. Based on the Brier Score assessment, the ensemble forecasts tend to overpredict Level 3 under normal conditions. However, the peak prediction rate indicates that approximately half of the ensemble members successfully predicted Level 3 floods. A case study of the Solo River using deterministic and ensemble forecasts suggests the effectiveness of flood forecasting 3-days in advance in the Cepu station.
Wijayanti et al. (2026) studied this question.