Unlocking watershed mysteries: Innovative regionalization of hydrological model parameters in data-scarce regions

Study region
The Upper Blue Nile Basin in Ethiopia, characterized by its complex hydrological system, is the focus of this study. The basin includes 76 gauged watersheds, which were analyzed to estimate parameters for ungauged locations using regionalization techniques.

Study focus
Regression-Based Approach (RBA), Physical Similarity Approach (PSA), and Spatial Proximity Approach (SPA), for estimating GR4J model parameters. A 25 km by 25 km fishnet-based grid was implemented to enable parameter prediction for ungauged watersheds. Principal Component Analysis (PCA) and k-means clustering were used to group gauged watersheds into three homogeneous clusters, with Beressa, Dedessa, and Gilgel Abay selected as pseudo-watersheds for validation. Model performance was evaluated using PBIAS, R², and NSE metrics.

New hydrological insights for the region
The RBA outperformed PSA and SPA in parameter transfer, achieving R² values of 0.57, 0.79, and 0.67; PBIAS values of 7.3, −1.5, and 2.6; and NSE values of 0.58, 0.78, and 0.67 for Beressa, Dedessa, and Gilgel Abay, respectively. Incorporating grid-based parameter values further improved model performance, with NSE values of 0.81 for Dedessa, 0.63 for Beressa, and 0.61 for Gilgel Abay. These findings highlight the effectiveness of the grid-based regionalization approach for accurate streamflow prediction in ungauged watersheds within the Upper Blue Nile Basin.