4:45 PM - 5:00 PM
[AHW29-06] Performance evaluation of grid-based regionalization in heterogeneous catchments: a case study of a distributed hydrological model by using geospatial datasets in Japan
Keywords:Regionalization, Geospatial datasets, Heterogeneous catchments, Distributed hydrological model
Parameter regionalization, which aims to transfer model parameters from donor basins (gauged basins) to receptor basins (target basins) has been extensively studied since the launch of Predictions in Ungauged Basins (PUB). Geospatial datasets (e.g., land-use, soil types, geological conditions), which can present the spatial variability within catchments, are recommended for improving the modeling performance. However, conventional regionalization methods, which typically transfer and apply a single parameter set at the basin scale, implicitly assume catchments are homogeneous, even employing distributed hydrological models, leading to unsatisfactory results.
In this study, we investigated a new type of regionalization pattern, called grid-based regionalization, which aims to identify a general parameter set for each geospatial type from highly homogenous donor basins and assign these parameters to heterogeneous catchments accordingly, using a distributed hydrological model, 1K-DHM. Seven representative model parameter sets of 1K-DHM including three land-use and four soil types were identified through calibration in 53 donor catchments, and a leave-one-out cross-verification within each parameter group, resulting in a parameter map at a 30-second grid resolution across Japan. The identified parameter (IP) sets yielded noticeable differences in the depth-discharge relationship in 1K-DHM, explaining variations in runoff characteristics among geospatial categories.
The performance of IP to heterogeneous catchments was validated against individual optimized parameter sets (OP) for 70 receptor basins using Nash-Sutliffe Efficiency (NSE) and Peak Discharge Error (PDE). Generally, the results show that IP performed comparably to OP among 823 rainfall-runoff cases, yielding a median NSE of 0.73 (IP) against 0.76 (OP) and a median PDE of 0.028 (IP) against 0.014 (OP). Furthermore, the modeling performance of IP and OP was analyzed from the perspective of catchment descriptors (CDs) (e.g., spatial ratio, physiography). These analyses indicate that both IP and OP exhibit similar trends in relation to most CDs, suggesting comparable reproducibility under different physical conditions. However, IP demonstrated a significant improvement when applied to andosol-dominant catchments compared to OP. These results demonstrate that the grid-based regionalization has high potential for addressing predictions in ungauged basins while identified parameters exhibit strong robustness in the andosol spatial type.
In this study, we investigated a new type of regionalization pattern, called grid-based regionalization, which aims to identify a general parameter set for each geospatial type from highly homogenous donor basins and assign these parameters to heterogeneous catchments accordingly, using a distributed hydrological model, 1K-DHM. Seven representative model parameter sets of 1K-DHM including three land-use and four soil types were identified through calibration in 53 donor catchments, and a leave-one-out cross-verification within each parameter group, resulting in a parameter map at a 30-second grid resolution across Japan. The identified parameter (IP) sets yielded noticeable differences in the depth-discharge relationship in 1K-DHM, explaining variations in runoff characteristics among geospatial categories.
The performance of IP to heterogeneous catchments was validated against individual optimized parameter sets (OP) for 70 receptor basins using Nash-Sutliffe Efficiency (NSE) and Peak Discharge Error (PDE). Generally, the results show that IP performed comparably to OP among 823 rainfall-runoff cases, yielding a median NSE of 0.73 (IP) against 0.76 (OP) and a median PDE of 0.028 (IP) against 0.014 (OP). Furthermore, the modeling performance of IP and OP was analyzed from the perspective of catchment descriptors (CDs) (e.g., spatial ratio, physiography). These analyses indicate that both IP and OP exhibit similar trends in relation to most CDs, suggesting comparable reproducibility under different physical conditions. However, IP demonstrated a significant improvement when applied to andosol-dominant catchments compared to OP. These results demonstrate that the grid-based regionalization has high potential for addressing predictions in ungauged basins while identified parameters exhibit strong robustness in the andosol spatial type.