Increasing Parameter Identifiability Through Clustered Time-Varying Sensitivity Analysis
Lu Wang,Yue-Ping Xu,Jiliang Xu,Haiting Gu,Zhixu Bai,Peng Zhou,Hongjie Yu,Yuxue Guo
DOI: https://doi.org/10.1016/j.envsoft.2024.106189
IF: 5.471
2024-01-01
Environmental Modelling & Software
Abstract:Hydrological models are becoming progressively complex, leading to unclear internal model behavior, increasing uncertainty, and the risk of equifinality. Accordingly, our study provided a research framework based on global sensitivity analysis, aiming at unraveling the process-level behavior of high-complexity models, teasing out the main information, and ultimately exploiting its usage for model parameterization. The Distributed HydrologySoil-Vegetation Model implemented in a mountainous watershed was used. Results indicated that 5 soil parameters and 5 vegetation parameters were most important to control the streamflow responses, while their importance varied greatly throughout the simulation period. Four typical patterns of parameter importance corresponding to different watershed conditions (i.e., flood, short dry-to-wet, fast recession, and continuous dry periods) were successfully distinguished. Using this clustered information, parameters with short dominance times were more identifiable over the clusters (time periods) in which they were most important. The reduced posterior parameter space also slightly improved the model performance.