Rainfall distributional properties control hydrologic model parameter importance when modeling medium size semi-arid watersheds

Authors: Meles, Menberu; Goodrich, David - Dave; Unkrich, Carl; GUPTA, HOSHIN - University Of Arizona; BURNS, IAN - University Of Arizona; HIRPA, FEYERA - One Concern, Inc; RAZAVI, SAMAN - University Of Saskatchewan; GUERTIN, D - University Of Arizona

Submitted to: Journal of Hydrology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/4/2024
Publication Date: 1/10/2024
Citation: Meles, M.B., Goodrich, D.C., Unkrich, C.L., Gupta, H.V., Burns, I.S., Hirpa, F.A., Razavi, S., Guertin, D.P. 2024. Rainfall distributional properties control hydrologic model parameter importance when modeling medium size semi-arid watersheds. Journal of Hydrology. 51. https://doi.org/10.1016/j.ejrh.2024.101662.
DOI: https://doi.org/10.1016/j.ejrh.2024.101662

Interpretive Summary: Evaluation of watershed models, and their capabilities to understand their importance in engineering, water management, or other analysis, requires knowledge of the factors and conditions that strongly control the model responses. In watersheds larger than the typical arid and semi-arid storm sizes, we evaluated the impacts of different properties of rainfall events on the dynamics of model parameters and the outputs over the model simulation periods. A Global Sensitivity Analysis (GSA) tool called VARS (Variogram Analysis of Response Surfaces) was used to show changes in model behavior to the different model input scenarios. Ten selected rainfall and runoff events in the watershed with an area of 16.6 km2 and three hypothetical storm intensities (high, medium, and low) localized at five different locations in WS10 in Walnut Gulch Experimental Watershed were used to study KINEROS2 parameter importance. Analysis of impacts of various storms on parameter importance revealed that storm proprieties such as intensity, spatial distribution, and distance of storm center to watershed outlet affect the relation between model response and parameters and provide constraining conditions for setting global parameter sets within the parameter spaces. The time-variant parameter importance analysis of the K2 model for multiple storms at the specified location (zonal calibration) can help to develop a spatially distributed K2 parameterization scheme and identification of top important parameters using rainstorm information such as the location of the storm center, intensity, and area covered by the storms.

Technical Abstract: The dynamics of parameter importance in Earth Systems Models have been the focus of substantial research in recent years to understand the factors and conditions that strongly control the watershed responses. To investigate the changing aspects of parameter importance in semi-arid environments, we implemented the VARS (Variogram Analysis of Response Surfaces) methodology to characterize the predictive uncertainty of the KINEROS2 physically-based distributed hydrologic model using ten selected rainfall and runoff events and three hypothetical storm intensities (high, medium, and low) localized at five different locations across the WS10 watershed. Time-variant (throughout the simulation period) and time-aggregate (average of a simulation) variogram-based parameter importance analyses were performed to explain model parameter control factors. We explored and quantified how parameter control varies in space and time due to the variability in rainfall fields in a relatively larger watershed than the typical semi-arid storm areas. Our analysis revealed two crucial outcomes that have far-reaching impacts on model parameterization. These are 1) parameter importance varies considerably depending on rainfall properties (intensity, depth, temporal rainfall distribution, and location and distance of the storms to the watershed outlet), and 2) time-variant analysis of parameter importance with the knowledge of key rainfall properties can be used to identify the dominant parameters controlling model responses, which leads to improved calibration approach for the KINEROS2 model without the need to run a comprehensive sensitivity analysis. Parameter importance indices showed a power law function to the changes in rainfall distribution index. Moreover, parameter consistency in zonal scale simulation provided additional constraining conditions for identifying unique spatially distributed calibrated values by further narrowing the parameter spaces. However, further extensive analysis, including different watersheds and models, is required to understand parameters controlling model responses and storm properties and the impacts on broader equifinality issues.