Model‐based analysis of the influence of catchment properties on hydrologic partitioning across five mountain headwater subcatchments

Abstract

Ungauged headwater basins are an abundant part of the river network, but dominant influences on headwater hydrologic response remain difficult to predict. To address this gap, we investigated the ability of a physically based watershed model (the Distributed Hydrology‐Soil‐Vegetation Model) to represent controls on metrics of hydrologic partitioning across five adjacent headwater subcatchments. The five study subcatchments, located in Tenderfoot Creek Experimental Forest in central Montana, have similar climate but variable topography and vegetation distribution. This facilitated a comparative hydrology approach to interpret how parameters that influence partitioning, detected via global sensitivity analysis, differ across catchments. Model parameters were constrained a priori using existing regional information and expert knowledge. Influential parameters were compared to perceptions of catchment functioning and its variability across subcatchments. Despite between‐catchment differences in topography and vegetation, hydrologic partitioning across all metrics and all subcatchments was sensitive to a similar subset of snow, vegetation, and soil parameters. Results also highlighted one subcatchment with low certainty in parameter sensitivity, indicating that the model poorly represented some complexities in this subcatchment likely because an important process is missing or poorly characterized in the mechanistic model. For use in other basins, this method can assess parameter sensitivities as a function of the specific ungauged system to which it is applied. Overall, this approach can be employed to identify dominant modeled controls on catchment response and their agreement with system understanding.

Exploring the impact of forcing error characteristics on physically based snow simulations within a global sensitivity analysis framework

“…There are fewer detailed studies focusing on forcing uncertainty relative to the number of parametric and structural uncertainty studies (Bastola et al, 2011;Benke et al, 2008;Beven and Binley, 1992;Butts et al, 2004;Clark et al, 2008Clark et al, , 2011bClark et al, , 2015aEssery et al, 2013;Georgakakos et al, 2004;Jackson et al, 2003;Kelleher et al, 2015;Kuczera and Parent, 1998;Liu and Gupta, 2007;Refsgaard et al, 2006;Slater et al, 2001;Smith et al, 2008;Vrugt et al, 2003aVrugt et al, , b, 2005Yilmaz et al, 2008). Di Baldassarre and Montanari (2009) suggest that forcing uncertainty has attracted less attention because it is "often considered negligible" relative to parametric and structural uncertainties.…”

Characterizing and reducing equifinality by constraining a distributed catchment model with regional signatures, local observations, and process understanding

“…The entire Tenderfoot Creek catchment was modeled using the distributed hydrology-soil-vegetation model (Wigmosta et al, 1994(Wigmosta et al, , 2002 The model framework and forcing data used to simulate Tenderfoot Creek were previously employed and described in Kelleher et al (2015). Key details for spatial and meteorological forcing data are described below.…”

“…8), and ranges for all 53 parameters included in the analysis. Sources for ranges are detailed in Kelleher et al (2015). tandem for cells with both canopy and tall canopy vegetation, with the exception of vegetation height and overstory fractional coverage (aerial percentage of each cell covered by canopy vegetation) which we expected to differ between these two vegetation classes.…”

“…In particular, a large number of model parameters, corresponding to a large number of degrees of freedom, may lead to equifinality Spear, 1980, 1981;Spear and Hornberger, 1980;Beven and Binley, 2014), a well-documented problem with respect to all models (Beven, 1989(Beven, , 1993(Beven, , 2001). Many studies have documented the presence of parameter equifinality in terms of parameter sensitivity, concluding that the potential for equifinality in distributed model applications is high in both time (Franks et al, 1998;van Werkhoven et al, 2008;Zhang et al, 2013;Kelleher et al, 2015;Ghasemizade et al, 2017;Guse et al, 2016) and space (Wagener et al, 2009;Herman et al, 2013;Moreau et al, 2013).…”

The spatial and temporal evolution of contributing areas

“…Fully distributed watershed models capable of tracking runoff source areas-or contributing areas-are often data intensive and include numerous parameters (e.g., DHSVM [Wigmosta et al, 1994], RHESSys [Band, 1993], and PIHM [Qu and Duffy, 2007]), which could increase the degree of equifinality [Beven, 2006] and may lead to various representations of internal system behavior [Kelleher et al, 2015]. Here we present a parsimonious but fully distributed modeling framework that incorporates topographically driven lateral water redistribution and eddy covariance derived spatially disaggregated evapotranspiration measurements to simulate streamflow and the spatial distribution of water stored in the watershed through time.…”