Assessing the performance of a physically based hydrological model using a proxy‐catchment approach in an agricultural environment

Abstract

Physically based models are useful frameworks for testing intervention strategies designed to reduce elevated sediment loads in agricultural catchments. Evaluating the success of these strategies depends on model accuracy, generally established by a calibration and evaluation process. In this contribution, the physically based SHETRAN model was assessed in two similar U.K. agricultural catchments. The model was calibrated on the Blackwater catchment (18 km 2 ) and evaluated in the adjacent Kit Brook catchment (22 km 2 ) using 4 years of 15 min discharge and suspended sediment flux data. Model sensitivity to changes in single and multiple combinations of parameters and sensitivity to changes in digital elevation model resolution were assessed. Model flow performance was reasonably accurate with a Nash-Sutcliffe efficiency coefficient of 0.78 in Blackwater and 0.60 in Kit Brook. In terms of event prediction, the mean of the absolute percentage of difference (μAbs diff ) between measured and simulated flow volume (Qv), peak discharge (Qp), sediment yield (Sy), and peak sediment flux (Sp) showed larger values in Kit Brook (48% [Qv], 66% [Qp], 298% [Sy], and 438% [Sp]) compared with the Blackwater catchment (30% [Qv], 41% [Qp], 106% [Sy], and 86% [Sp]). Results indicate that SHETRAN can produce reasonable flow prediction but performs less well in estimation of sediment flux, despite reasonably similar hydrosedimentary behaviour between catchments. The sensitivity index showed flow volume sensitive to saturated hydraulic conductivity and peak discharge to the Strickler coefficient; sediment yield was sensitive to the overland flow erodibility coefficient and peak sediment flux to raindrop/leaf soil erodibility coefficient. The multiparameter sensitivity analysis showed that different combinations of parameters produced similar model responses. Model sensitivity to grid resolution presented similar flow volumes for different digital elevation model resolutions, whereas event peak and duration (for both flow and sediment flux) were highly sensitive to changes in grid size. K E Y W O R D S catchment modelling, model sensitivity, proxy catchment, sediment yield, SHETRAN

Simulated event‐scale flow and sediment generation responses to agricultural land cover change in lowland UK catchments

“…Results showed similar model performance in both catchments (Blackwater R 2 = 0.8 [Qv], 0.6 [Qp]) 0.4 [Sy] 0.2 [Sp]) and Kit Brook R 2 = 0.8 [Qv], 0.6 [Qp]) 0.4 [Sy] 0.3 [Sp]); which corresponded to larger differences (Pct diff ) in Kit Brook (48% [Qv], 66% [Qp], 298% [Sy] and 438% [Sp]) than Blackwater catchment (30% [Qv], 41% [Qp], 106% [Sy] and 86% [Sp]). Kit Brook exhibited higher permeability than Blackwater based on measured flow duration curves (Escobar‐Ruiz et al, 2019), supporting the adjustment of the K sat for the 6th soil layer and resulting in considerable improvement in model performance for Kit Brook (i.e. 0.69 [NSE], 44% [Qv], 58% [Qp], 202% [Sy], 319% [Sp]) with no variations in the coefficient of determination.…”

“…For SHETRAN, this might be associated with model representation of physical processes (e.g. dimensional formulation, Bathurst, 2011), the choice of grid scale (Escobar‐Ruiz et al, 2019) and basis for estimating parameter values (Bathurst et al, 2004; Ewen & Parkin, 1996). Therefore, it is important to recognize this uncertainty when interpreting model results.…”

“…Therefore, the larger the sample size, the greater likelihood of detecting a significant difference (if present in the data). The 84 events over the 4‐year simulation were selected to obtain a systematic error in the simulated scenarios and used in the previous model calibration and evaluation (Escobar‐Ruiz et al, 2019).…”

“…The SHETRAN model has previously been tested on an event basis (Adams et al, 2005; Adams & Elliott, 2006; Bovolo & Bathurst, 2012; De & Bathurst, 2002; Elliott et al, 2012). Event‐scale performance using SHETRAN has been reported for the studied catchments based on analysis of 53 events in Blackwater and 46 in Kit Brook during the study period (Escobar‐Ruiz et al, 2019). Events were selected based on data quality (e.g.…”

“…The catchment is modelled as a set of grids with a limit in number of 300 Â 300, with each cell containing soil and vegetation information (Ewen et al, 2000). The reported parameters to which the water flow component of SHETRAN is most sensitive are actual evaporation to potential evaporation ratio (AE/PE), Strickler coefficient (Stk) and saturated hydraulic conductivity (K sat , [Bathurst, 1986;Anderton et al, 2002;de Hipt et al, 2017;Escobar-Ruiz et al, 2019]). In terms of the sediment transport components, the most sensitive parameters include raindrop soil erodibility coefficient (D r ), and overland flow erodibility coefficient (D f , [Adams & Elliott, 2006;de Hipt et al, 2017;Escobar-Ruiz et al, 2019]).…”