Revealing spatial patterns of lateral hydraulic conductivity through sensitivity analysis of wflow_sbm 

Moving toward high-resolution gridded hydrologic models asks for novel parametrization approaches. The use of transfer functions and advances in scaling and regionalization play an important role to ensure flux matching across scales. However, for some processes no transfer functions are yet available or simplified approaches, such as a fixed vertical-to-horizontal saturated hydraulic conductivity ratio, are being used. To get insight into the spatial variability of the vertical-to-horizontal saturated hydraulic conductivity ratio we performed a sensitivity analysis on one parameter of the wflow_sbm model across England, Wales and Scotland exploiting the CAMELS-GB dataset. The wflow_sbm models were setup using reproducible workflows based on HydroMT (https://deltares.github.io/hydromt/stable/) for each CAMELS-GB basin.  To investigate the sensitivity to rainfall forcing all derived wflow_sbm models were first run using a default ratio of 100 with both EOBS and CEH GEAR rainfall data. The sensitivity analysis was only based on the high quality CEH GEAR rainfall dataset. In the sensitivity analysis, the vertical-to-horizontal saturated hydraulic conductivity ratio was varied over a large range from 1 – 10,000 and results were assessed using the non-parameteric KGE (which focuses more on recession/baseflow performance). Even with a fixed uniform vertical-to-horizontal saturated hydraulic conductivity ratio results show a big impact of the precipitation forcing on the model results.  The uncertainty analysis shows that wflow_sbm model results have a high sensitivity to the vertical-to-horizontal saturated hydraulic conductivity ratio. For the optimal ratios, we obtain high KGE values (median=0.84). In addition, when plotting the optimal ratios across the GB clear patterns emerge that seem to coincide with geological features. The resulting optimized lateral saturated hydraulic conductivity values seem realistic when compared with literature values. When compared to Grid2Grid model results the wflow_sbm model shows similar performance for most stations. However, for parts in the south of the England where the geology consists of chalk, the performance of Wflow_sbm is poor, but this is likely caused by the used soil depth map when constructing the models which limits the soil depth often to 30-60cm while it is known that the chalk below the soil is also hydrologically active.