Analysis of uncertainties associated with different methods to determine soil hydraulic properties and their propagation in the distributed hydrological MIKE SHE model

Complex hydrological models require a significant amount of data as input. The necessary measurement campaigns to determine input variables and parameters can be extremely expensive and time consuming, particularly at catchment scale. A subset of the inputs of hydrological models is the set of soil hydraulic parameters. Pedo-transfer-functions (PTFs), relating easily measurable soil properties to soil hydraulic parameters, can deliver candidate approximations for the required soil hydraulic properties. In the present study, uncertainties, resulting from four ways to obtain soil hydraulic parameters, are compared and evaluated with respect to their resulting uncertainties on different model outputs. These four methods are: (i) moisture retention lab measurements, (ii) prediction via PTFs using field texture measurements, (iii) prediction via PTFs using USDA texture classes, and (iv) prediction through the bootstrap-neural network approach using field texture measurements. The effect of parameter uncertainties on simulated catchment response was investigated using the spatially distributed, physically based hydrological MIKE SHE model in a joint deterministic-stochastic approach, based on the Latin Hypercube Sampling. As expected, different results are found for the different model outputs: discharge, ground water level, and soil water content. Including the PTF model as well as measurement fitting error, next to soil heterogeneity, when quantifying the input distributions, has a major impact, which cannot be neglected. Scaling issues were disregarded and parameters presumed to be grid-effective. The assumption of equal medians of the soil hydraulic functions, providing the input for the MIKE SHE model, generally cannot be rejected, but the uncertainties differed. The neural network approach consistently provides the smallest uncertainty, but exhibits different median values as well as uncertainty, and as such its application requires further research. No significant conclusions can be inferred for the ground water elevations--the model behaved differently for the separate methods, indicating even non-behavioural parameter sets. Soil water content and cumulative discharge uncertainty followed the iv, i, ii, iii order. K(sat) (ground water recharge, runoff-infiltration ratio) and theta(s) (soil water contents) can be established as influential parameters. Methods ii and i provide for similar in- and output, however their input distributions do not necessarily correspond to grid-effective values. Depending on the objective of the model application, approximation methods to assess soil hydraulic parameters can be a valid option.