Sensitivity of a large-scale hydrologic model to quality of input data obtained at different scales; distributed versus stochastic non-distributed modelling

Abstract

The amount of information available to run a spatially distributed model is often very much less than the ideal. The aim of this study is to estimate the impact of degradation of information about the spatial distribution of input parameters, and second the scale at which this information is obtained. A well defined agricultural catchment, with an important database concerning spatial and temporal observations has been used for this purpose: the agricultural catchment of la Côte St André, 60 km North East of Grenoble in the South East of France. The methodological framework is based on the Areal Non-point Source Watershed Environmental Response Simulation model. A 3 year simulation using georeferenced variables (crops and soil types) and annual changes in crop rotation is first developed as a reference. This is compared to simulations results obtained during the same period, with the same climatic data, but with the following degradation of quality of other inputs: firstly, the spatial distribution of soils and crop is ignored; both variables being defined by their areal coverage obtained from local information; secondly, the same inputs are deduced from a database obtained at the European scale. In both cases, a Latin Hypercube Sampler is used to stochastically generate sets of samples corresponding to the probability distribution of variables. The study is based on comparisons between modelled outputs: drainage of water and leaching of nitrate below the root zone of crops at the catchment scale. When information is local, and in absence of lateral flow (runoff), distributed modelling and purely stochastic modelling provide identical catchment average values; on the contrary, the use of the European database may introduce important biases concerning the proportion of land uses and of soils. In both cases, however, the lack of information concerning the location of sensitive areas in terms of risks of pollution may be considered as an important weakness of stochastic models. This work was done in the frame of ‘CAMSCALE—Upscaling predictive models and catchment water quality’ a European Union DGXII—Environment funded programme co-ordinated by SSLRC, Silsoe, UK.