Abstract:
Variability with respect to model input data is recognised as a potential source of uncertainty in model predictions. The aim of this study is to estimate the sensitivity of computed values of water balance terms, in particular drainage below the root zone of crops, and that of nitrate leaching, to variability of soil transport parameters within a soil class, and to quantify the domains of sensitivity as a function of soil type. The methodological framework is based on the concept of Areal Non-Point Source Watershed Environmental Response Simulation, coupled with a Latin Hypercube Sampler, to obtain a stochastic model. Two applications are considered. First, a case study is made of the experimental catchment of LaCote St Andre, predominantly a loam soil, where intensive experimentation has been carried out from 1991 to 1995. Second, a generalisation to different types of soil is carried out.It is shown that for this model within-class variability has no effect in long-term simulations for soils with saturated hydraulic conductivity Ks higher than 100mm/day. For these soils, the concept of representative elementary area is fully acceptable and convenient. Corresponding soil classes can each be described by a single set of parameters (the barycentre (centrod) of the class) with a very small loss of information compared to a very important gain in terms of input data requirements and simulation time. This has important consequences for large-scale distributed models, since it reduces considerably the number of measurements necessary to describe the soil; in particular there may be no need, in this range, to account for spatial variability of textural parameters within a class.In contrast, within-class variability of transport parameters becomes an important source of uncertainty for soil classes below this threshold value of saturated hydraulic conductivity. An estimation of errors resulting from aggregation of transport parameters values to those corresponding to the centrod of the soil class is given. These errors are obviously dependent on Ks values and rainfall intensity.
