Sensitivity of the process-based model DNDC on microbiological parameters

Abstract

L'archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d'enseignement et de recherche français ou étrangers, des laboratoires publics ou privés. Overview Process-based model such as DNDC rely on a large numbers of parameters which were defined by the model developer on the basis of existing references. Subsequently, some values have been changed to improve model performance for specific applications, often without adequate documentation. Many of these parameters are thus estimates of the real values appropriate for local conditions introducing approximation errors for applications at larger scales. Spatially explicit datasets might be required for some parameters for which model output is highly sensitive. We will present a sensitivity analysis of 38 mainly micro-biological internal parameter of DNDC-EUROPE. Methods We test 38 mostly, currently internal, soil microbiological parameters required to simulate rates of nitrification, denitrification, and soil organic matter turnover on its sensitivity to estimated rates of N 2 O fluxes and nitrate leaching. The selected parameter include microbiological parameters describing population dynamics of denitrifying and nitrifying organisms as well as the characterization of soil carbon pools and their mineralization rates. The shape of the probability density functions was assumed to be triangular, when the range of a parameter could be obtained from literature data. In such cases, the default value was set as the mode and the maximum and minimum values of the rages were set as the extremes of the distribution. When no information on the range was available, we opted for a normal distribution with a CV of 20% around the nominal values. Input data for simulations of N 2 First, a list of sample points from the marginal pdfs of the k input parameters using a quasi-Monte Carlo generator (Sobol', 1967) is generated. It consists of blocks of with k+2 rows, where the first two rows are the coordinates of two independent realizations A and B and the subsequent k rows represent points having in turn all coordinates from A but one that is from B. In order to achieve convergence we generated 166 blocks resulting in 6640 simulations to be carried out at each simulation unit.