Adaptive surface module

The University of Bonn will develop the surface module for the MiKlip seamless decadal prediction system. To be able to exploit the memory effects of the subsurface soil moisture a complete 3-dimensional hydrological and a dynamical vegetation model will be used. It would be computationally inhibitive to run such a model for the full model domain. Therefore, the surface module will use the so-called adaptive parameterisation approach. In an adaptive parameterisation scheme, the computation is split into a more complex, intrinsic parameterisation and a simple, adaptive extrinsic parameterisation. The adaptive surface module calls the intrinsic parameterisation only in a few catchments. Because the intrinsic parameterisation is not run for the full model domain, its total computational cost goes down. To generalise the results to the full domain, a simpler extrinsic scheme is run for the full domain. The difference between the simple and the complex domain can be computed in the catchments in which both schemes are run. This difference will be interpolated in an intelligent way.

Examples

A simple vegetation model for NWP models may have a fixed annual cycle of the leaf area index (LAI) for every land use type. This may often be a good parameterisation, but for example in the summer of 2003 it would have failed in Europe. In this hot summer many smaller plants or at least part of their leaves died, which a dynamical vegetation model would notice. Exploiting this difference between the simple and the dynamical vegetation model in a few model catchments will likely be sufficient to correct most of this problem; as drought is a large scale phenomenon, the difference can probably be generalized well to the full model domain by interpolating a multiplication factor. Similar correction are likely possible for differences in the water table depth, etc.