Improving a subgrid runoff parameterization scheme for climate models by the use of high resolution data derived from satellite observations

In this study it is shown that the availability of a very high resolution dataset of land surface characteristics leads to the improvement of a surface runoff parameterization scheme. The improved parameterization scheme was developed for application in global and regional climate models and is a further development of the Arno scheme that is widely used in climate models. Here, surface runoff is computed as infiltration excess from a "bucket" type reservoir which takes the subgrid variability of soil saturation within a model gridbox into account. Instead of prescribing a distribution of subgrid scale soil water capacities as in the original Arno scheme, the array of high resolution soil water capacities taken from a global 1 km dataset of land surface parameters is used to obtain individual fractional saturation curves for each model gridbox. From each saturation curve, the three parameters (a shape parameter describing the shape of the subgrid distribution of soil water capacities, subgrid minimum and maximum soil water capacity) required in the modified formulation of the scheme are derived via optimization. As in the original Arno scheme applied in the ECHAM general circulation model and the REMO regional climate model, topography variations will influence the distribution of saturated subgrid areas within a model gridbox. At most gridboxes the net effect of these changes is such that more runoff is produced for high soil water contents and less runoff for low soil water contents. A validation of simulated discharge computed with a simplified land surface scheme applied to reanalysis data of the European Centre for Medium-Range Weather Forecasts and a hydrological discharge model has shown that these changes lead to a more realistic simulation of the annual cycle of discharge for several catchments. In particular this could be shown for the Yangtze Kiang and Amur catchments where adequate input data are available.