A framework to evaluate and identify development requirements for land-surface models at km-scale resolution: Application to a semi-arid and mountainous region

Earth-system and weather forecasting models are moving to km-scale resolutions to provide more pertinent information to society on extreme events or the impacts of climate change. As some parametrized processes can be represented explicitly, increasing spatial resolution is expected to be beneficial for the atmosphere and oceanic components. It is not obvious that the same benefits will be achieved for land-surface models (LSMs), as landscape organizing processes start to play a role. To evaluate the consequences of increasing resolution, six LSMs driven by 3-km resolution forcings are compared with their reference simulation at 50-km resolution. These high-resolution atmospheric forcing data are developed over a region covering all catchments flowing off the Pyrenees. It is shown that these forcings capture the contrasts in atmospheric conditions between mountainous areas and valleys absent at coarser resolutions. At finer resolution, the LSMs display reduced evaporation over semi-arid catchments, which cannot be explained by differences in the atmospheric forcings. The cause has to be sought in the lack of spatial redistribution of water within the catchments. The observed diurnal amplitude of land-surface temperature shows that the models do not reproduce the local minima along rivers and in irrigated areas caused by increased evaporation. We conclude that, at km-scale resolution, lateral transfers of water that organize landscapes play an important role in predicting evaporation correctly. At resolutions of a few deca-kilometres, the contribution of grid-cell lateral flows to evaporation can be neglected. However, at higher resolutions, groundwater, riparian recharge, and human water management for irrigation need to be simulated to represent realistic spatial contrasts in the surface fluxes that drive the atmosphere. We call upon the community to invest in the development of representation of these processes in LSMs, so that they are ready for higher resolution applications.