Simulation of permafrost and seasonal thaw depth in the JULES land surface scheme

Abstract

Abstract. Land surface models (LSMs) need to be able to simulate realistically the dynamics of permafrost and frozen ground. In this paper we evaluate the performance of the LSM JULES (Joint UK Land Environment Simulator), the stand-alone version of the land surface scheme used in Hadley Centre climate models, in simulating the large-scale distribution of surface permafrost. In particular we look at how well the model is able to simulate the seasonal thaw depth or active layer thickness (ALT). We performed a number of experiments driven by observation-based climate datasets. Visually there is a very good agreement between areas with permafrost in JULES and known permafrost distribution in the Northern Hemisphere, and the model captures 97% of the area where the spatial coverage of the permafrost is at least 50%. However, the model overestimates the total extent as it also simulates permafrost where it occurs sporadically or only in isolated patches. Consistent with this we find a cold bias in the simulated soil temperatures, especially in winter. However, when compared with observations on end-of-season thaw depth from around the Arctic, the ALT in JULES is generally too deep. Additional runs at three sites in Alaska demonstrate how uncertainties in the precipitation input affect the simulation of soil temperatures by affecting the thickness of the snowpack and therefore the thermal insulation in winter. In addition, changes in soil moisture content influence the thermodynamics of soil layers close to freezing. We also present results from three experiments in which the standard model setup was modified to improve physical realism of the simulations in permafrost regions. Extending the soil column to a depth of 60 m and adjusting the soil parameters for organic content had relatively little effect on the simulation of permafrost and ALT. A higher vertical resolution improves the simulation of ALT, although a considerable bias still remains. Future model development in JULES should focus on a dynamic coupling of soil organic carbon content and soil thermal and hydraulic properties, as well as allowing for sub-grid variability in soil types.