The Greenland ice sheet: modelling the surface mass balance from GCM output with a new statistical downscaling technique

Abstract. The aim of this study is to derive a realistic estimation of the Surface Mass Balance (SMB) of the Greenland ice sheet (GrIS) through statistical downscaling of Global Coupled Model (GCM) outputs. To this end, climate simulations performed with the CNRM-CM5.1 Atmosphere-Ocean GCM within the CMIP5 (Coupled Model Intercomparison Project phase 5) framework are used for the period 1850–2300. From the year 2006, two different emission scenarios are considered (RCP4.5 and RCP8.5). Simulations of SMB performed with the detailed snowpack model Crocus driven by CNRM-CM5.1 surface atmospheric forcings serve as a reference. On the basis of these simulations, statistical relationships between total precipitation, snow-ratio, snowmelt, sublimation and near-surface air temperature are established. This leads to the formulation of SMB variation as a function of temperature variation. Based on this function, a downscaling technique is proposed in order to refine 150 km horizontal resolution SMB output from CNRM-CM5.1 to a 15 km resolution grid. This leads to a much better estimation of SMB along the GrIS margins, where steep topography gradients are not correctly represented at low-resolution. For the recent past (1989–2008), the integrated SMB over the GrIS is respectively 309 and 243 Gt yr^–1 for raw and downscaled CNRM-CM5.1. In comparison, the Crocus snowpack model forced with ERA-Interim yields a value of 245 Gt yr^–1. The major part of the remaining discrepancy between Crocus and downscaled CNRM-CM5.1 SMB is due to the different snow albedo representation. The difference between the raw and the downscaled SMB tends to increase with near-surface air temperature via an increase in snowmelt.