[en] A number of high resolution reconstructions of the surface mass balance (SMB) of the Greenland ice sheet (GrIS) have been produced using global re-analyses data extending back to 1958. These reconstructions have been used in a variety of applications but little is known about their consistency with each other and the impact of the downscaling method on the result. Here, we compare four reconstructions for the period 1960–2008 to assess the consistency in regional, seasonal and integrated SMB components. Total SMB estimates for the GrIS are in agreement within 34% of the four model average when a common ice sheet mask is used. When models' native land/ice/sea masks are used this spread increases to 57%. Variation in the spread of components of SMB from their mean: runoff 42% (29% native masks), precipitation 20% (24% native masks), melt 38% (74% native masks), refreeze 83% (142% native masks) show, with the exception of refreeze, a similar level of agreement once a common mask is used. Previously noted differences in the models' estimates are partially explained by ice sheet mask differences. Regionally there is less agreement, suggesting spatially compensating errors improve the integrated estimates.
Modelled SMB estimates are compared with in situ observations from the accumulation and ablation areas. Agreement is higher in the accumulation area than the ablation area suggesting relatively high uncertainty in the estimation of ablation processes. Since the mid-1990s each model estimates a decreasing annual SMB. A similar period of decreasing SMB is also estimated for the period 1960–1972. The earlier decrease is due to reduced precipitation with runoff remaining unchanged, however, the recent decrease is associated with increased precipitation, now more than compensated for by increased melt driven runoff. Additionally, in three of the four models the equilibrium line altitude has risen since the mid-1990s, reducing the accumulation area at a rate of approximately 60 000 km2 per decade due to increased melting. Improving process representation requires further study but the use of a single accurate ice sheet mask is a logical way to reduce uncertainty among models.