Supplemental information to: Model simulations of arctic biogeochemistry and permafrost extent are highly sensitive to the implemented snow scheme

Prior to the evaluation of the large-scale performance of the new, Dynamic snow scheme, we conducted a single-site comparison to examine the validity of the results. These detailed snow pack observations from Zackenberg helped to determine whether the Dynamic scheme can simulate internal snow pack dynamics, snow depth and snow density. We established the ability of the new snow scheme to simulate snow conditions by comparing a simulated snow pack with snow depth and density observations from Zackenberg (2013-2014 snow season). Figure S2 presents the observed and simulated snow pack by the Dynamic and Static schemes. This figure shows that the Dynamic scheme simulates higher snow depth due to lower density values in the mid and top layers of the snow pack. Density values are compared qualitatively, since it is difficult to accurately align the observational and modelled layer densities.

Thermal properties in snow layers are derived from density, and this entity is especially important in the Dynamic snow scheme. The comparison to observations shows that the modelled density compares well to observations. There are lower densities early in the snow season and fresh snow has a low density, while density increases in late spring during the melt season. The Static scheme with constant snow density simulated a somewhat higher than observed snow depth. The difference in snow depth between the Static and Dynamic simulations is small -as indicated in Fig. S2, bottom panel. Figure S2: Snow pack dynamics at the Zackenberg GeoBasis station. Density values for the layers are extrapolated -from three and five layers for the observational and modelled data, respectively. The colours of the snow pack indicate snow density.
Overall, the new scheme reproduces the snow dynamics over the cold season better than the Static scheme. Please note that the model's climate forcing is a crucial controlling factor when simulating snow conditions. For this study, we used a global climate forcing dataset, which may explain some of the observed model-observation differences. Taken together, these results suggest that the Dynamic scheme is skilled in simulating the snow pack's internal structure and dynamics. Since the Static scheme has a constant snow density throughout the snow season, the Dynamic scheme is expected to better capture the seasonal behaviour of snow and soil conditions. The Zackenberg site comparison indicated that the Dynamic scheme successfully integrated these key processes affecting the density over the snow season. The mismatch between snow observations and simulations is influenced by the use of a global model forcing dataset instead of site-specific temperature, precipitation or snowfall series.

S2.2 Site simulation details
Years of observational data used for the site simulations on Abisko, Bayelva, Kytalyk, Samoylov and Zackenberg sites (PAGE21 sites) can be seen in Table S1. The computed RMSE between observed and modelled near surface soil temperature and air-soil temperature difference is shown in Table S2.  1986-2020 1998-2009 2011-2013 1996-2013 1996-2011 soil T 2012-2015 1998-2017 2004-2011 2012-2014 1995-2017 climatic zone sub-arctic high arctic low arctic low-arctic high arctic Table S2: RMSE for soil temperature and ∆ T for the applied snow schemes, and temperature regimes at the Russian sites (Sect. 3.2).      Figure S7: Simulated heterotrophic respiration normalised by soil carbon content, using the Static and Dynamic snow schemes and their difference for winter and summer. Figure S8: Simulated NPP using the Static and Dynamic snow schemes and their difference for winter and summer. Figure S9: Simulated NEE using the Static and Dynamic snow schemes and their difference for winter and summer.

S3.4 Nitrogen cycling
Besides the carbon-related fluxes, we also assessed the impact of snow on nitrogen cycling. Figure S10 shows the nitrogen mineralisation ( Fig. S10 (a)) and leaching ( Fig. S10 (b)) normalised by soil carbon content. Nitrogen mineralisation only changed markedly during the summer season within the permafrost region. Leaching is higher for the Dynamic scheme in Eastern-Canada and Northern-Russia. Nitrogen use efficiency (NUE) on panel (c) was calculated as the ratio between NPP and nitrogen uptake. The Dynamic scheme simulates a lower NUE than the Static scheme, which indicates a higher N uptake per unit productivity.   Figure S11: Mean fractional water content of the upper soil column in April, May, June and July, using the Static, Dynamic schemes and their difference.

S3.5 Vegetation dynamics
Sites where PFT dominance changed between the Static and Dynamic simulations is shown in Fig. ??. These transition sites are scattered across the Arctic, but there are some clear hotspots in Eastern-Russia, the Scandinavian coastline and Northern-America.