Investigating the impact of reanalysis snow input on an observationally calibrated snow-on-sea-ice reconstruction

Abstract. A key uncertainty in reanalysis-based snow-on-sea-ice reconstructions is the choice of reanalysis product used for snowfall input. Although reanalysis products have many similarities in their precipitation output over the Arctic Ocean, they nevertheless have relative biases that impact derived snow-on-sea-ice estimates. In this study, snowfall from the ERA5, JRA-55 and MERRA-2 reanalysis products is used as input to the NASA Eulerian Snow On Sea Ice Model (NESOSIM). A Markov chain Monte Carlo (MCMC) approach is used to calibrate the wind packing and blowing snow parameters in NESOSIM run with these different snowfall inputs. A multi-input-averaged snow-on-sea-ice product is then constructed from NESOSIM run with the three reanalysis products. JRA-55 shows the largest departure from the previously-used values (Bayesian priors) when the MCMC calibration is run, and also has the largest posterior uncertainty due to parameter uncertainties. The MCMC calibration reconciles snow depths between NESOSIM run with different reanalysis snowfall inputs, but produces larger discrepancies in snow densities, due to the sensitivity of snow density in NESOSIM to parameter values and weak observational constraints on density. Regional climatologies and trends in the calibrated products are examined and compared to another reanalysis-based snow-on-sea-ice reconstruction, SnowModel-LG. NESOSIM and SnowModel-LG show close agreement in snow depth climatologies in the Central Arctic Ocean region, but differ more in peripheral seas. Trends are found to be region-dependent, and the magnitude of Central Arctic Ocean snow depth trends is more sensitive to the choice of reanalysis input than to the choice of model.