Current CMIP6 climate models (such as CESM2 and NESM3) use constant snow density, while those models that focus on snow depth and density changes (such as SnowModel-LG) use empirical snow density formulas. Comparing the modeled snow depth with those observed by the CryoSat-2 satellite, it is found that from the perspective of the spatial distribution and average value of the snow depth, it is difficult to detect the effects of varying snow density on the simulation of snow depth in the Arctic Ocean. The model improvement and its mechanism from varying snow depth is still to be further studied. Here an empirical snow density model considering meteorological factors such as air temperature, wind etc., is applied to the SNOTEL observational site to carry out the following sensitivity experiments for different factors: A. snow density model considering all meteorological factors; B. constant snow density model; C. same as A but the influence of wind on the densification is not considered and D. same as A but the influence of temperature on the densification is not considered. The root mean square error of snow depth simulated by experiments A, B, C and D from November 1, 2018 to May 10, 2019 are 4.2 cm, 4.8 cm, 25.9 cm, and 4.2 cm, respectively. The results show that the mean snow density and depth simulated by the varying snow density model are close to the results using constant snow density, but the root mean square error of the simulated snow depth from Case A is the smallest, and the Case A simulation can reproduce the high frequency variations of snow depth on the time scale of several days to ten days. In the meantime, the relative errors in the period with high-frequency snow depth variations are also reduced as they are found to be related. In addition, it is also found that the influence of temperature on snow densification is much smaller than that of wind.