Snow Trends The snow record can be divided into 3 periods: 1950-1978 model snow with empirical correction 1979-198x x is around 5. early snow analyses 198x-present newer snow analyses The CORe snow output is WEASD which is water equivalent of the snow depth (mass). The CORe forecast model takes the snow field and adds the snow fall and subtracts the snow melt and snow sublimation. The model will adjust the snow density as older snow is densier than new snow. When there is a new snow analysis (every 24 hours in the 1979+ period). The cycle program then examines both the snow analysis and the modeled snow from the previous cycle. If there is both analysed snow and modeled snow at the same grid point, then a snow analysis is modified with a mixture of the modeled and analyzed snow. The reasoning being that the snow analyses has poorer accuracy for deep snows (2+ m). For the period 1950-1978. the model snow with an empirical correction was used. Every 24 hours, an empirical correction was applied. This corrections minimized the error in the global snow cover. The early snow analyses were use old satellite technology which wasn't as good as today. The algorithms were more primative and the computer processing was more limited. The snow analyses were not reprocessed for modern reanalyses. For the satellite period (1979+), the details of the snow analysis is given in snow.txt -------------------------------------------- Modeled Snow Correction We ran CORe in 2016 with model snow and observed snow analysis. We found that the model snow had more snow cover which locally leads to cooler 2 meters temperatures in regions that have variable snow cover. Since many people live in regions that see a variable snow cover, this would affect their long term temperature trends. The snow correction was based on physical reasoning. For a shallow snow depth, the natural variablility of the terrain would mean that some fraction of the grid cell would be snow free. So the grid cell would behave radiatively as partially snow covered and partially snow free grid cell. There is no provision in the land surface model for a fractional snow cover. So even a minute amount of snow would behave radiatively as snow covered. So we set the snow depth to zero when the snow ammount was smaller than some empirically determined amount. The amount was found by minimizing the error of the global snow cover relative to observations. This correction was done every 24 hours to allow snow to accumulate from storms. It was observed that the mountain snows tended to last longer into the summer than observed. This suggests that land surface model needs some improvement or the land surface model was optimized for the higher resolutions of the NWP model than the low resolution CORe. This effect is clearly seen in WEASD time series in mountainous regions. Trends in Snow Cover vs Snow Depth The global snow cover anomaly is shown in ./core_snow_q.png This is the plot of the % land snow cover with the 1991-2020 climatology removed. The lower plot shows the % land snow cover. The period from 1950-1978 uses model snow. From 1979-2023 uses the CFSR snow which is based on the AFWA snow depth analyses and the IMS snow cover analyses. There is roughly a 3.5% greater snow coverage in the 1950-1978 period relative to the 1979-2023 period. The top plot shows a transition between 1978-1979. On the other hand, the land snow cover between 30N-60N (./core_snow_nm_q.png) doesn't show a strong transition but a gradual transition from 1985-1990. Plots from other latitude bands show a strong 1978-1979 transition. The global averaged snow depth is shown in ./g_snod_core.png. The assimilation streams are clearly seen. Since the snow cover didn't show the stream structure, it appears that the permanently snow covered grid points had increased snow depth. The model doesn't have the physics that would limit the snow depth.