Land surface snow processes, which are challenging to simulate in Earth system models, are of critical importance in northern high-latitude regions, where they influence many other biogeophysical and biogeochemical processes. Capturing them in models is particularly difficult because of highly heterogeneous surface properties and lack of reliable data in remote and harsh regions. Here we present an offline land surface simulation using the Energy Exascale Earth System Model (E3SM) Land Model (ELM) over northern high-latitudes (>=60°N) at a half-degree spatial resolution. Results are evaluated using the ILAMB package and the NCAR Land Diagnosis Tool (as shown on https://elm-ngee-webserver.ornl.gov). For assessing snow processes and their consequences on plant phenology, we derived DOYs (day of year) of snow melt and ground cover of 1998-2019 from northern hemisphere daily snow cover products by the US National Ice Center’s Interactive Multi-sensor Snow and Ice Mapping System (USNIC-IMS).

USNIC-IMS data showed that earlier snow melt occurred in 2000-2010 (from DOY ~160 to ~150), but then began to reverse (DOY ~155 in last 3 years), with large spatial-temporal variations. Combined with slightly earlier ground snow-cover since around 2010, yearly snow-free period increased from ~115 to 130 days around 2010 and then dropped to ~120 days in recent years. Remarkably, offline ELM simulations, with GSWP3v2 forcing, exhibited those trends., and simulated yearly-averaged 2.4+/-3.0 days earlier snow-cover and 6.4+/-3.4 days earlier melt, and thus 6.6+/-3.9 days longer snow-free season, compared to the observations.

Model-data discrepancies may be caused by model forcing or snow algorithms or both, which partially contributed to vegetation phenological shifts (and to spatiotemporal LAI mismatches) in some Arctic regions identified by two Diagnosis Tools. For example, severe under-estimation of LAI (and thus SOM) and phenological shifts apparently exist in Northeastern Siberian Russia and Northeastern Canada. Those biases could be a consequence of late snow melt and a short snow-free season due to heavy or extended winter snowfall. Our study demonstrates that data integration, model development and fidelity assessments are critical to further improve ELM performance in the pan-Arctic.