Dry snow densification over ice sheets in the ORCHIDEE land surface model

The Antarctic and Greenland ice sheets are particularly vulnerable to global warming. Surface melt and runoff are increasing over Greenland, inducing a decrease in surface mass balance. Projections suggest that this process will accelerate in the future and could also affect the Antarctic ice sheet. Over ice sheets, snowpacks can reach several tens of meters and have the capacity to store and refreeze liquid water. This process directly impacts the amount of runoff and is strongly dependent on the physical characteristics of the snowpack, particularly the snow density governed by metamorphism and overburden pressure. Consequently, understanding and modelling the evolution of ice sheets requires an accurate representation of surface and internal snowpack processes.  However, many Earth system models have simplified snowpack schemes, often evaluated and adapted for seasonal snow but not for polar snow conditions.

Here we present an automatic method for initialization and calibration of densification in snowpack models, applied  to the ORCHIDEE model, the land surface scheme of the IPSL-CM Earth system model. ORCHIDEE includes an intermediate complexity representation of the snowpack with 12 snow layers and 8 ice layers. In this work, we use ORCHIDEE in offline conditions with atmospheric forcings from the polar-oriented regional atmospheric model MAR. We develop a snowpack initialization method adaptable to any snowpack thickness and model. To address the limitations of densification parameterizations for polar regions identified in ORCHIDEE, we use  an automatic tuning method known as History Matching to calibrate free parameters of the densification formulations. Calibration of 1D simulations over two characteristic dry-snow locations in Greenland and Antarctica enable us to improve densification across the rest of the ice sheets. We apply this method for two different types of density parameterizations and obtain similar good agreement with observed density profiles from the SUMup database. In the future, this methodology can be extended to other free parameters of the model, such as those associated with the albedo parameterization.