Differences to the COSIPY Model

The FRICOSIPY model has been developed as an adapation of the original COSIPY v1.4 model of Sauter et al. (2020). This section explains the principal differences between the two models.

Note

The listed changes are relative to the point of divergence at version 1.4; subsequent releases of COSIPY may be substantially different and incorporate some of the new features of FRICOSIPY.

\((1)\) Model Structure

The principal difference between the COSIPY and FRICOSIPY models lies in the redesigned model structure. The original COSIPY model uses a single 3-dimensional dataset \((x,y,t)\); FRICOSIPY instead decouples this large dataset into 3 seperate static \((x,y)\), meteorological \((t)\) and illumination \((x,y,t)\) input files. This design alteration is highly beneficial to simulations with large spatio-temporal resolutions as the combined memory overhead of these 3 input files is much smaller than original input file. In the FRICOSIPY model, the Dask distribution client simply indexes the relevant data for each spatial node \((x,y)\) and peforms any spatial interpolation during each nodal simulation.

Similary for the output dataset, the original COSIPY model possessed no means to reduce its size; the temporal frequency of the output dataset is directly coupled to the input. The FRICOSIPY model instead offers data aggregation into user specified output timestamps (eg. annual means or cumulative sums of variables).

The final significant design change is the implementation of periodic writing of output data into the result file, rather than waiting until all nodes have simulated. These adaptions mean that the FRICOSIPY model typically requires significantly less RAM than the original COSIPY model and is therefore much more accessible to users with modest computational resources.

Users with limited memory may be unable to activate parellisation (simulating multiple nodes simulatenously), but they should always be able to run the model – albeit at a slower speed.

\((2)\) Model Improvements

\((3)\) New Parameterisations

The FRICOSIPY model introduces a range of new paramaterisations in order to enhance the model's versatility. As of the latest 1.3 release, these are as follows:

Oerlemans and Klok, 2002 (solar radiation)
Bougamont et al. (2005) (surface albedo)
Essery and Etchevers, 2004 (turbulent fluxes)
Mattea et al. (2021) (three-phase precipitation model)
Hirashima et al., 2010 (Darcy's law percolation)
Marchenko et al. (2017) (preferential percolation)
Calonne et al. (2012) (hydraulic conductivity)
Shimzu (1970) (hydraulic conductivity)
Sturm (1997) (thermal conductivity)
Calonne et al. (2019) (thermal conductivity)
Yen (1981) (specific heat capacity)
Ligtenberg et al. (2011) (firn densification)
Katsushima et al. (2009) (snow metamorphosis)