access_LandMask - ACCESS-NRI/accessdev-Trac-archive GitHub Wiki
Land mask and associated ancillary files
[Martin Dix 2011-04-01]
Procedure for generating ancillary files
Dave Bi generated a fractional land mask on the UM N96 grid by using the SCRIP interpolation with the MOM4 mask. Version 2 of this is vayu:~dhb599/SCRIP/um96_landseamask_fraction_20101208.cdf. Points with a land fraction of less than 1% were set to zero. This truncation removes 88 points, but only reduces the global mean land fraction by 0.001%.
Compared to the earlier change from the HadGEM3 mask to the ACCESS v1 mask, differences here are quite small. There are small decreases in land fraction around many coastlines. The major change is an increase in land fraction in far northern Canada where changes were made to improve the sea-ice dynamics. Global land fraction has decreased from 29.6% to 29.2%. There are 7 new land points created.
The script cherax:~dix043/src/python/um/access_landmask.py reads the netcdf file, applies the 1% truncation and writes the data into preexisting UM ancillary files qrparm.landmask and qrparm.landfrac. (It's much easier to overwrite the data in existing files than to get all the header fields correct from scratch.)
The Met Office ancillary program is installed on solar and cherax. Calculations here were done on solar.
The standard HadGEM2 orography (e.g. in vayu:/data/projects/access/HG2AO_ancils/qrparm.orog_new) was generated from the GLOBE1 orography while HadGEM3 (/data/projects/access/HG3AO40_umdata/qrparm.orog) uses the higher resolution GLOBE30. At N96 resolution this makes only a small difference. There are small differences in the namelist options for the two models, mainly to do with the degree of filtering applied. For the ACCESS orography I chose the HadGEM3 options. However the L_AUTO_LAKES=T case didn't work as expected and gave height variations over the Mediterranean and Black Sea. Using L_AUTO_LAKES=F gave better results.
export OROGOPTS="-d GLOBE30_AUS -n /g/sc/data/dix043/ancil.vn7.7/n96_access/orog.n96_access"
$ANCIL_PARENT/bin/ancil_top -model n96 -variant _access
Using the standard HadGEM2 grid as a test gives a height field that differs by at most a few m from the HadGEM3 version. The other subgrid scale fields used for the gravity wave drag are also very similar.
The GLOBE data has some very odd values over Australia compared to the AUSDEM 9s dataset from GA (see HighResolutionAustralianOrography). Surprisingly substantial differences show up even at N96 resolution. Note that the orography is filtered to remove wavelengths shorted than 4 grid lengths, so the Great Dividing Range is lower than it would be in a simple grid box average. The final file here uses the AUSDEM data over Australia.
Generating soil and vegetation
There is a multitude of options for generating the vegetation and soil fields. HadGEM2 uses a Clapp Hornberger soil hydrology scheme while HadGEM3 uses a van Genuchten scheme. These use ancillary files with the same parameters (e.g. soil suction) but the physical interpretation and values are quite different. It's vital to use parameters appropriate for hte model settings.
There are also several different soil datasets that may be used. Here we follow the Met Office choices in using IGBP soil data to generate the Cosby parameters for HadGEM3 and HSWD to generate the van Genuchten parameters for HadGEM3.
Note that CABLE uses a Clapp Hornberger hydrology scheme so must use the HadGEM2 parameters.
These options only affect the soil properties, the vegetation properties are identical. The dust ancillary differs because of the different source datasets. However both HadGEM2 and HadGEM3 use the IGBP dust file.
Options for HadGEM3 are
export VEGFOPTS="-g4 -s -t0.4 -h -rdefault -qmodis -3 -5 -A1 -B1 -CHWSD -DVG"
and for HadGEM2/CABLE
export VEGFOPTS="-g4 -s -t0.4 -h -rdefault -qmodis -3 -x -A1 -B1 -CIGBP -DCH"
Note there's an extra runtime argument related to filling in missing data. Run as
$ANCIL_PARENT/bin/ancil_top -model n96 -variant _access -search 200 -search_row 5
with soil files renamed with a _vg or _ch suffix.
Files produced are qrparm.veg_frac, qrparm.veg_func, qrparm.soil.dust_igbp, qrparm.soil_ch, qrparm.soil_vg. This process produces the MOSES 9 tile vegetation files.
The soil field that's most different between the schemes and so best used for checking which version is which is the saturated soil water suction. This is typically 0.15 with the CH parameters and 0.30 with VG.
The UM format ancillary files are in raijin:~/access/data/ancil/access_v2 and cherax:~dix043/access_mask_v2.# Attachments