Skip to content

Latest commit

 

History

History
69 lines (48 loc) · 3.79 KB

tests.md

File metadata and controls

69 lines (48 loc) · 3.79 KB

Description of tests available in clim-sanity-checker

Below you can find a brief description about each test available in this tool.

Note that each test only checks for a single metric. It is therefore important to compare results across tests to get a good overall impression of the confidence level for each field-

Welch's t-Test

This test compares the annual mean values against a set of references. It is able to detect whether two samples have equal means. The function used to do the statistical test is:
scipy.stats.ttest_ind
If a p-value is below 5 %, the test is not passed.

Additional processing for Welch's t-Test

After the standard preprocessing another processing step using CDO takes place:
cdo yearmean -fldmean -vertsum

This command sums the fields vertically (in case 3-D fields are present), reduces 2-D fields to one value, and computes the mean values per year.

Field Correlation Test

This test compares the spatial correlation of fields. To do so the CDO operator -fldcor is applied. To have more sensitive results the values are squared with the operator -sqr.

If the R^2 values is below 0.98 the test is not passed.

In order to apply this test, a reference netCDF is needed. Per default a reference is downloaded from the ftp-server of ETHZ. The link needs to be defined in a file having the same name and location as the desired f_vars_to_proc. As an example vars_echam-hammoz.csv and ftp_echam-hammoz.txt is a matching pair.

In case one wants to use a custom reference file, the file can be passed with the -f_pattern_ref argument.

Additional processing for Field Correlation Test

After the standard preprocessing another processing step using CDO takes place:
cdo timmean -yearmean -vertsum

This command sums the fields vertically (in case 3-D fields are present), computes the mean values of each gridcell per year,
and finally averages these values over the entire period (usually 10 years).

Normalized Root Mean Square Error Test

This test is looking at the normalized RMSE of a reference and the experiment. To do so the CDO operators sqrt -fldmean -sqr -sub are used.

If the normalized RMSE is above 0.15 the test is not passed.

In order to apply this test, a reference netCDF is needed. The handling of the references is identical as for the Field Correlation Test.

Additional processing for Normalized Root Mean Square Error Test

After the standard preprocessing another processing step using CDO takes place:
cdo timmean -yearmean -vertsum

Finally, to be independent from the absolute values of a field and its unit, a normalization step for both the reference and the experiment data takes place. For the normalization the CDO operators fldstd and fldmean are used.

Emission Test

This test checks if the emissions fed into the model as input are correctly processed during the model run. It computes the relative deviation of the averaged emissions over the entire simulated period. As a reference, the .csv file emis_base_ref.csv is taken.

This is not ideal an should be removed at some point, see Issue 4 https://github.com/C2SM-ICON/clim-sanity-checker/issues/14

In general a relative deviation of up to 1 % is still ok due to rounding errors of floating point operations.

Additional processing for Emission Test

After the standard preprocessing another processing step using CDO takes place:
cdo timmean -fldmean -vertsum

This command sums the fields vertically (in case 3-D fields are present), reduces 2-D fields to one value, and computes the average over the entire period.