verification script of spanish team

[konradmayer]

related #6

[konradmayer]

To be able to run with the DeepR code, our output needs to be modified:

• datasets merged (as xarray has problems to merge the files) with cdo mergetime
• rename of variable for baseline and model dataset to prediction using cdo chname
• removing of the grid_mapping attribute in the samos data and selection of the prediction variable (i.e. removing the crs variable - if grid_mapping is existing cdo and nco operators won't remove the variable) with ncatted -a grid_mapping,prediction,d,, samos.nc followed by cdo select
• rename dimensions in baseline and model data to match the cerra dimension names: for the baseline e.g. ncrename -d x,longitude era5_regridded.nc, ncrename -d y,latitude era5_regridded.nc, then cdo chname,x,longitude,y,latitude era5_regridded.nc out.nc
• the validation script cannot handle na values - there are some present on the edges due to the cdo bilinear interpolation in the baseline as well as samos output. clip all data on the edges with ncks -d latitude,5,159 -d longitude,2,237 era5_regridded.nc -O out.nc

modified data currently at /scratch/klifol/kmayer/tmp/code4earth/validation

[konradmayer]

config file at resources/configuration_validation_netcdf.yml

# Define the anchor for the 'locations' list
locations_anchor: &locations
  [
    { "name": "Ibiza-Baleares", "lat": 38.9067, "lon": 1.4206 },
    { "name": "Mallorca-Baleares", "lat": 39.6953, "lon": 3.0176 },
    { "name": "Pyrenees-Spain", "lat": 42.5751, "lon": 1.6536 },
    { "name": "Madrid-Spain", "lat": 40.4168, "lon": -3.7038 },
    { "name": "Barcelona-Spain", "lat": 41.3851, "lon": 2.1734 },
    { "name": "Picos_de_Europa-Spain", "lat": 43.1963, "lon": -4.7461 },
    { "name": "Alicante-Spain", "lat": 38.3452, "lon": -0.4810 },
    { "name": "Valencia-Spain", "lat": 39.4699, "lon": -0.3763 },
    { "name": "Malaga-Spain", "lat": 36.7213, "lon": -4.4213 },
    { "name": "Almeria-Spain", "lat": 36.8381, "lon": -2.4597 },
    { "name": "Alboran_Sea-Spain", "lat": 35.9393, "lon": -3.2231 },
    { "name": "Balearic_Sea-Spain", "lat": 39.8223, "lon": 2.6480 }
  ]

validation:
  model_name: "samos"
  model_predictions_location: /scratch/klifol/kmayer/tmp/code4earth/datasets/model/ #/ssea/SSEA/C4E/DATA/TESTING/SAMOS/postprocessed
  baseline_name: "bicubic"
  baseline_predictions_location: /scratch/klifol/kmayer/tmp/code4earth/datasets/baseline/ #/ssea/SSEA/C4E/DATA/TESTING/PREPROCESSED/ERA5_regridded
  observations_name: "cerra"
  observations_location: /scratch/klifol/kmayer/tmp/code4earth/datasets/cerra/ #/ssea/SSEA/C4E/DATA/TESTING/PREPROCESSED/CERRA
  visualization_types:
    metrics_global_map:
      show_baseline: True
      color_palette: None
    sample_observation_vs_prediction:
      number_of_samples: 10
    time_series_for_a_single_site:
      locations: *locations
      temporal_subset: None
      aggregate_by: ["1D", "7D", "15D", "1M"]
      color_palette: None
    error_time_series_for_a_single_site:
      locations: *locations
      temporal_subset: None
      aggregate_by: [ "1D", "7D", "15D", "1M" ]
      color_palette: None
    error_distribution_for_a_single_site:
      locations: *locations
      temporal_subset: None
      color_palette: None
    boxplot_for_a_single_site:
      locations: *locations
      group_by: ["hour", "month", "season"]
      color_palette: None
  validation_dir: /scratch/klifol/kmayer/tmp/code4earth/validation/

[konradmayer]

job gets killed after running out of memory on vsci-dev (64GB RAM)

[konradmayer]

currently try to make it run on our SLURM cluster, but for some reason the scheduler does not give me more than 16GB RAM. IT is already investigating

[konradmayer]

high RAM demand obviously came from inconsistent dimension naming among datasets (and therefore higher dimensional output). there is another error now which i'll investigate further tomorrow

[konradmayer]

works now after all changes to out output as outlined in https://github.com/ECMWFCode4Earth/tesserugged/issues/11#issuecomment-1713829586 (still need to put this into a script)