model-evaluation

Multi-model evaluation of from the Low-Resolution Archive (LRA): three coupled storm-resolving models — IFS-FESOM, IFS-NEMO, and ICON — are compared against observational references (ERA5, ESA-CCI, EN4) over the historical period 1990–2014.

Developed on the DKRZ Levante HPC system.

Scientific purpose

DestinE Climate DT models run at storm-resolving resolution; their LRA output (r100, ~1° regular grid) enables systematic multi-decadal evaluation. These notebooks assess whether the three models reproduce observed climate signals consistently across independent components of the climate system:

• 2 m temperature — warming trends and detrended variability vs. ERA5
• Total cloud cover — means, trends, and Taylor statistics vs. ESA-CCI
• Ocean potential temperature — time–depth evolution vs. EN4
• Sea ice — concentration and thickness climatologies (inter-model)
• Surface energy closure — a process-space diagnostic linking radiation, clouds, and turbulent fluxes

Notebooks / workflow

Notebook	Purpose
comLRA_t2m.ipynb	2 m temperature: global-mean series, decadal trend maps vs. ERA5, detrended variability, seasonal-cycle removal, zonal means.
comLRA_tcc.ipynb	Total cloud cover: annual/multi-year means, model−ESA-CCI difference maps, Taylor diagram, 1990–2014 trends.
comLRA_thetao.ipynb	Ocean potential temperature: domain-mean time–depth (Hovmöller) anomalies for 3 models + EN4.
comLRA_si.ipynb	Sea-ice concentration & thickness: multi-year consolidation and climatological comparison maps.
comLRA_SLC.ipynb	Surface energy closure: process-space consistency of radiation, clouds, and turbulent fluxes.

Each notebook starts with "Why this notebook matters" and "Key techniques demonstrated" sections, then is self-contained and follows the same pattern:

1. Configure data locations (documented placeholder variables in the config cell near the top).
2. Load & consolidate monthly LRA NetCDF files (1990–2014) with xarray.
3. Compute climatologies, trends (per-gridpoint linear regression), variability, or domain means.
4. Compare against the observational reference (regridding with xesmf where needed).
5. Plot global maps (cartopy), Hovmöller diagrams, Taylor diagrams, and time series.

Example outputs

The notebooks retain their executed outputs, so representative figures can be viewed directly inside each notebook. Representative outputs include:

• 2 m temperature trends (models vs. ERA5)
• Arctic sea-ice concentration climatology (March & September)
• Total cloud cover trends (models vs. ESA-CCI)

Required Python packages

See environment.yml. Core dependencies: xarray/netcdf4, numpy, scipy, matplotlib, cartopy, xesmf (regridding), astropy-healpix (HEALPix quick-look section only).

The notebooks retain their executed outputs, so all results and figures can be browsed directly on GitHub without installing anything.

Expected input data

Paths are set via placeholder variables documented in each notebook's config cell:

Placeholder	Content
LRA_ROOT	DestinE LRA Phase 2 archive: climatedt-o25.1/<MODEL>/historical-*/r100/monthly/<var>_*.nc for IFS-FESOM, IFS-NEMO, ICON
EVA_DIR	Working directory for derived NetCDF files (consolidated series, trends, variability)
ERA5_DIR	ERA5 2 m temperature (regular lat/lon, 1940–2024)
ESACCI_DIR	ESA-CCI cloud fraction record (monthly, 1982–2016)
EN4_DIR	Extracted EN4.2.2 monthly ocean analysis files
FESOM_DATA_ROOT, MN5_DIR	Auxiliary data used only in optional side checks

Limitations

• The input data are not included. DestinE LRA output, ERA5, ESA-CCI, and EN4 datasets reside on DKRZ Levante and are subject to their own licences; the notebooks cannot be re-executed without access to equivalent data. Set the placeholder variables to your own copies.
• All embedded outputs and figures document the analysis as originally performed; printed paths in outputs show sanitised placeholders (e.g. <LRA_ROOT>).

Notes

• The original climate datasets are not included because of their size and access restrictions.
• Absolute HPC paths have been replaced with placeholder variables for portability.
• These notebooks are provided as examples of scientific analysis workflows and are intended primarily as a software engineering and research portfolio.