Tier II: Extended variables — OHC, surface fluxes, clouds, IWV

[duncanwp]

Description

Implement scoring for extended (optional) Tier II variables.

Variables and datasets

• Ocean heat content (0–100m, 0–2000m): Argo, Cheng et al., Levitus
• Surface fluxes (latent, sensible, LW, SW): CERES SYN, Ameriflux, BSRN, OceanSITES
• Cloud properties (LWP, cloud fraction, CTP): CERES, MODIS, CloudSat/CALIPSO, ISCCP
• Integrated water vapor: RSS, ERA5, radiosondes

Specification

• Same CRPS/NRMSE/spatial correlation framework as core variables
• Optional but contributes additional credit to overall score

[duncanwp]

Proposed solution: adapt ICONEval's cloud profiles and maps

ICONEval has extensive cloud diagnostics directly applicable here:

• Clouds/IWV: recipe_clouds_maps computes multi-obs mean for clivi, clt, clwvi, lwcre, lwp, prw, swcre against CLARA-AVHRR, CloudSat, ERA5, ESACCI-CLOUD, MERRA2, MODIS
• Cloud profiles: recipe_clouds_profiles provides vertical structure evaluation
• Surface fluxes: recipe_basics_timeseries includes hfls, hfss timeseries with ERA5/MERRA2 reference
• OHC: Not covered by ICONEval — continue using the existing Python pipeline (model_benchmark.py OHC path with thetao/so)

Port the cloud recipes with CMIP6 multi-model placeholders and add RMSE/MAE scoring to the results CSV.

Phase: 3 (Tier II observational benchmarks) in the integration plan
ICONEval reference: recipe_clouds_maps.yml, recipe_clouds_profiles.yml, recipe_basics_timeseries.yml