.. _evaluation_icon: ICON Evaluation =============== This page provides evaluation of the ICON historical and projection simulations. For model description and forcing details, see :ref:`icon_model_historical_projection`. Assessment of the Historical Simulation ---------------------------------------- The Performance Index table below provides a comprehensive evaluation of model biases across a wide range of variables and regions. For most variables, ICON exhibits near-neutral performance, with values close to 1. Performance is particularly strong (PI < 0.5) for specific humidity, winds, air temperature, and sea surface salinity. Temperature-related variables over land and ocean show weaker performance (PI ~1.5), reflecting a persistent cold bias that also affects precipitation and radiation. .. _icon_eval_1: .. figure:: ../../../evaluation/general_evaluation/figures/climate_metrics.performance_indices.climatedt-o25.1.ICON.historical-1990.r1.png :align: center :width: 100% Performance Index table for the historical simulation. Performance values refer to the mean of CMIP6 models, with values below 1 indicating improved performance and values above 1 degraded performance. The cold bias is evident in the Gregory plot below (right panel), which relates global mean 2 m temperature to the top-of-atmosphere (TOA) energy imbalance over the simulation period. In 1990, ERA5 has a global mean temperature of 14.24°C, while ICON is colder by ~0.6°C and exhibits a negative TOA imbalance of ~−0.59 W m\ :sup:`-2`. ICON lies outside the ERA5 temperature spread throughout the simulation, confirming a persistent cold bias. At TOA, global net radiation rapidly falls within the CERES range, indicating a fast atmospheric adjustment. .. _icon_eval_2: .. figure:: ../../../evaluation/mn5/figures/ICON-Historical_SSP370_Tco2559_timeseries_Gregory_absoluteT-ICON.png :align: center :width: 100% Left\: Time series of the globally averaged annual surface air temperature in ERA5 and the historical and scenario ICON simulations. Fit-lines representing the trends over the overlap period between ERA5 and the simulations are also included. Middle\: Same figure but for the timeseries of the net heat fluxes at the top of the atmosphere (TOA), including observations from CERES. Right\: Gregory plot of the combined ICON simulations. The mean values and ranges of the observed TOA fluxes and global mean surface air temperatures are included for reference. Near-surface temperature ^^^^^^^^^^^^^^^^^^^^^^^^ The surface air temperature bias maps below further illustrate the cold bias in ICON. ICON displays a cold bias over the tropics (land and ocean) and a warm bias over land above the mid-latitudes. The land warm bias is likely linked to the Louis turbulence scheme, which enhances vertical mixing and increases surface temperatures. In the Pacific basin, ICON exhibits a cold tongue bias, which is also present in :ref:`icon_model_control`, suggesting an issue with the ocean spin-up. ICON's regional biases are comparable to the CMIP6 multi-model mean over both land and ocean. Biases in eastern boundary upwelling regions (e.g. Humboldt, Benguela) are reduced relative to CMIP6, likely due to the higher model resolution. .. _icon_eval_3: .. figure:: ../../../evaluation/levante/Fig/ICON_nk_tas_annual_bias_combined_cropped.png :align: center :width: 100% Spatial maps of the climatological biases of annual surface air temperature in the historical ICON simulation and the CMIP6 multi-model mean. Biases are computed against Berkeley Earth climatology over the period 1990–2014. Mean sea level pressure exhibits large-scale biases (see the Performance Index table above), with strong negative biases over the Southern Ocean, positive biases over the tropics, and slight negative biases over the Northern Hemisphere mid-latitudes, as shown in the sea level pressure bias maps below. The origin of these biases remains under investigation. .. _icon_eval_4: .. figure:: ../../../evaluation/levante/Fig/ICON_nk_psl_annual_bias_combined_cropped.png :align: center :width: 100% Spatial maps of the climatological biases of annual mean sea level pressure in the historical ICON simulation and the CMIP6 multi-model mean. Biases are computed against ERA5 climatology over the period 1990–2014. The tropical cold tongue bias shifts the intertropical convergence zone (ITCZ) and associated rainfall. The precipitation bias maps below show the regional bias of ICON in comparison to MSWEP and the CMIP6 multi-model mean. ICON produces a double ITCZ structure in the Pacific, consistent with a displacement driven by altered SST gradients. The shift of the Atlantic rainband reduces precipitation over the Amazon, resulting in a pronounced negative bias. Over the maritime continent and warm pool, ICON shows a positive and negative bias, respectively. .. _icon_eval_5: .. figure:: ../../../evaluation/levante/Fig/ICON_nk_pr_annual_bias_combined_cropped.png :align: center :width: 100% Spatial maps of the climatological biases of annual precipitation in the historical ICON simulation and the CMIP6 multi-model mean. Biases are computed against MSWEP climatology over the period 1990–2014. The TOA energy imbalance time series below shows the global mean TOA energy imbalance from 1990 to 2014, with CERES observations available from 2000 onwards. ICON captures the seasonal cycle accurately, and its annual mean is closer to CERES than the CMIP6 multi-model mean. Consistent with the Gregory plot above, ICON shows a negative energy imbalance at the start and a fast atmospheric adjustment. .. _icon_eval_6: .. figure:: ../../../evaluation/levante/Fig/ICON_nk_radiation_imbalance_timeseries.png :align: center :width: 100% Time series of the global-mean net top-of-atmosphere radiation (Earth's energy imbalance) from 1990 to 2014, comparing ICON with CERES observations and the CMIP6 multi-model mean. Consistent with the Performance Index table above, ICON simulates less sea ice than OSI-SAF observations, as shown in the sea ice extent time series below. ICON reproduces a strong seasonal cycle, but with systematically lower values. In the Southern Hemisphere, it shows a similar seasonal cycle to observations, while the Northern Hemisphere seasonal amplitude is underestimated. ICON shows a negative sea ice extent trend of about 0.56 million km\ :sup:`2` per decade in the Northern Hemisphere, comparable to OSI-SAF. In the Southern Hemisphere, ICON displays a neutral trend, whereas OSI-SAF shows a positive trend of ~0.31 million km\ :sup:`2` per decade. .. _icon_eval_7: .. figure:: ../../../evaluation/levante/Fig/ICON_nk_sea_ice_extent_timeseries.png :align: center :width: 100% Time series of monthly and annual-mean sea ice extent for the Northern and Southern Hemispheres, comparing ICON against OSI-SAF satellite observations and a CMIP6 multi-model ensemble. .. rubric:: Further evaluation Additional evaluation plots for the ICON simulations are available in the `Climate DT Evaluation Charts `_.