Editing IPCC:AR6/WGI/Chapter-1 (section)

==== 1.5.3.2 Model Tuning and Adjustment ====

<div id="h3-30-siblings" class="h3-siblings"></div>

When developing climate models, choices have to be made in a number of areas. Besides model formulation and resolution, parameterizations of unresolved processes also involve many choices as, for each of these, several parameters can be set. The acceptable range for these parameters is set by mathematical consistency (e.g., convergence of a numerical scheme), physical considerations (e.g., energy conservation), observations, or a combination of factors. Model developers choose a set of parameters that both falls within this range and mimics observations of individual processes or their statistics.

An initial set of such choices is usually made by (often extensive) groups of modellers working on individual components of the Earth system (e.g., ocean, atmosphere, land or sea ice). As components are assembled to build an ESM, the choices are refined so that the simulated climate best represents a number of pre-defined climate variables, or ‘tuning targets’. When these are met the model is released for use in intercomparisons such as CMIP. Tuning targets can be one of three types: mean climate; regional phenomena and features; or historical trends ( [[#Hourdin--2017|Hourdin et al., 2017]] ). One example of such a goal is that when the simulated climate system receives energy from the sun in accordance with what we observe today, the resulting mean equilibrium temperature should also be close to observations. Whether tuning should be performed to facilitate accurate simulation of long-term trends such as changes in global mean temperature over the historical era, or rather be performed for each process independently such that all collective behaviour is emergent, is an open question ( [[#Schmidt--2017|Schmidt et al., 2017]] ; [[#Burrows--2018|Burrows et al., 2018]] ).

Each modelling group has its own strategy and, after AR5, a survey was conducted to understand the tuning approach used in 23 CMIP5 modelling centres. The results are discussed in [[#Hourdin--2017|Hourdin et al. (2017)]] , which stresses that the behaviour of ESMs depends on the tuning strategy. An important recommendation is that the calibration steps that lead to particular model tuning should be carefully documented. In CMIP6 each modelling group now describes the three levels of tuning, both for the complete ESM and for the individual components (available at https://explore.es-doc.org and in the published model descriptions, Annex II: Models). The most important global tuning target for CMIP6 models is the net top-of-the-atmosphere (TOA) heat flux and its radiative components. Other global targets include: the decomposition of the energy fluxes at TOA into a clear sky component and a component due to the radiative effect of clouds, global mean air and ocean temperature, sea ice extent, sea ice volume, glacial mass balance, and the global root mean square error of precipitation. The TOA heat flux balance is achieved using a diversity of approaches, usually unique to each modelling group. Adjustments are made for parameters associated with uncertain or poorly constrained processes ( [[#Schmidt--2017|Schmidt et al., 2017]] ), for example the aerosol indirect effects, adjustments to ocean albedo, marine dimethyl sulfide (DMS) parameterization, or cloud properties ( [[#Mauritsen--2020|Mauritsen and Roeckner, 2020]] ).

Regional tuning targets include: the AMOC, the Southern Ocean circulation, and temperature profiles in ocean basins ( [[#Golaz--2019|Golaz et al., 2019]] ; [[#Sellar--2019|Sellar et al., 2019]] ); regional land properties and precipitations ( [[#Mauritsen--2019|Mauritsen et al., 2019]] ; [[#Yukimoto--2019|Yukimoto et al., 2019]] ) ; latitudinal distribution of radiation ( [[#Boucher--2020|Boucher et al., 2020]] ); spatial contrasts in TOA radiative fluxes or surface fluxes; and stationary waves in the Northern Hemisphere ( [[#Schmidt--2017|Schmidt et al., 2017]] ; [[#Yukimoto--2019|Yukimoto et al., 2019]] ).

Even with some core commonalities of approaches to model tuning, practices can differ, such as the use of initial drift from initialized forecasts, the explicit use of the transient observed record for the historical period, or the use of the present-day radiative imbalance at the TOA as a tuning target rather than an equilibrated pre-industrial balance. The majority of CMIP6 modelling groups report that they do not tune their model for the observed trends during the historical period (23 out of 29 groups), nor for ECS (25 out of 29). ECS and TCR are thus emergent properties for a large majority of models. The effect of tuning on model skill and ensemble spread in CMIP6 is further discussed in [[IPCC:Wg1:Chapter:Chapter-3#3.3|Section 3.3]] .

<div id="1.5.3.3" class="h3-container"></div>

<span id="from-global-to-regional-models"></span>