Imperative 1: Improved atmosphere and ocean component models of Earth System Models

Reduce the negative impact of biases in model representations of atmospheric and oceanic processes 

Prepared by: S. Bony, S. Griffies and Tim Stockdale.

Rationale

The Earth's climate has changed over the last century and is expected to further change in the future. Predicting the evolution of the climate system on time scales of seasons to decades and longer has never been more required than today to inform the decisions that society must take to adapt to natural climate variations or anthropogenic climate change. Climate models are the tools through which such predictions can be made. Some recent advancees in climate models based on the CLIVAR CMIP experiments are summarised below:

• More experiments to quantify model sensitivity and feedbacks.
• Earth System Models - interactive carbon cycle, atmospheric chemistry, ozone chemistry, land-surface schemes.
• Increased resolution - including NWP models, eddy permitting ocean models, stratospheric-resolving atmospheric models.
• Decadal prediction experiments to test a variety of initialisation techniques.

However, in spite of these advances, the skill of seasonal and longer forecasts is still substantially limited by model biases in the representation of atmospheric and oceanic processes. These biases in turn handicap the ability of modellers to deliver reliable information about regional climate fluctuations and change, with this information crucial for policy makers to implement adaptation measures.  

Scientific Background and Major Challenges

Opportunities: Many resources are currently available and potentially useful to evaluate models and help their development. For example, on the atmospheric side, a new generation of satellite observations using both passive and active sensors makes it possible to evaluate multiple aspects of the model simulations with consistent datasets and in a much more constraining manner (e.g. the CloudSat and Calipso satellites make it possible to evaluate for the first time the vertical structure of clouds). On the oceanic side, information from the ARGO profiling floats, complemented by satellite measurements, provide means for evaluating the water mass properties, energetics of currents, and fluctuations of surface ocean fields in ocean climate simulations. In addition, several instrumented sites, field campaigns, and observational arrays have been collecting a large amount of in-situ data, that can be used to better understand processes and to carry out process-oriented evaluations of models. Meteorological and oceanographic analyses have greatly improved during the last decade, and coupled ocean-atmosphere analyses are emerging. On the modelling side, the increase of computing power makes it possible to refine model resolution, and to use very fine-resolution models (useful to develop parameterisations) on larger domains and longer time scales. Finally, the modelling community is also becoming more coordinated at the international level. Through WGCM, WGOMD and WGSIP, coordinated experiments (e.g. CMIP5, CORE, CHFP) have been designed that will be used over the next five years to address issues such as the relative merits of different model initialisation techniques for intraseasonal to decadal predictions.

Model biases: Model biases occur over a wide range of time and space scales. Some biases have been persistent and long-standing problems for climate modellers, such as: the tendency of coupled ocean-atmosphere models to simulate a double ITCZ; a too far extended cold tongue; an erroneous diurnal cycle for tropical convection; difficulties in simulating correct characteristics of tropical atmospheric waves and the associated intraseasonal variability; biases in ocean water mass properties that impact heat and tracer storage and sea level; and in general with properly representing modes of climate variability. Incorrect simulations of mean climate patterns are problematic for seasonal prediction, and biases in the simulation of short-term climate variability (e.g. synoptic variability, intraseasonal oscillations) negatively affects the simulation of climate on longer time scales (e.g. interannual or decadal variability). Biases in the simulation of some processes (e.g. cloud and moist processes) are problematic for the simulation of climate on all time and space scales. 

Model Sensitivity: To gain confidence in the simulations of future climate changes, it is not sufficient to assess the quality of the models' mean climate. The sensitivity of climate and processes to a change in environmental conditions, or to an external forcing, also should be assessed and improved. It is the case in particular of cloud properties, whose sensitivity to changes in temperature, static stability or large-scale dynamics is critical for assessing cloud-climate feedbacks, but is poorly simulated by current climate models.  Additionally, parameterizations of ocean eddies must properly respond to changes in atmospheric forcing, such as observed wind stress changes in the Southern Ocean.

Earth System Models: Earth System Models (ESMs) must incorporate a broad suite of physical and biogeochemical processes. Interactions between these processes make biases in one component (e.g. simulation of precipitation by the atmospheric component) problematic for other components (e.g. the simulation of carbon fluxes between continental surfaces and the atmosphere). Therefore, the increasing complexity of climate models, and the development of ESMs, has not reduced the models' biases. On the contrary, it has made improvement of the basic atmospheric and oceanic components more imperative than ever.

Model resolution and initialisation: There is evidence that refining the horizontal and vertical resolution of models can reduce some biases (e.g. biases in the intensity of extreme precipitation events; simulation of mid-latitudes atmospheric baroclinic waves; structure of the oceanic boundary currents), but not all of them. Improving climate models therefore requires an enhanced fundamental physical basis upon which the models are framed. This need is particularly important for the parameterisations of subgrid-scale processes, as well as the treatment of boundary conditions and couplings between components. The initialisation of climate models is another aspect to be improved. Seasonal prediction has given us some experience, but is not clear how appropriate the methods used are for other parts of the climate system or for longer timescales. Improving the techniques of coupled model initialisation for use at a full range of timescales is another major challenge for the modelling community.

Strategic Plan

General strategy: The number of processes needing improvement in climate models is immense. To ensure a rapid improvement of ESMs, it is therefore important to identify the aspects of climate models associated with their most critical deficiencies, and for which the available resources are the most likely to lead to improvement. Some guidance is needed to identify these processes and to set priorities.

In addition, for cultural or practical reasons, it might not be straightforward to use the available observational or modelling resources for model improvement. In these cases, bridges need to be built between the large-scale modelling community and the communities involved in satellite observations, field campaigns or fine-scale process modelling.

Once the root cause of model deficiencies is identified, developments have to be made to improve models. This process usually requires expertise and experience, and is generally associated with a very small community of researchers. It is thus imperative to highlight and foster activities associated with model development, to attract and engage young scientists into this area, and recognise this activity as a fundamental part of climate research.

Guidance: The in-depth analysis of climate model simulations coordinated by WGCM (CMIP), WGOMD (CORE) or WGSIP (CHFP) will help to identify the systematic model errors that are problematic for several space and time scales and for many applications (e.g. both for synoptic variability, seasonal climate prediction and climate change projections). This leadership will then provide the necessary guidance for prioritising the model improvements that are most critical. The upcoming CMIP5 analysis holds this potential. Namely, as each model will be used in a large variety of experiments and configurations (coupled, atmosphere-only, aqua-planet), it will be possible to test robustness of the model biases and to better identify possible root causes. For these analyses to contribute to the improvement of climate models, it will be imperative to synthesise the results obtained from all analyses, and to facilitate discussions and the communication between the different WCRP and IGBP communities (e.g. those focused on the ocean, the stratosphere or the cryosphere) so that the possible cause of biases may be more clearly identified and specific improvements proposed. Specialised workshops are required for this purpose.

Some errors commonly seen in long-term climate simulations may be revealed after relatively short model integrations (a few days to a season) when models are run in a prediction mode. This approach facilitates the testing of hypotheses about the origin of these errors and of likely remedies. Experimental protocols to help modellers run climate models in weather prediction mode (aka transpose-AMIP) are being established to foster this type of evaluation. These new modes of evaluation of climate models should be used more widely over the next few years.

Bridges: Comparing climate model simulations and satellite observations is not always straightforward. In some cases, it requires bridges across communities, such as satellite simulators to diagnose from model outputs what satellites would observe if they were flying above an atmosphere similar to that simulated by the model. This approach will be widely used over the next few years for evaluating the atmospheric component of ESMs using the A-Train constellation of satellites (e.g. CloudSat, Calipso, Parasol), geostationary satellites (e.g. ISCCP), and hopefully other satellites (e.g. TRMM, GPM).

To test and improve models through comparisons between model outputs and in-situ data from instrumented sites or field campaigns, a promising approach consists in running a single-column model version of a climate model over this site by using large-scale forcings from meteorological analyses. Such an approach is at the basis of the GCSS (GEWEX Cloud System Study) strategy to improve atmospheric parameterisations using in-situ data. It should be developed over the next few years.

From the ocean perspective, there has traditionally been a disconnection between oceanographers focused on particular physical processes, and ocean climate scientists analysing features of CMIP simulations. For CMIP5, however, there is a broad new suite of fields to be archived from the ocean model components that will foster studies that ask questions about how model parameterisations impact simulated climate mean, variability, predictability, and stability. This work represents just a beginning, with more input by process scientists needed for the design of large model comparison projects, such as CMIP5 and CORE, to ensure that models archive information necessary to address mechanistic questions related to fundamental processes.

Finally, better bridging between observations and modelling might require access by the modelling community to a large ensemble of observations. Those ensembles may be distributed upon the example of climate model simulations (e.g. the multi-model CMIP3 database at PCMDI), using a single portal, and a specified format and documentation.

More importantly, it is essential that once specific deficiencies have been identified among models, specific Climate Process Teams (CPTs) be established to tackle the problem through a combination of observational analysis, theories, climate modelling, and process studies. Such an approach has been very fruitful in the past (e.g. US-CLIVAR CPTs).

Climate Process Teams are set up to tackle key problems - a few strong collaborative actions may then be organised as CPTs or task forces among WCRP projects (GEWEX, CLIVAR, SPARC, CLIC) and working groups (WGCM, WGOMD, WGSIP), in collaboration with WGNE (for the development of parameterisations), WWRP (NWP techniques for models' initialisation and climate prediction) and IGBP (e.g. AIMES for biogeochemical processes).

Collaborations among these different bodies have already been in place for the CMIP5 experimental design. This should facilitate further collaborations on the improvement of climate models.

Back to About Us