The field of climate modeling has grown tremendously over the past several decades, and much of that growth has occurred in the international community. In the first Intergovernmental Panel on Climate Change (IPCC) report (IPCC, 1990), only three coupled ocean-atmosphere models were used for estimates of the transient evolution of global temperature in response to changing greenhouse gases. Those models were all from the United States (the Geophysical Fluid Dynamics Laboratory [GFDL], the National Center for Atmospheric Research [NCAR], and the Goddard Institute for Space Studies). Since that time the growth in climate modeling has been substantial—for the IPCC Fourth Assessment Report in 2007, 23 models were used from 11 countries around the world (Table 10.4 of the Fourth Assessment Report), and even more will likely be used in the upcoming Fifth Assessment Report, scheduled for completion in 2013. These include climate modeling centers in a wide range of countries, including Canada, the United Kingdom, Germany, France, Norway, Russia, Italy, China, Japan, Korea, and Australia. Computational resources associated with these international centers have likewise grown, including facilities such as the Earth System Simulator in Japan.1
INTERNATIONAL COORDINATION, ESPECIALLY AS IT RELATES TO THE INTERGOVERNMENTAL PANEL ON CLIMATE CHANGE
Systematic comparison of simulations using these models has proved highly beneficial. Since the 1990s the leading climate modeling efforts around the world have exchanged information and coordinated their efforts under the umbrella of the World Climate Research Programme (WCRP), an activity of the World Meteorological Organization of the United Nations. A number of working groups have sought to facilitate interactions and coordination of modeling activities. The Working Group on Numerical Prediction (WGNE) has coordinated activities involving weather prediction models. The Working Group on Seasonal to Interannual Prediction (WGSIP) has coordinated
efforts in developing and using coupled ocean-atmosphere models for seasonal to interannual prediction, with its primary focus on the El Niño/Southern Oscillation phenomenon. The Working Group on Coupled Modeling (WGCM) has coordinated coupled ocean-atmosphere models that are primarily developed and used for the study of decadal to centennial climate change projections. A subset of the WGCM, the Working Group on Ocean Model Development, has fostered the development worldwide of the ocean component of coupled models to improve the representation of the ocean component of coupled models.
The community as a whole, under the aegis of the WGCM and the WGNE of WCRP, with links to the International Geosphere-Biosphere Programme, comes to consensus on a suite of experiments, which they agree would help advance scientific understanding. The WGCM sponsors the Coupled Model Intercomparison Project (CMIP), a project that seeks to foster and coordinate the design and execution of simulations using models around the world that are subjected to a common experimental protocol. Meehl and Bony (2011), Stouffer et al. (2011), and Doblas-Reyes et al. (2011) describe the current protocol and how it has evolved. All the major modeling groups participated in defining the experiments and protocols and have agreed to the CMIP5 suite2 as a sound basis for advancing the science of secular climate change, assessing decadal predictability, and so forth (Taylor et al., 2012). The use of this common protocol is designed to facilitate the comparison of the various models used. Model output is freely available over the Web. The Program for Climate Model Diagnosis and Intercomparison (PCMDI), sponsored through the U.S. Department of Energy, has played a key role in archiving this model output and facilitating its wide public dissemination.
These common experiments have evolved significantly over the years. The first experiments were performed in the early 1990s with atmosphere-only models as part of the Atmospheric Model Intercomparison Project (AMIP). A key aspect of this early effort that set the tone for future success was an emphasis on making model output available for use by a wide community of users. This early AMIP effort then spawned a number of model intercomparison projects, including an Ocean Model Intercomparison Project, the Paleoclimate Model Intercomparison Project, and the widely known CMIP.
In addition to the output of such coordinated experiments, the various working groups serve as important mechanisms for exchange of information and ideas among modeling scientists around the world. U.S. scientists have benefited greatly from such interactions. These working groups sponsor internationally coordinated experiments
2 Currently ongoing at the time of this report.
with climate models, diagnostic projects across models, and international workshops to synthesize model results and foster increased understanding.
The intraseasonal to interannual community agrees on similar multimodel approaches for seasonal forecasting, for example, through the WCRP WGSIP and its Climate System Historical Forecast Project.3 A globally coordinated suite of experiments is then run, and results are shared for a comparative study of model results.
The data archives that result from all of these coordinated campaigns have spawned an entire new field of research in the interpretation of multimodel ensembles (e.g., Reichler and Kim, 2008; Santer et al., 2009), including studies of model genealogy and cladistics (see, e.g., Masson and Knutti, 2011) and uncertainty quantification (Tebaldi and Knutti, 2007). The data are stored in petabyte-scale (and soon exabyte-scale; see Overpeck et al., 2011) distributed archives. Providing access to these data, especially for users who may not be climate experts, is one of the primary challenges of the decade.
Finding 8.1: U.S. climate modelers are extensively involved in internationally coordinated activities, including the Coupled Modeling Intercomparison Project, the IPCC, and a suite of observational and modeling programs that are designed to advance climate models by improving processes-based understanding of important aspects of the climate system, such as clouds and their feedback on the climate system.
Finding 8.2: The U.S. involvement in such international activities contributes significantly to advances in U.S. climate modeling through leveraging international resources that are applied to climate modeling.
Finding 8.3: Modeling intercomparison projects create vast amounts of data that need to be curated, managed, made readily available, and analyzed.
INTERNATIONAL ACTIVITIES IN PROCESS-BASED STUDIES AND OBSERVING SYSTEMS THAT CAN LEAD TO IMPROVED MODELS
Under the auspices of WCRP, there are also numerous international activities aimed at testing the fidelity of model simulations of various specific physical processes, for instance sea ice, carbon dioxide (CO2) fluxes from vegetated surfaces, aerosol transport, or tropical cirrus clouds. Examples of such activities are the Global Atmospheric System Study coordinated through the Global Energy and Water Cycle Experiment
(GEWEX), the GEWEX Atmospheric Boundary Layer Study, and the Year of Tropical Convection (YOTC).
Typically, such activities evaluate how well a process is simulated by comparing a relatively new data set with data from a suite of participating models (e.g., from a field experiment, a ground observing network, or a satellite instrument). Participation is voluntary; usually a case leader will specify details of how the models are to be run (what days, boundary conditions, what fields to output) and then process all the model output for direct comparison with the observations.
This process is rarely as straightforward as it may appear. A focus on a single process (e.g., cirrus microphysics) requires other related processes to be constrained (e.g., cumulonimbus convection that first creates the cirrus) using observations or a best guess at the related meteorology; often the model results themselves suggest how to better do the comparison. An international intercomparison leverages the effort involved in setting up both the observations and the modeling protocol; most modeling groups can participate with relatively minor effort once the case is defined, and they get valuable analysis of their results essentially for free.
Many recent U.S.-led field experiments have from the start been designed in part for such an intercomparison. A partial list of such projects includes DYCOMS-II (the Second Dynamics and Chemistry of Marine Stratocumulus field study); the Rain in Cumulus over Ocean project, sponsored by the National Science Foundation (NSF); the North American Monsoon Experiment (NAME); the Tropical Warm Pool International Cloud Experiment; and the VAMOS Ocean-Cloud-Atmosphere-Land Study, a part of the Variability of the American Monsoon Systems (VAMOS) project. Intercomparisons have also been based around observation networks such as the Atmospheric Radiation Measurement sites, the AERONET aerosol monitoring network (AEROsol robotic NETwork), or the AMERIFLUX array of CO2 monitoring sites, or using new satellite data sets (e.g., the CFMIP Observation Simulator Package, within the Cloud Feedback Model Intercomparison Project [CFMIP] project).
U.S. funding agencies have supported these types of projects (e.g., as part of the NAME and VOCALS science plans) because of their potential for speeding the pace at which new observations are used to test and improve process representation in models. In some cases (e.g., National Oceanic and Atmospheric Administration/NSF Climate Process Teams), U.S. climate modeling groups such as GFDL, NCAR, and the National Centers for Environmental Prediction (NCEP) also have received dedicated funding to aggressively use such intercomparisons to improve their models.
Model intercomparisons allow modelers to see weaknesses of their simulations in focused settings, and also to see whether other parameterization approaches clearly work better. They are only one part of the road to actual model improvement because different process parameterizations can strongly interact so that a change in one parameterization (e.g., cumulus convection) may not have the same effect on overall results when applied to different climate models. However, leading modeling groups such as GFDL and NCAR in the United States and the U.K. Met Office, Max Planck Institute, and the European Centre for Medium-Range Weather Forecasts in Europe are typically participating in many intercomparisons at any one time. Their model development teams see great merit in this approach. NCEP has participated less, perhaps because of a lack of available manpower. Overall, the committee’s assessment is that voluntary process-oriented international intercomparisons greatly benefit U.S. climate models rather than being a distracting drain on resources.
Finding 8.4: International model intercomparison projects have proven to be effective mechanisms for advancing climate models because they leverage the effort involved in setting up both the observations and the modeling protocols used for testing, and they allow modelers to see weaknesses of their simulations in focused settings.
CURRENT CMIP/IPCC EFFORTS
In support of the Fifth Assessment Report of the IPCC, the CMIP has established an extensive suite of common modeling experiments that many centers around the world are executing. A goal is to make the output from such experiments widely and easily available so that scientists from around the world can analyze that model output in time for the results of such analyses to inform the Fifth Assessment Report of the IPCC (as of this writing, any paper that will be cited in the next IPCC report must be submitted by July 31, 2012). However, the CMIP archive will be available for many years to come, so that additional studies using this archive will likely occur well after the IPCC. The previous (CMIP3) suite of simulations from 2005-2006 had been used in 595 publications as of January 20124 and is still heavily utilized. There are more than 6,700 registered users of the CMIP3 archive, with new users continually registering for access; data are being downloaded from the archive at a rate of approximately 160
4http://www-pcmdi.llnl.gov/ipcc/subproject_publications.php (accessed October 11, 2012).
TB per year, with over 1 PB of data downloaded since the start of the CMIP3 project in 2005, corresponding to 3 million files.5
The CMIP activity has evolved from common experiments using models of just the atmosphere to models of the full Earth system, including oceans, interactive aerosols, and biogeochemical cycles. As described below, the current suite of CMIP experiments involves models of differing levels of complexity; these span a range from atmosphere-only models to more comprehensive coupled ocean-atmosphere models that include representations of ecosystems and various biogeochemical cycles, including the carbon cycle. In addition to model comprehensiveness, the suite of experiments conducted under CMIP has grown in diversity over the years. While the initial protocols consisted of very simple, idealized experiments, the full protocol for CMIP56 is extremely complex, entailing many thousands of years of model simulations (note, however, that there are several tiers of simulations of different priority levels, of which the highest-priority tier is less computationally demanding, and there is no requirement for even a leading modeling center to perform all requested simulations). This allows a much fuller examination of simulations but also entails significant costs. This general issue is discussed below. The experimental protocol entails designs for both long-term and near-term climate change predictions and projections, as well as a focused effort on evaluating the role of biogeochemical cycles and changes in the climate system and their potential future change. The model output from this archive is used to investigate a host of issues. These range from detailed analyses of the physical processes that operate in models in order to assess their credibility, to using this model output to assess the impact of projected climate change for various regions to estimate climate vulnerability and adaptation.7
Finding 8.5: CMIP outputs, including model outputs from models outside the United States, are a valuable resource for a wide range of activities, including estimating climate change impacts and adaptation planning.
BENEFITS AND COSTS OF INTERNATIONAL COLLABORATIVE EFFORTS
One of the fundamental challenges in the use of climate models for projections of future change is the very limited understanding of the uncertainties embedded within
5 Karl Taylor, PCMDI, personal communication, 2011.
6http://cmip-pcmdi.llnl.gov/cmip5/experiment_design.html?submenuheader=1 (accessed October 11, 2012).
7 See, for example, http://www-pcmdi.llnl.gov/ipcc/diagnostic_subprojects.php for a list of projects using CMIP3 output (accessed October 11, 2012).
any single model projection. The construction and use of a climate model represents a series of choices on many topics, including physical parameterizations and scenarios of future changes in emissions of radiatively active atmospheric constituents. Each of these is highly uncertain, and yet climate modelers are constrained in our choices due to various resource limitations. Thus, a projection based upon one model represents a single point within a very large parameter space.
A more robust assessment of future climate change arises with fuller coverage of this parameter space of uncertainty in model formulation and scenarios of future radiative forcing changes. Thus, many recent assessments of future climate change draw not only on the output of a single simulation, but also on the full suite of possible outcomes as drawn from the archives of past CMIP experiments. Although this assessment is still very far from a satisfactory estimate of the full range of possible future climates, it represents an invaluable guide. Thus, participation of modeling centers around the world in the CMIP suites of experiments contribute both to better estimates of future climate change and to model development and improvement. Such international coordination and exchange of information provide a vital exchange of ideas and techniques that improve climate modeling in the United States and around the world.
Model intercomparison programs, such as CMIP, provide timelines for model development and the execution of coordinated experiments. The process of climate model development is one in which there are often not obvious ending points. Models can be changed in an almost continuous fashion, with each change producing new simulations that must be carefully evaluated. This process often has no natural closure points and generally becomes longer as models become more comprehensive. However, participation in activities such as CMIP can provide clear schedules for the conclusion of such model development processes that can be used very effectively by modeling centers to define completion of the model development cycle. In fact, many of the model development cycles at centers around the world are now timed to the schedule of IPCC assessment reports. Although such a schedule can be a benefit by providing firm deadlines for concluding model development cycles, it can also be a serious detriment by artificially constraining the model development process and placing enormous strain on already underresourced model development efforts.
Participation in CMIP-like activities can also produce a healthy sense of competition among national and international modeling centers. The output from such coordinated experiments is routinely made available to researchers around the world, who provide evaluations and comparison among the models. Such activities often show
which models are among the world’s elite, and this can produce a very positive feedback in the ongoing model development process.
Finding 8.6: There are many benefits to the participation of U.S. climate models in the CMIP process, including defining timelines for model development and creating healthy competition among modeling centers.
However, there are also costs to participation in such efforts. Bringing closure to the model development process on any timeline is a difficult task, especially because modeling centers want to have the best possible physics and numerics in their models. These typically involve recently developed physical parameterizations based on new observational and theoretical research, and their behavior in complex models can be difficult to predict. The often unpredictable nature of newly developed model processes can create an environment of intense pressure to finalize a model with simulation characteristics that are superior to its predecessor model and to competitor models. This pressure can be exacerbated by the CMIP-derived timelines for coordinated model experiments and can lead to model decisions being heavily influenced by artificial time pressures rather than the best possible science. The effect of this process can lead to “burn out” among those most deeply involved in the model development process.
In addition, as described in Chapter 7, model development is an enormous task requiring substantial human and computational resources, yet the vast majority of this effort, including the production of model runs for CMIP activities, does not lead to peer-reviewed publications. Because such publications are usually the metric by which scientists are evaluated, participation in the model development process can sometimes hurt a young scientist’s career, at least in the short term as measured by publications. As noted in Chapter 7, a culture of coauthorship with model developers and the careful citation of model development papers could change this.
The benefits of CMIP-related model experimentation also have to be weighed against some lost opportunity costs, especially for the scientists directly involved in model development. Fundamental advances and new findings are often the result of research that is curiosity driven or inspired by an idea or question. The more time that a scientist devotes to large-scale science, as embodied by programs and activities such as CMIP, the less time is available for small-scale or curiosity-driven research. In addition, the full suite of CMIP simulations requires an enormous computational effort that can consume a substantial fraction of the computational resources available to a modeling center for a year or longer.
Finding 8.7: One cost associated with the effort and time pressure of participating in the CMIP/IPCC is the reduction of time and computational resources that model developers have to devote to fundamental research that produces results on longer time scales.
THE WAY FORWARD
For decades, the United States has sustained the largest climate research enterprise in the world. The first climate simulation model was developed in the United States, and the United States continues to support a diverse range of approaches to better understanding future extreme weather and climate on all space and time scales. A robust international climate modeling community has evolved, including state-of-theart efforts in Europe in regional and global modeling, as well as growing efforts in Asia supported by large new investments in computing. This has led to Earth system models that simulate the current climate more accurately and comprehensively than in the past, and the application of these and finer-scale, more specialized regional models to many societal and scientific problems, although model-related uncertainties in future climate projections remain substantial. In response to IPCC-type assessments, the international community, led by the United States, has pioneered mechanisms for distributing an ever-growing set of standardized outputs from international suites of models. These are a major resource for the U.S. climate community.
On balance the CMIP activities are a clear positive for U.S. climate modeling activities. These activities help to keep U.S. models and model-based research at the leading edge of activity around the world. However, the costs associated with these activities imply a need for balance among the various sorts of activities in modeling centers to achieve some optimal outcome, especially in light of the rapidly growing scope of CMIP experiments. These activities are important enough to be considered an expected part of the model development process and thus warrant sustained support. This includes support for participation in the CMIP/IPCC activities and for the systems to archive model output in a way that is freely and easily available to users. Such support would include (a) software specialists for the development and maintenance of data storage and distribution systems that meet the needs of the climate community and (b) the required hardware, including storage, transmission, and analysis capabilities. This support would likely include resources at the modeling centers that run the climate model simulations, as well as support for a centralized capability that coordinates this activity within the United States.
In addition, it is anticipated that over the coming decades climate change assessments will be conducted in the United States that focus on both national and regional scales, with an increasing emphasis on adaptation. The utility of such assessments is greatly enhanced through the active use of climate models both from the United States and from institutions around the world. The utility of a large number of models enhances the credibility of any such assessments by providing the potential for an improved assessment of the uncertainty of climate change projections. The U.S. participation in CMIP and related activities greatly facilitates the use of multimodel ensembles incorporating U.S. and international models.
U.S. modeling centers should be encouraged to participate in international activities, including the execution of internationally coordinated numerical experiments such as CMIP, and to make that data publicly available. In addition, there should be sustained support and encouragement for the participation of U.S. scientists in international activities in support of climate modeling and the use of climate models, such as those organized by WCRP, and for the systems to archive model output from leading U.S. climate models, and to make that output freely and easily accessible (this is discussed in Chapter 10).
Recommendation 8.1: To advance in the next 10-20 years, U.S. climate modeling efforts should continue to strive for a suitable balance among and support for
• the application of current generation models to support climate research activities, as well as national and international projects such as CMIP/IPCC;
• near-term development activities that lead to incremental but meaningful improvements in models and their predictions; and
• the investment of resources to conduct and capitalize on long-lead-time research that offers the potential for more fundamental and transformational advances in climate modeling.
Recommendation 8.2: The United States should continue to support the participation of U.S. scientists and institutions in international activities, such as model intercomparisons, including support for systems to archive model output, because such activities have proven effective in robustly addressing user needs for climate information and for advancing U.S. climate models.
Recommendation 8.3: To enhance their robustness, national and regional climate change/adaptation assessments should incorporate projections from leading international climate models as well as those developed in the United States.