Strategic Plan for the
Climate Change
Science Program
Final Report, July 2003

  This chapter also available as PDF file.

Authors of this Chapter

Final Report also is available as single PDF file (8.8 Mb)

See also:
 

Vision for the Program and Highlights of the Scientific Strategic Plan

Acronyms, Abbreviations,
and Units

This chapter introduces goals and objectives intended to provide accelerated scientific improvements in climate and climate impact models responsive to the needs of the Climate Change Science Program's (CCSP's) scientific research and decision support activities.

Models are essential tools for synthesizing observations, theory, and experimental results to investigate how the Earth system works and how it is affected by human activities. Models can be used in both a retrospective sense, to test the accuracy of modeled changes in Earth system forcing and response by comparing model results with observations of past change, and in a prognostic sense, for calculating the response of the Earth system to projected future forcing. Comprehensive climate system models provide the primary quantitative means to integrate scientific understanding of the many components of the climate system and, thus, are the principal tools available for making quantitative projections.

The CCSP modeling strategy consists of three goals: Improve the scientific basis of climate and climate impacts models; provide the infrastructure and capacity necessary to support a scientifically rigorous and responsive U.S. climate modeling activity; and coordinate and accelerate climate modeling activities and provide relevant decision support information on a timely basis. In order to achieve these goals, three modeling arenas will be implemented: 1) diverse and disparate research activities that represent new process understanding in models; 2) assimilation and integration efforts that employ new types of observations and tools and next generation understanding and model coupling; and 3) "high-end" climate models run for scenarios required in periodic scientific assessments or to achieve higher resolutions.

The CCSP strategy envisions two complementary streams of climate modeling activities. The first is principally a research activity, which will maintain strong ties to the global change and computational science research communities to rapidly incorporate new knowledge into a comprehensive climate and Earth system modeling capability. Closely associated with the research activity, but distinct from it, will be the sustained and timely delivery of predictive model products that are required for assessments and other decision support needs. The CCSP will ensure that a productive partnership is maintained between product-driven modeling activities and the discovery-driven modeling research program that will underpin its credibility and future success.

In his 11 June 2001 speech, the President asked his administration to work to "develop state-of-the-art climate modeling that will help us better understand the causes and impacts of climate change." In response to this directive, the program is addressing the following overarching question:

Based on recommendations in National Research Council reports on U.S. climate modeling (NRC 1999b, 2001d), the CCSP agencies initiated new activities to strengthen the national climate modeling infrastructure. These activities will accelerate the delivery of improved model products that are especially important for making climate simulations, predictions, and projections more usable and applicable to the broader research, assessment, and policy communities (see Annex D for definitions of "prediction" and "projection."). The new activities form the basis of a longer-term solution that will maintain the pace and progress of the basic research, while simultaneously creating a path for the rapid exploitation of new knowledge in model development, testing, and applications.

Virtually all comprehensive climate models project a warmer Earth, an intensified hydrologic cycle, and rising sea level as consequences of increasing atmospheric concentrations of greenhouse gases. However, projections of the details about the magnitude, timing, and specific regional impacts and consequences are variable (IPCC, 2001a, b). The program is placing the highest priority on research aimed at addressing known modeling deficiencies (see Chapter 4, Question 4.1). The following objectives in pursuit of Goal 1 describe CCSP's long-term approach.

The climate system responds in complex ways to changes in forcing that may be natural (e.g., variations in the magnitude of solar radiation reaching the top of the atmosphere) or human-induced (e.g., changing atmospheric concentrations of greenhouse gases). Several of the program's science elements will provide climate modelers with the best scientific estimates of past and expected future climate forcing factors -- for example, the Climate Variability and Change element for solar variability, the Atmospheric Composition, Carbon Cycle, and Human Contributions and Responses elements for radiatively-active trace gases and aerosols, and the Land-Use/Land-Cover Change and Ecosystems elements for land surface cover changes and energy exchanges.

The direct response of the climate to a change in forcing may be either diminished or amplified by feedback processes within the climate system itself. For example, warmer upper oceans will result in increased evaporation and, thus, increased concentrations of atmospheric water vapor -- itself a strong greenhouse gas -- a positive or amplifying feedback. Increased water vapor will alter cloudiness, which may be either a positive or a negative feedback, depending on the cloud height and type. Climate-induced changes at the land surface (e.g., through more intense and higher frequency droughts) may in turn feed back on the climate itself, for example, through changes in soil moisture, vegetation, radiative characteristics, and surface-atmosphere exchanges of water vapor.

Enhanced understanding and improved representation in models of the processes that influence climate will improve confidence in model forecasts and projections. Reductions in uncertainty will be measured by the degree to which differences between the major climate models as well as differences between observations and relevant model fields are reduced.

Past emphasis in climate modeling research has been on the development and testing of physical aspects of coupled atmospheric and ocean general circulation models (GCMs). This occurred primarily because climate models have their roots in numerical weather prediction models (atmospheric GCMs with some ocean coupling that treat primarily physical processes), and the available observations were those derived for the application of weather prediction models. Significant advances have been achieved on the physical aspects of climate modeling, but much more research is required (see Chapter 4).

In the 1990s, motivated by unresolved questions about long-term climate change, modeling efforts expanded to include additional components of the climate system, such as chemistry and biology, that are important to longer-term climate processes. Here too, much has been accomplished, but much remains to be done. In parallel with continued research into physical climate processes and modeling, the program will enhance efforts to more fully develop the chemical and biological components of climate models, including their human dimensions, in the context of a coupled interactive system, and also expand the atmospheric domain to include middle-atmosphere processes. This priority is motivated by the need to provide answers to pressing questions about long-term change and variability that may result from human-induced climate forcing involving chemical, biological, and human-induced processes.

Figure 10-1: The development of climate models over the last 25 years showing how the different components are first developed separately and later coupled into comprehensive climate models. Source: IPCC (2001a).

The Role of the Middle Atmosphere in Climate

The Interaction of Aerosols, Chemistry, Ecosystems, and the Hydrologic Cycle

Atmospheric Composition

Biogeochemistry and Ecosystems

High-Resolution Climate Model

Climate models and observations are intimately connected, as described under Objectives 1.3-1.5 below. Models must be evaluated and constrained by observations, which also serve to initialize models used for prediction. Models provide a dynamically consistent framework into which diverse climate observations can be assimilated to produce "value-added" data sets of gridded, continuous time series of hybrid field observations and modeled data.

Given the complexity of climate models, it is difficult to ascertain why a particular model performs "better" or "worse" than others in any given situation. Generally speaking, a measure of a model's quality is its ability to simulate the current climate (global averages, annual cycle, major modes such as ENSO and NAO, etc.) and the climate of the past centuries (19^th and 20^th) as measured by available observations. Therefore it is essential to carefully identify and document model deficiencies, such as systematic biases ("bias" refers to the tendency of a model's prediction to drift toward the model's climatology, which may be at variance with the real world). Furthermore, thorough diagnostics of model deficiencies (e.g., intercomparison of various models, comparison of models to observations, analysis of physical mechanisms using simplified or conceptual models, and carrying out model sensitivity experiments) are essential to identify sources of model errors, which then provide the basis for model improvement.

Near-Term Priorities

Some key examples of global climate model intercomparison projects include the Atmospheric Model Intercomparison Project (AMIP), the Coupled Model Intercomparison Project (CMIP), and the Paleoclimate Model Intercomparison Project (PMIP). The AMIP was initiated in 1989 for the purpose of identifying and documenting the differences among the various atmospheric models and observations, to provide a basis for model improvement. Nearly all atmospheric models in use today have been calibrated using the AMIP experimental protocol. The AMIP is closely integrated with the CMIP, and provides a calibration for understanding how coupled models respond to increasing levels of greenhouse gases. Model-model and model-measurement intercomparisons have been carried out for a number of component models that will ultimately need to be incorporated into global climate models. For instance, multi-dimensional stratospheric composition models have been intercompared with both each other and with observations in a series of "Modeling and Measurement" papers.

The CMIP, like the AMIP, is an activity of the World Climate Research Programme (WCRP). It was initiated in 1995 with the goal of collecting and intercomparing simulations from global coupled climate models -- models that operate over the complete global three-dimensional domain of the climate system, with components typically consisting of atmosphere, ocean, sea ice, and land surface. Virtually every global coupled model group worldwide is participating in the CMIP, including groups from Australia, Canada, China, France, Japan, Germany, South Korea, Russia, the United Kingdom, and the United States.

The PMIP, launched in 1994 and endorsed by the WCRP and the International Geosphere-Biosphere Programme, is an international project involving members of all the major modeling groups worldwide. The aims of the PMIP project are to improve understanding of the mechanisms of climate change by examining such changes in the past, and to evaluate the ability of climate models to reproduce paleoclimate conditions radically different from present-day climate.

Over the past several decades, progress in improving numerical weather forecasting has been achieved primarily through the continuing increase in model resolution and improvement in initializing (specifying the values of the model's variables at the start of the forecast) the prediction models. Initialization has been improved through the increase in available observations, particularly from remote-sensing platforms, and by advances in data assimilation techniques.

Improving climate prediction can follow a similar path. Although seasonal climate forecasting using coupled climate models is still at an early stage of development, the most significant impact on forecasts in the past decade has come through the use of ocean data assimilation to assimilate in situ observations for initializing the coupled models for El Nino-Southern Oscillation (ENSO) forecasts.

There is a critical need for an ongoing effort to provide complete descriptions of the present and past state of the atmospheric and oceanic components of the global climate system, together with continually updated data sets compiled in consistent ways to enable comparison of models with observations (see Chapter 12). As new climate observations are obtained, it is essential to place them in a historical context to enable the accurate evaluation of departures from normal, and trends and change in variability. New observations can provide additional information about climate when they are put in the context of past observations at uniformly-spaced points in space. This is accomplished through the systematic processing and integration of climate observations using the state-of-art climate models and data assimilation methods.

Regional climate models (RCMs, also called mesoscale models) are used in conjunction with GCMs to provide "downscaling" of climate variables for regional-scale predictions or projections. RCMs operate on scales that could not be accessed directly in GCMs due to computational capacity limitations. Thus, while global climate models will be approaching 100 km resolution in the foreseeable future, it is unlikely that in the near future they will reach the 10 km scale that regional models can simulate. RCMs represent the most prevalent current approach to dynamic downscaling. An alternative approach is statistical downscaling, for which the applicability to climate change is uncertain. Other dynamic approaches include stretched-grid GCMs, with uneven horizontal resolution, which can provide regional-scale resolution over certain domains.

RCMs can also be used for "upscaling" of information to test GCMs. To the extent that many small-scale processes, parameterized in general circulation models as "sub-grid scale," can be better simulated in regional models, they can provide feedback on the adequacy and limitations of the coarse-grid parameterization schemes.

While the finer resolution provided by regional-scale models is desirable, the quality of the resulting solutions is limited, except perhaps in better delineating orographic (mountain region) precipitation. There are several broad issues involved in the use of regional models. While such models may allow many physical processes to be incorporated on the scale at which they occur (e.g., rainfall in the vicinity of mountains), the physics on regional scales is often uncertain. The uncertainties have led to the existence of multiple versions of most RCMs. When utilizing them for climate change simulations, the different versions can produce results that differ from each other by as much as the observed climate changes they are attempting to simulate, which questions their use by policymakers and decisionmakers.

A second issue is consistency. Regional-scale models of necessity generally utilize physical parameterizations that differ from those of the GCM, if for no other reason than because parameterizations are often scale-dependent. The GCM supplies boundary conditions generated with one physics package, and the RCM utilizes these boundary conditions in conjunction with different physics. This effect is made worse in "two-way coupling," in which the mesoscale result is fed back to the GCM. Using a mesoscale model over the Rocky Mountains, for example, and not at the same time over the Himalayas, provides inconsistent forcing for the GCM planetary wave structure, which is affected by both mountain chains.

The principal U.S. agencies that support climate model development and application commissioned the National Research Council (NRC) to analyze U.S. modeling efforts as well as to suggest ways that the agencies could further develop the U.S. program so that the need for state-of-the-art model products can be satisfied. The NRC reports, The Capacity of U.S. Climate Modeling to Support Climate Change Assessment Activities (NRC, 1999b) and Improving the Effectiveness of U.S. Climate Modeling (NRC, 2001d), provided valuable guidance on how to improve U.S. climate modeling efforts. Also, the U.S. Global Change Research Program (USGCRP) issued High-End Climate Science: Development of Modeling and Related Computing Capabilities (2000), a report commissioned by the Office of Science and Technology Policy to make recommendations on climate modeling activities.

These documents emphasized four key points: 1) the acknowledged U.S. leadership in basic climate research that generates the knowledge base, which underpins both domestic and international modeling programs; 2) the limited ability of the United States to rapidly integrate the basic climate research into a comprehensive climate modeling capability; 3) the challenges, including software, hardware, human resource, and management issues, to routinely produce comprehensive climate modeling products; and most important, 4) the need to establish a dedicated capability for comprehensive climate modeling activities, including the global climate observations and data that support modeling.

The production of global model predictions of climate variability and change with sufficient resolution and veracity to provide useful regional information to decisionmakers requires a comprehensive computational infrastructure of computing resources, data centers, networks, and people. The success of this enterprise is predicated on a long-term commitment of the program to sustain the institutional support and investment required to maintain a resilient and state-of-the-art computational and information technology infrastructure.

High-end computing needs are often divided into the two broad and overlapping categories of capability needs and capacity needs. The former refers to applications that require the capability to do sophisticated, cutting-edge simulations that were impossible just a few years ago because the computing platforms to execute them did not exist. Typically, these simulations require the dedicated use of the most powerful computer available for several weeks at a time. Ensemble simulations with the current and next generations of coupled models fall into this category. In addition to capability resources, the CCSP requires a large amount of capacity resources to carry out the bulk (in terms of total computing cycles) of its modeling and analysis needs.

Currently, capability resources needs are met through a combination of dedicated and shared resources. The dedicated resources -- particularly the computers at the NOAA, NASA, DOE, and NCAR laboratories -- do not fully meet the needs of the modeling community. The shortfall is made up by additional resources acquired at several shared-access supercomputer centers by the researchers themselves, through individual proposals. These centers are not supported by the CCSP and provide computing through a competitive review process among many researchers from many fields of science and engineering. CCSP objectives cannot be met without these additional computing resources.

One example of such computing resources is DOE's National Energy Research Scientific Computing (NERSC) facility at the Lawrence Berkeley National Laboratory. To help meet computing needs for U.S. climate modeling, 10% of the computing cycles at NERSC will be made available to the broader scientific community in an open competition, with a special emphasis on climate modeling. Those cycles that might be allocated for climate modeling in that open competition would supplement the cycles at NERSC already used for climate modeling.

Possible collaborations with the Earth Simulator center in Japan, which has a computer with 20-50 times the capability of any existing U.S. machine for climate model applications, may help meet some of the near future needs. Unfortunately, this resource is now not directly accessible to U.S.-based researchers, because its lack of appropriate communication and mass storage and retrieval capacity requires users to be physically located at the center. Data generated at the center must be transferred to storage media that are then physically transported to data archives in the United States. This problem may be resolved in the near future.

The divergence in high-end computing architectures over the last decade has made developing models for high-end capability machines more labor-intensive, requiring the addition of a software engineering component to model development.

In addition to these primary capability computational resources, the CCSP requires a network of available capacity computing engines, data archives, and associated information technology infrastructure to make the model products readily available and accessible, so that further analysis and the development of secondary products, such as downscaled model information, can be used for research and assessment. While the current archive of model results totals several tens of terabytes (trillions of units of information), future model data archives, and associated observational data to evaluate the models, will consume tens of petabytes (thousand terabytes).

The development, testing, and application of climate system models requires environmental and computer technicians and scientists with expertise in a broad range of disciplines. The scientific disciplines include atmospheric physics and chemistry; ocean physics, chemistry, and biology; ice physics; biological ecology; geology; applied mathematics; and the interactions among them. The activity also requires computer software engineering to develop, test, and manage model code (which, for a state-of-the-art climate system model, is currently about 500,000 lines and is projected to grow to about one million lines over the next five years).

The NRC review of U.S. climate modeling (NRC, 2001d) recommended the following as high priorities for the Nation:

The dispersed and diverse nature of climate research, including climate modeling research, in the United States requires an integrating strategy to rapidly infuse new knowledge into the most complete models used to simulate and predict future climate states. At the same time, the ability to provide decision support requires a robust and ready modeling capability to perform specialized projections and simulations to inform policymaking. A multi-tiered CCSP strategy has evolved over the last several years to address shortcomings identified in NRC reports on U.S. integrated climate modeling efforts (NRC 1999b, 2001d), as well as to meet anticipated future demands. The strategy combines a "bottom-up" approach required to solve difficult basic research problems with a "top-down" approach to focus a component of modeling research on the needs for decision support.

At the most fundamental level of the strategy is the large number of basic research projects at universities, federal laboratories, and in the private sector, which, along with the larger centers, produce new knowledge required to further improve climate models.

The middle level includes modeling centers that conduct essential research and development for climate, weather, and data assimilation applications. These centers are of two types, the first of which are the large modeling centers that develop and maintain state-of-the-science global models but have primary missions other than century-scale global climate projection, although they may have some activities in that area. These centers include the NOAA National Center for Environmental Prediction, which conducts operation weather and climate prediction; the NASA Goddard Institute for Space Studies, which focuses on using satellite data to provide representations of climate forcing fields and uses these to model and study the climate's sensitivity to natural and human forcings; the NASA Goddard Space Flight Center, which focuses on the use of satellite data to generate research-quality data sets, to improve climate models, and to improve weather and coupled model predictions; the International Research Institute for Climate Prediction, which prepares and distributes internationally, seasonal-to-interannual climate prediction and impacts products; and the Center for Ocean-Land-Atmosphere Studies, which focuses on studies of the predictability of climate. The second type is a number of smaller centers that have focused research interests on specific questions in climate research.

The strategy includes, at its third level, two "high-end" climate modeling centers that will continue to develop, evaluate, maintain, and apply models capable of executing the most sophisticated simulations, such as those required for assessments by the Intergovernmental Panel on Climate Change (IPCC). These two centers, one based at the NOAA GFDL and the other based on the CCSM coordinated by NCAR, are complementary, cooperative, and collaborative. Both high-end modeling groups have a long legacy of successful climate change research that predates the IPCC process and have led U.S. participation in international modeling evaluations and assessments.

A healthy balance of resources (financial, human, and computer) among these three levels is essential to maintain a strong U.S. applied modeling program. Researchers collaborate extensively across all three tiers, ensuring the rapid flow of knowledge and understanding, as well as the definition of new problems. One example of such collaboration is support for a common modeling infrastructure and the Earth System Modeling Framework (ESMF) to optimize modeling resources and enable meaningful knowledge transfer among modelers. By adopting common coding standards and system software, researchers will be able to test ideas at any of the several major modeling centers and the centers themselves will be better able to exchange model components.

The multi-tiered strategy provides a structure by which the diverse contributions of the basic research community can be quickly utilized and integrated in state-of-the-art models used for climate simulation and prediction. At the same time, the strategy supports the decision support requirements of the CCSP to make model simulations available for policy and impacts studies.

A major CCSP objective is to develop scientifically-based global, time-dependent, multi-century projections of future climate change for different scenarios of climate forcing caused by natural variations and human activities. These projections are a primary form of scientific information to support decisionmaking about options to address the potential consequences of climate change. Development of capabilities to produce world-class climate change projections on demand to meet the needs of international and national assessments and other decision support requirements will bring new cohesion and coherence to the efforts of the U.S. climate modeling community. Given current scientific uncertainty and gaps in knowledge, it is essential that the United States maintain more than one high-end modeling center focused on long-term climate change, so that differing approaches to unsolved problems can be explored.

Both the CCSM and the GFDL support ongoing development of comprehensive climate system models, involving chemistry, biogeochemistry, and ecological processes. The centers are engaged in the development and application of distinct models. Their methods, innovations, and working hypotheses differ regarding many of the outstanding unresolved theoretical and modeling issues. The different approaches are essential at this stage, given the highly complex nature of the climate system with its numerous feedback mechanisms across a broad range of temporal and spatial scales.

Further, a comprehensive U.S. climate modeling strategy benefits from a measure of differentiation between the roles of the two centers. Despite an apparent overlap in responsibilities, the missions and structures of the two centers are more complementary than duplicative. The CCSM is an open and accessible modeling system that integrates basic knowledge from the broad, multidisciplinary basic research community for research and applications. The GFDL model development team participates in these community interactions and will focus on model product generation for research, assessments, and policy applications as its principal activity. GFDL models and products are integral to the development of the NOAA Climate Services program, which provides operational climate products and services to policymakers and resource managers. The GFDL maintains dedicated computer resources that can be allocated flexibly to meet mission requirements.

Climate scientists who conduct observational and empirical research into climate processes are often not well connected with modeling centers and model developers. The U.S. Climate Variability and Predictability (CLIVAR) program has developed and promoted a research strategy that involves the formation of Climate Process and Modeling Teams (CPTs) in order to improve collaboration between researchers and modelers. CPTs consist of process-oriented observation specialists, researchers, and individual process and parameterization modelers, working collaboratively with climate model developers. The teams are organized around an issue, model deficiency, and/or parameterization(s) that are generic to all climate models. An important distinction between CPTs and other more conventional model development research is the emphasis on directed teamwork, demonstrated progress, and delivery of products that will be tested and possibly implemented in climate models.

One of the great strengths of atmospheric, oceanic, and climate modeling in the United States is the variety, availability, and wide use of models. But this diversity has also led to duplication of effort and a proliferation of models and codes which, due largely to technical reasons, cannot interoperate (i.e., cannot be used interchangeably) and have been unable to keep up with and exploit advances in computing technology.

Climate models are increasingly being used to support decisionmaking. The predictive requirements are becoming more stringent. The demand for interoperability of climate model components has intensified as various modeling groups are engaging in collaborative research. Without exchangeable model components, it is often difficult to point to a component as a clearly identifiable cause of divergent results when one model is compared against another or against observations. In order to optimize modeling resources and enable meaningful collaborations among modelers, it is necessary to build common and flexible modeling infrastructure at the major modeling centers.

The common modeling infrastructure that will be implemented by the two centers will enable the exchange of model components between different modeling systems and facilitate analyses and intercomparisons of model results by adopting common coding standards for model components and common output formats, and developing common diagnostics packages. To achieve part of this commonality, the Earth System Modeling Framework (ESMF) project has been established. The ESMF is a community-wide software engineering effort to develop common software to facilitate interoperability of climate models on various hardware platforms, especially on massively parallel architecture platforms (i.e., many -- hundreds to thousands -- individual computational units operating in parallel, connected by data communication links).