Ideally, models are the only viable tool for such efforts, taking up where human minds become limited, by being able to incorporate as many of the interacting processes and influences as are understood in a quantitative manner that subdivides the world into tens of thousands of finite domains and that does not leave out what may incorrectly be thought to be minor influences which can have amplifying or moderating effects. While computers can do the needed hundreds of trillions of calculations that are necessary, scientists are careful to make sure they understand what is happening before they place high confidence in model results. To help in this process, the model experiments provide vast amounts of data to help understand how well models are matching real world behavior, why changes and events are occurring, whether the events were a result of human activities or simply natural fluctuations, and whether the results are dependent on aspects of the Earth system behavior that we understand well (and so should have high confidence in) or aspects that are relatively uncertain and not in accord with records of past climatic behavior (in which case we should withhold our confidence and focus our attention on research to further our understanding of the model results and to make indicated improvements).

Important examples of where GCMs in the United States and in other nations have demonstrated increasing skill are in their ability to represent seasonal changes in climate, to represent the departures from the normal seasonal to interannual pattern of changes in the low latitudes where the El Nino cycle increases eastern Pacific Ocean surface temperatures, to represent the few year cooling and subsequent recovery following major volcanic eruptions such as Mt. Pinatubo in 1991, to represent the climates of periods in the geologic past that had significantly warmer or colder climates, and to represent the recent gradual warming that may be attributable to human activities.

Gaining an Understanding of Model Results

Models are a tool for helping scientists understand what is happening. The results of models are evaluated in the context of all that is understood about Earth system behavior. One step in the process of gaining acceptance of model results is the publication of peerreviewed journal articles describing the results, a very rigorous process that both forces careful analysis by the model developers and calls into question those results where models and observations do not match. The progress of science is slowed when those putting forth and those criticizing model results do not equally participate in this process.

To achieve a synthesizing of scientific understanding that considers results from all sides of the spectrum, various bodies convene review and assessment panels. Most prominent internationally is the Intergovernmental Panel on Climate Change (IPCC) which periodically publishes assessment reports that are drafted by an international team of expert authors, reviewed by an international array of scientific experts, and then reviewed by the countries of the world, each responsible for organizing their own review effort. Preparation of the Second (pentadal) IPCC Assessment is underway, and the United States country review process has involved hundreds of invited expert scientists (spanning a wide range of perspectives on the issues), agency program leaders, and stakeholders from industrial and environmental groups. The IPCC and other reviews, for example by the National Academy of Sciences, provide a highly considered analysis of the state of scientific understanding and deserve great weight by decision makers.

Independent evaluations of the scientific results are another means for gaining understanding; however, they are often difficult to perform well and completely due to the wide range of knowledge needed and the limited time to pursue the process. This GAO report is an example of an effort for an independent analysis. Based on the request from Congressman Dingell, the GAO has investigated and analyzed research underway regarding GCMs in order to provide an enumeration of present limitations in the best available models. While

we have no significant difference with the GAO analysis in response to the very specific request, the U. S. Government's modeling agencies believe that contemporaneously providing a summary of what models do well is needed in order to provide perspective on understanding where science stands in making useful projections about future conditions.

What Models Are Projecting for the Future

To help focus in particular on the performance of models in support of the GAO study (and for other reasons), the Subcommittee on Global Change Research convened a group of scientists ranging from modelers to model critics and asked them to attempt to provide a consensus summarization of the current state of scientific understanding. While some doubted that this could be achieved, it has been accomplished. In brief, the group of scientists (after considerable discussion and written reviews, but ultimately without recorded dissent) agreed upon a series of statements with the appropriate degree of confidence indicated and that are supported both by model results and by overall scientific understanding. The resulting statement of the participants in the USGCRP Model Forum has just been published (in slightly condensed form) in EOS (the transactions of the American Geophysical Union) and is being published in full by the U. S. Global Change Research Program (a prepublication draft that may be released is included as an appendix to this response).

Extracting from the full report (which should be read in full to appreciate the basis for the statement), the scientists agree that:

• Human activities are increasing atmospheric concentrations of gases and this will enhance the natural greenhouse warming effect,

• Aerosol concentrations are increased due to human activities and they can exert a cooling influence on climate,

• Greenhouse gases (but not the aerosols) are generally long-lived in the atmosphere and the Earth will thus have a prolonged warming influence from human activities,

• Projected emissions of these gases will increase their concentrations significantly in the future,

• Stratospheric cooling and surface warming to be expected from past emissions of gases and aerosols have already started,

• Further warming over the next century is very probably in the range of 0.5 to 2 C (1 to 4 F),

• Sea level will very probably rise at an accelerating rate,

• Global precipitation and evaporation will increase but not uniformly everywhere, Arctic lands will very probably experience an amplified winter warming and midlatitude continents will probably experience drier summer conditions,

Changes in climate variability (including changes in tropical storm intensity) are possible but too uncertain to specify,

• Details of changes over the next 25 years are uncertain due to possible natural variations in the climate, and

• Biospheric feedbacks omitted from the models, while uncertain, could somewhat amplify or moderate the changes.

This set of statements indicates that significant progress in understanding has been made over the past decade; however, there was agreement that, while progress will be occurring each year, it will require another decade and more of research to significantly improve confidence in the projected regional details of the anticipated changes. The participants then pointed out nine specific opportunities where sustained or intensified research would bring important gains in understanding and predictive capabilities, a set that is closely related to the set of limiting uncertainties identified in the GAO report.

USGCRP Research on GAO-Identified Limiting Uncertainties

The USGCRP goal is to improve predictive understanding of the Earth system. Its research program is thus focused on reducing the limitations that restrict the abilities of models to make more accurate projections. In particular, significant resources are being devoted to removing and reducing the limitations identified by the GAO. We would offer the following comments on what we are doing to respond for each of the limitations identified by the GAO (we would note that the identified set of limitations is somewhat overlapping, and the set of USGCRP responses and activities is similarly so).

1. Inadequate Representation of Processes Affecting the Climate. The USGCRP is devoting approximately 30% of its $1.8B budget to conducting research aimed at improving scientific understanding of processes controlling and influencing the climate. Process studies, which are generally conducted with international cooperation, include major studies to understand the global water cycle, ocean circulation, cloud-radiation interactions, land surface processes, atmospheric chemistry, and other processes. In addition, the USGCRP is supporting a series of model intercomparison studies to improve understanding of how well models represent the present climate and, thereby, to identify which processes are not adequately represented and on which research should be focused. An important recent accomplishment in the development of models that more adequately represent Earth system processes is the development of the second generation Community Climate Model (CCM2) at the National Center for Atmospheric Research.

2. Exclusion of Critical Processes. It is important to understand that not all processes are equally important; most models include all of the processes that are most important, including radiation, water vapor, convection, sea ice, land surface exchanges, and many more. The USGCRP is supporting efforts by major modeling groups to include an even more complete set of processes in their models and is supporting a major international model intercomparison project (AMIP) to identify systematic errors in GCMs. Processes now receiving attention include sulfur aerosols, cloud water and cloud microphysics, surface exchange processes including vegetation, and horizontal and vertical mixing in the oceans. Substantial progress is now being made toward accurate simulation of longer term climate components, particularly the deep ocean, sea-ice, and terrestrial vegetation. A new global ocean model developed through an unprecedented collaborative effort among NCAR, the Naval Postgraduate School, and the Los Alamos National Laboratory has a resolution of less than 20 km (about 13 miles), which results in a significantly improved representation of oceanic heat transport. When coupled to improved atmospheric models (e.g., CCM2), significant improvements in long-term simulations will be realized.

There also remain critical biospheric and chemical interactions to include. In addition, and not mentioned specifically in the GAO report, it is essential to fully represent land and cryospheric components and their interactions in climate models. The USGCRP is moving actively to support such efforts.

3. Inadequate Representations of Interactions Among Variables [the text actually refers to Earth system components rather than variables]. The USGCRP is strengthening efforts to couple models of the atmosphere, the oceans, sea ice, and the land surface in order to provide models that can adequately represent the exchanges of fluxes which couple the system components. It is essential that all of these components, including also land glaciers, be represented in simulations of the long-term climate. Two specific examples of USGCRP efforts include the Climate System Modeling program of NSF and the Earth system modeling program of EPA.

4. Inadequate Representations of or Accounting for Feedback Mechanisms. Present models include representations of many of the most important feedbacks, but more and more

intense observational studies are indicating shortcomings that currently are present in the parameterizations. As indicated in response to the first point, the USGCRP is devoting about 30% of its budget to improving understanding of processes and feedback mechanisms. As one example, a number of programs are underway to better understand the role of cloud-radiation interactions, particularly as they relate to absorption of solar radiation, cloud changes in height and extent as climate changes, etc. The USGCRP is also studying past climatic conditions as a means of trying to identify potential surprises and changes that nature may have hidden from those studying the present climate.

5. Insufficient Computer Power. The Earth system is large and complex, with processes taking place on spatial scales from the microscopic to the global and on temporal scales from seconds to centuries (and beyond). Representing the necessary elements of this system needed to make projections of future climate requires extensive computer resources. While it does no good to compute when processes are not well enough understood, we understand enough about the most important processes to make informative calculations that are significantly beyond the computer resources that are currently available for this purpose. More computing power is needed not only for storing and analyzing vast amount of data but also for running high resolution GCMs for long-term simulations, and doing so multiple times to improve the statistical validity of the results. Such long-time simulations are needed to test how well the interannual to decadal variabilites are simulated, to test the climate changes under various scenarios (e.g., doubled carbon dioxide), and to carry out predictability studies for various forcing sensitivities.

To address the near-term problem, the USGCRP through NSF has established a special-use, dedicated climate system modeling computing facility known as the Climate Simulation Laboratory (CSL) in cooperation with NCAR. This facility will provide stateof-the-art computer resources and data storage systems for use in major modeling research simulations, especially those that will support the IPCC and other assessment efforts where computer resources are short. The facility is open to all investigators funded or supported by a U.S. university or federal or private not-for-profit laboratory. The CSL encourages simulations of multi-100-year runs with coupled climate models and very large ensembles of seasonal to interannual predictability runs. For the longer-term, DOE's CHAMMP program is already making progress in overcoming the challenges of using the new massively parallel computers(with their promise of greatly reduced cost per compute cycle) to address the climate issue.

6. Cold Start Error. There are three causes of the "cold start" error. One is that we do not have detailed observations of the oceans back in time as we do for the atmosphere; as a result, we can only start with conditions that approximate the actual conditions in the oceans at various times in the past. A second cause of the "cold start" error is the lack of understanding of ocean and atmospheric processes. To help alleviate these problems, the USGCRP is supporting, along with other nations, major field programs to observe and understand the oceans. New data sets, such as that from the TOPEX/POSEIDON satellite, will enable more realistic initialization of the global ocean, a major factor in "cold start" errors. The third cause of the "cold start" error is the unavailability of adequate computer resources. With adequate physical and spatial representation of the oceans and with the ability to carry out very long simulations, scientists will be better able to reduce the problem. Already, models with finer spatial resolutions are showing reduced flux imbalances between the atmosphere and oceans and this will help in not exacerbating the cold start problem.

7. Inability to Project Regional Changes in Climate. The ultimate goal of the USGCRP is to be able to project climatic changes on scales of interest to those evaluating the risks and benefits of climatic change. Achieving regional resolution in climate models requires both incorporation of additional processes and significantly increased computer resources (requirements increase by about a factor of ten for each halving of the spatial resolution).

As indicated above, the USGCRP is supporting both process research to improve understanding of what processes must be treated and to provide increased computer resources. As a result of these efforts, models are beginning to show improved representations of regional features over subcontinental areas such as the Sahel and the central United States.

In addition, to begin to explore regional changes, the USGCRP is supporting use of high-resolution, sub-continental scale models to look at domains such as the western United States. Such models cannot be run very long times, so do not yet provide climatic information but do provide some indications of how specific global climate changes may be manifested at the regional scale.

Summary

Overall, the USGCRP finds the GAO report an interesting and useful perspective on the most important factors limiting the credibility of the results of GCMs. At the same time, there are many important successes of GCMs that give credibility to major aspects of their results. We have attempted to describe these as well as the steps we are taking to address the important scientific limitations identified by the GAO in its report.