CHAPTER THIRTEEN

Strategies for Optimizing U.S. Institutional Arrangements

The current U.S. institutional structure for climate modeling consists of multiple centers that develop and use climate models in largely independent efforts. These institutions coincide primarily with U.S. funding agencies, and this structure has arisen primarily for administrative and historical reasons. Large global climate models are primarily run at larger modeling centers (described below). Universitybased research helps efforts to better understand processes in the climate system that can advance theoretical understanding of the climate system and lead to improved parameterizations in models, often utilizing models and model output from the large centers. Model development efforts involving both these communities are fostered by activities such as National Science Foundation (NSF)/National Oceanic and Atmospheric Administration (NOAA)-sponsored Climate Process Teams (CPTs). Regional climate modeling is mainly done by small groups at universities and national laboratories.

Climate modeling in the United States has efforts aimed at both global and regional modeling. There is overlap and interaction between the two, but in this report they are discussed separately for ease of presentation.

CURRENT GLOBAL CLIMATE MODELING ACTIVITIES IN THE UNITED STATES

There are several core global climate modeling efforts within the United States, complemented by scientists at a variety of other institutions. For this discussion, a “core modeling effort” is an activity that meets most or all of the following criteria:

•  builds complete climate models for use on seasonal to centennial time scales, and includes state-of-the-art representations of the ocean-atmosphere-landice system, as well as carbon and biogeochemical cycling;

•  develops models with spatial resolution and scientific capabilities that are consistent with state-of-the-art models used internationally; and

•  has efforts that are not continually divergent, but that periodically bring together model branches into a central core for ongoing coordinated development.



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 223
CHAPTER THIRTEEN Strategies for Optimizing U.S. Institutional Arrangements T he current U.S. institutional structure for climate modeling consists of multiple centers that develop and use climate models in largely independent efforts. These institutions coincide primarily with U.S. funding agencies, and this struc- ture has arisen primarily for administrative and historical reasons. Large global climate models are primarily run at larger modeling centers (described below). University- based research helps efforts to better understand processes in the climate system that can advance theoretical understanding of the climate system and lead to im- proved parameterizations in models, often utilizing models and model output from the large centers. Model development efforts involving both these communities are fostered by activities such as National Science Foundation (NSF)/National Oceanic and Atmospheric Administration (NOAA)-sponsored Climate Process Teams (CPTs). Regional climate modeling is mainly done by small groups at universities and national laboratories. Climate modeling in the United States has efforts aimed at both global and regional modeling. There is overlap and interaction between the two, but in this report they are discussed separately for ease of presentation. CURRENT GLOBAL CLIMATE MODELING ACTIVITIES IN THE UNITED STATES There are several core global climate modeling efforts within the United States, com- plemented by scientists at a variety of other institutions. For this discussion, a “core modeling effort” is an activity that meets most or all of the following criteria: • builds complete climate models for use on seasonal to centennial time scales, and includes state-of-the-art representations of the ocean-atmosphere-land- ice system, as well as carbon and biogeochemical cycling; • develops models with spatial resolution and scientific capabilities that are consistent with state-of-the-art models used internationally; and • has efforts that are not continually divergent, but that periodically bring together model branches into a central core for ongoing coordinated development. 223

OCR for page 223
A N AT I O N A L S T R AT E G Y F O R A D VA N C I N G C L I M AT E M O D E L I N G It is the assessment of this committee that a number of efforts in the United States meet some or all of these criteria. The two core modeling efforts that meet all of the criteria are • the National Center for Atmospheric Research (NCAR), supported by NSF and the Department of Energy (DOE); and • the Geophysical Fluid Dynamics Laboratory (GFDL), supported by NOAA. Additional efforts that meet some of the criteria are • the Goddard Institute for Space Studies (GISS), supported by NASA, focusing on decadal to centennial climate change; • the National Centers for Environmental Prediction (NCEP), supported by NOAA, focusing on seasonal prediction; and • the Goddard Global Modeling and Assimilation Office (GMAO), supported by NASA. This somewhat distributed system for U.S. model development has evolved over more than three decades with a legacy of funding support among different agencies, as well as differing modeling missions. Early results from the GFDL and GISS models provided much of the basis for the NRC (1979) assessment of the climate change expected from increasing carbon dioxide. NCAR began global modeling activities in the 1960s. Efforts in global modeling were also initiated at a number of universities, such as the University of California, Los Angeles. In part because of the large infrastructure that is required on an ongoing basis, the efforts at comprehensive global climate system modeling in the United States are primarily sustained at large national centers, while drawing upon expertise from universities and other partners. The GFDL and NCAR modeling efforts continue to focus on modeling of climate change and variability on time scales of seasons to centuries, with a strong emphasis on long-term projections. NCAR has partnered with DOE and the university research community to help provide the scientific and computational resources needed to sustain its effort. NASA-GISS, at a much smaller level, has also continued to focus on long-term climate change. All three centers have contributed to the Intergovernmental Panel on Climate Change assessments since they began. GFDL and NCAR were designated as the two primary U.S. climate modeling centers in the 2003 report of the U.S. Climate Change Science Program (CCSP1) (CCSP, 2003). Other major U.S. global modeling activities have focused on other objectives: NCEP on operational weather and climate predictions on time scales from days to seasons, and 1  CCSP is now known as the U.S. Global Change Research Program (USGCRP). 224

OCR for page 223
Strategies for Optimizing U.S. Institutional Arrangements NASA’s GMAO on the the simulation and global gridded analysis of current climate, in conjunction with the assimilation of satellite and other data. These centers vary in the size and scope of their activities devoted to modeling (see Chapter 7 for a discussion on the climate modeling workforce). The two largest centers in the United States are NCAR and GFDL. USGCRP (2011) estimates that of the $2.18 billion spent annually by federal agencies on climate research, 11 percent (~$239 mil- lion) is allocated for “improving our capability to model and predict future conditions and impacts.” That spending supports activities in both global and regional model- ing at the large modeling centers, as well as smaller activities in federal laboratories, universities, and private companies. Finding 13.1: The United States has a distributed system for global climate mod- eling, with a small number of “core modeling efforts.” These efforts have a long history and their structure derives from both modeling functions and agency funding structures. There is a separation of modeling activities across time scales, with operational weather and seasonal prediction centers largely sepa- rated from longer-term climate variability and change efforts. CURRENT REGIONAL MODELING ACTIVITIES IN THE UNITED STATES Regional climate modeling activities are focused on developing and using climate models with fine spatial resolution to better resolve small-scale climate features over a limited geographic domain. These models can be defined only over this limited domain, with specified boundary conditions at the perimeter of the domain, or they can be global models with varying spatial resolution in which the fine resolution is focused over the region of interest. For the limited domain regional models, boundary conditions can be supplied from a reanalysis or from some other climate model, for example, from a global simulation of future climate change. Regional modeling activities are also distributed in the United States. Some have primary affiliation in universities, while others have strong affiliations with some of the modeling centers described above. Most of the regional models used are derived from models developed at one of the global modeling centers. For example, a number of regional modeling efforts use the Weather Research and Forecasting (WRF) regional model that is developed through efforts involving NCAR and NOAA. This modeling system is then tailored to specific applications in various institutions according to their scientific foci and goals. WRF supports many different options for physical parameter- ization, and a centralized effort has not yet evolved to quantitatively evaluate which of these options are most appropriate for a regional climate model. Based on local expe- 225

OCR for page 223
A N AT I O N A L S T R AT E G Y F O R A D VA N C I N G C L I M AT E M O D E L I N G rience and history, different institutions are making different choices of such options; WRF is a “multiflavored” climate modeling platform. The multiple options available allow and even foster innovation but also make it much more challenging for a user of such regional climate model simulations to be assured of their credibility. MIPs like CORDEX2 and NARCCAP3 will be helpful in assessing the credibility of regional climate simulations. Finding 13.2: The United States has a distributed system for regional climate modeling, hosted both at national laboratories and at universities. The underly- ing models are used in a variety of applications and have not been as systemati- cally evaluated and intercompared for climate applications as global models. STRENGTHS AND WEAKNESSES OF CURRENT INSTITUTIONAL ARRANGEMENT Strengths The current institutional arrangements have many advantages and have fostered world-class climate modeling activities in the United States. One of the strengths has been the development of a cadre of talented scientists at each institution that con- tribute to the development and use of state-of-the-art models on a long-term basis. Model development is a long-term enterprise, so a stable team of scientists, supported by stable funding, is needed. Such teams provide important institutional memory. The current system has also effectively entrained talented researchers at institutions outside the primary centers into the model development activities. As mentioned above, activities like the CPTs, which are funded by NSF and NOAA, seek to leverage the talents in both universities and national laboratories to make progress on major uncertainties in climate models. The existence of multiple climate modeling centers in the United States has led to a healthy diversity of activity and the benefits of competing approaches. For example, focused comparisons of model development activities between NCAR and GFDL have strengthened each modeling effort. However, it could also be argued that such healthy competition could come from a single U.S. modeling effort in competition with inter- national efforts. 2  http://wcrp.ipsl.jussieu.fr/SF_RCD_CORDEX.html (accessed October 11, 2012). 3  http://www.narccap.ucar.edu/index.html (accessed October 11, 2012). 226

OCR for page 223
Strategies for Optimizing U.S. Institutional Arrangements The current arrangement has produced somewhat stable funding that is concentrated along existing agency lines. Long-lead-time research activities need such stability, al- though within this arrangement there can be short-term swings in funding that have negative long-term consequences. For example, short-term budget reductions can lead to reductions in the hiring of postdocs or young scientists; these missed opportu- nities have negative consequences for many years to come. Finding 13.3: Some positive aspects of the current U.S. institutional arrange- ment for climate modeling are the general stability of the funding that sustains the various efforts, as well as the diversity of approaches to solve problems and healthy competition that follow from having multiple modeling activities. Weaknesses One of the primary weaknesses in U.S. climate modeling is that modeling efforts are subcritical in key areas. Increased model complexity and greater societal expectation and demand for climate information create pressure for expanded climate modeling capacity, while human resources within individual modeling groups have not ex- panded commensurately (Chapter 7). There are at least two reasons for this: • funding that, while substantial overall, is inadequate to support the number of major modeling efforts; and • inadequate career development rewards, especially for young scientists. Scientific and applications-driven demands for increasing realism and comprehensive- ness of climate models also require major modeling groups to seek access to con- stantly increasing computational capacity, which requires increasingly sophisticated software development to efficiently exploit (Chapter 10). This software development requires additional human resources that core modeling groups struggle to support. These are serious impediments to progress. At the national level, maintaining the cur- rent structure of several quasi-independent Earth system modeling efforts cuts into the resources available for each group, pacing progress and creating stress by requir- ing modeling groups to spread expertise thinly across a broad spectrum of topics. In the current structure, computational resources for U.S. climate modeling are largely aligned along agency structures. This arrangement has some advantages in terms of stability, with multiple computing platforms providing some level of overall reliability to the availability of U.S. climate computing. Even if one agency’s computing platforms were cut, there would remain other platforms available for U.S. climate computing. 227

OCR for page 223
A N AT I O N A L S T R AT E G Y F O R A D VA N C I N G C L I M AT E M O D E L I N G However, this fragmentation invites duplication of effort and suboptimal alignment of national climate modeling priorities with computational resources. An additional weakness of the U.S. institutional structure is that modeling activities for long-term climate change are not well connected with the main U.S. operational center for weather and short-term climate prediction at NOAA’s NCEP. In some other countries, such as the United Kingdom, modeling activities for both short-term weather prediction and long-term climate are integrated within a single institution. In this arrangement the models used for weather prediction and climate projections share much of the same software infrastructure and physics, although the models used on the two time scales are not identical. As articulated in Chapter 11, there would be a significant potential for overall advancement in the United States if there were tighter integration between modeling activities across time scales. Two strategies would be (a) enhanced interactions between scientists that are developing and using models for long-term climate change, intraseasonal to decadal climate change, and for weather prediction and (b) development of a single unified modeling system for prediction on all time scales. Finding 13.4: Some limitations of the current institutional structure are that most U.S. climate modeling centers are individually subcritical with respect to exper- tise and funding, it is difficult to attract talented young scientists into model development, and the separation of operational and research modeling efforts can be a barrier to advances. THE WAY FORWARD A national strategy for advancing U.S. climate modeling should optimize or modify existing structures while adding critical new ingredients, as supported by the lessons learned from previous reports on U.S. climate modeling (Chapter 2). The committee believes it is productive to focus on actions that develop a greater level of unification by combining high-level cross-cutting leadership with science-motivated grassroots efforts. Several key aspects of a national strategy that contributes to this focus are de- scribed below. This discussion applies both to core modeling efforts for global climate and to regional climate modeling activities. Regular National Climate Modeling Forum In a distributed modeling system, the various model development and applications or user groups need mechanisms to communicate progress, share results, and discuss 228

OCR for page 223
Strategies for Optimizing U.S. Institutional Arrangements and plan common strategies for effective collaboration. Modelers can learn about each others’ progress at conferences and through scholarly journals, but for a diverse and decentralized community, this can be slow, haphazard, and inefficient. For this pur- pose, the committee recommends the establishment of an annual “U.S. Climate Model- ing Forum,” in which scientists engaged in both global and regional climate model development and analysis from across the United States, as well as interested users, would gather to focus on timely and important cross-cutting issues related to U.S. cli- mate modeling. While NCAR hosts an annual Community Earth System Model (CESM) meeting that is widely attended, it is largely focused on the needs of its own global modeling activities and would not be ideal for the broader purposes the committee envisions. The proposed National Climate Modeling Forums would provide regular interactions between scientists from the various U.S. regional and global modeling activities, including operational modeling. The Forum should also include end users of climate model output. The committee recognizes that one meeting may not be able to meet all the goals that are set forth below, and there will need to be experimentation about how to design a Forum that is most effective as a community-building institu- tion for climate modeling and its applications. The proposed Forum would, at a minimum, provide a periodic synthesis of current U.S. climate modeling capabilities and an opportunity for community discussion of near- term plans. It would also provide a venue for wide-ranging communication across a spectrum of climate model developers and users of climate model information. In the spirit of favoring a science-motivated grassroots approach, the Forum would provide the opportunity for the community to work together in ways that make sense at the scientific level, but which are sometimes difficult to anticipate in detail or to prescribe in advance. The Forum would • serve as an important mechanism for informing the community of the current and planned activities at core modeling centers and regional modeling efforts; • provide an important venue for fostering interactions among scientists in the core modeling efforts, regional modeling efforts, and other institutions includ- ing universities; • facilitate a more coordinated approach to global and regional model devel- opment and use in the United States; this approach would likely include the design of common experiments using multiple models that seek to improve our understanding and representation of key climate processes, and sharing the results and analyses of such experiments, as well as the formation of joint development teams to focus on addressing limiting biases or shortcomings in the current generation of models in the spirit of the current U.S. CPT approach, funded through multiagency competitively awarded grants; 229

OCR for page 223
A N AT I O N A L S T R AT E G Y F O R A D VA N C I N G C L I M AT E M O D E L I N G • provide an important vehicle to enhance and accelerate communication among climate modeling groups at research and operational modeling cen- ters, especially regarding the status and requirements of operational models and potential collaboration; • offer an opportunity to facilitate the development and implementation of a shared national software infrastructure through sustained, regular interac- tions between the infrastructure software developers and model developers and users, as well as by providing demonstrations of the benefits of such an approach; • offer a vital opportunity for end users of climate model information to both learn about the strengths and limitations of models, and to provide input to modelers on the critical needs of end users that could feed back onto the model development and application process; these exchanges could include offering short update courses that would satisfy a continuing education re- quirement for “climate modeling interpreters;” and • provide an opportunity for regular broad-based discussion of strategic priori- ties for the national climate modeling enterprise. Although current institutions may be subcritical in many areas, frequent interactions addressing the needs of all U.S. models with attractive and varying thematic foci would help to gather a critical mass of scientists across the United States to attack key problems in a coordinated fashion, and tighten the exchanges between global and regional modeling efforts. These interactions would include in-depth communications on activities, progress, and plans of the major research and operational centers and promote the advancement of specific aspects of climate modeling across the United States. The Forum would be a particularly appropriate venue for discussing and planning more systematic comparisons and evaluations of regional climate models using standardized metrics, and for model development projects (e.g., scale-aware param- eterizations) that try to bridge between the scales of regional and global models. It would also be an opportunity to broadly discuss the evaluation and communication of model uncertainty. Because this activity involves coordination across multiple modeling groups and agencies, it would be most likely to succeed if it were organized through a strong coordinating institution. While other organizations such as the American Meteorologi- cal Society, the American Geophysical Union, or the World Climate Research Program could in theory serve this role, the USGCRP might be a natural choice for taking the lead in organizing the Forum and associated activities given its mission to coordinate 230

OCR for page 223
Strategies for Optimizing U.S. Institutional Arrangements climate research activities in the United States. The USGCRP has stated in its strategic plan that “the global change research community as a whole would benefit from an increased and more systematic dialogue” and that “USGCRP will play an important role in facilitating this dialogue” (USGCRP, 2012). Meeting overload is always a concern and that makes it important that the Forum be seen as exciting and attractive. However, it is not expected that every modeler be at the proposed Forum. Instead the emphasis would be on transferring information be- tween modeling communities and interacting with user communities. Representatives of each major modeling group should attend all the meetings, and many more model- ers should be encouraged to attend through their interest in discussion of intercom- parison projects and various changing themes. One potentially unique attraction of this meeting would be users giving more substantial talks about their experiences and issues with using climate model output and how closely existing simulations meet their needs. This thread could lead to the Forum being a nexus for modelers to interact with the National Climate Assessment, depending on how that evolves. Common Software Infrastructure Chapter 10 advocated that a national computing and data infrastructure be a major component of a national strategy for climate modeling; here we discuss some of its institutional benefits and challenges. One of the weaknesses identified in the current U.S. structure is that efforts can be subcritical. The distributed U.S. modeling system has some some tendency for multiple institutions to develop modeling capabilities that partly duplicate efforts at other centers. A common software infrastructure can increase returns from existing structures across the U.S. modeling institutions. One goal of such a structure would be to allow the easy exchange and adoption of model- ing components. For example, if certain model components are viewed as “relatively mature,” or if there is one facility that is acknowledged as premier in developing some component (e.g., sea ice), those could become the de facto standard in the U.S. mod- eling community. This designation could effectively liberate resources at the other centers to focus on their strengths and address other critical topics, such as simulation of cloud feedbacks, which might benefit more from a diversity of approaches. The adoption of common software infrastructures has been advocated previously (e.g., Dickinson et al., 2002), and individual modeling centers have since internally adopted such infrastructures to allow a variety of configurations of their modeling system for different applications (see discussion in Chapter 10, including Box 10.2 on ESMF, an infrastructure that was intended for community use by multiple model- 231

OCR for page 223
A N AT I O N A L S T R AT E G Y F O R A D VA N C I N G C L I M AT E M O D E L I N G ing groups). In the process, much has been learned about how best to do this (see also Chapter 2), and it is now worth investing in the adoption of a common approach across all U.S. modeling centers over the next 5-10 years. The committee anticipates that the proposed annual Forum could play a key role as a venue for working strategic discussions on how to make this happen. To make a common national software infrastructure a reality, there need to be compel- ling incentives and benefits for all modeling centers to adopt a common approach, beyond facilitation of collaboration and code exchange. As noted in Chapter 10, the committee believes that cross-laboratory intercomparison experiments are a crucial part of the path forward to advancing U.S. climate models and a common national software infrastructure has the potential to facilitate in-depth comparison between models, including interchanging individual model components. Other compelling reasons for evolution to a common software infrastructure include the move toward fundamentally new computer architectures that will need to be adapted to; another could be enhanced opportunities to exploit high-end computing capabilities facili- tated by this approach; and a third could be to facilitate data standards that allow users to easily analyze results from different models with a common set of visualiza- tion and analysis tools. Decisive cross-agency endorsement of this approach will be needed to allow the climate modeling and software engineering community to collectively design and test the infrastructure and to provide the resources to transi- tion current major models to it. The adoption of such an infrastructure will facilitate interactions among scientists engaged in the full hierarchy of U.S. modeling efforts, thereby leading to their greater unification and coordination and allowing the climate model enterprise to better serve national needs and advance more efficiently. It is important that this infrastructure should entrain major regional modeling efforts as well as global climate modeling centers, and be adaptable to both research-ori- ented and operational modeling, to facilitate cross-fertilization between these model types and their developer and user communities. Computational Capabilities for Climate Modeling As described in Chapter 10, in order to meet the climate data and information needs of decision makers and users, U.S. climate models will need substantially increased computing capacity in the coming 10-20 years. This capacity will be distributed over a range of models and applications, ranging from pilot simulations for model devel- opment to large ensembles of lower-resolution simulations to extremely long pa- leoclimate simulations to decadal global and regional simulations at the finest grid 232

OCR for page 223
Strategies for Optimizing U.S. Institutional Arrangements resolution feasible. Storage and usability of large model data sets will also be key con- siderations. As discussed below, the committee recommends a two-pronged approach that involves the continued use and upgrading of dedicated computing resources at the existing modeling centers, complemented by an intensive research program on efficient implementation of high-resolution climate models on architectures requir- ing extreme concurrency (as also called out in Recommendation 10.2). This section also discusses possible pros and cons of a more radical step—establishment of a new national climate-specific computing facility of higher performance that any current U.S. climate modeling institution can afford to maintain. Existing climate modeling centers typically use computing resources that are largely dedicated to their institution. These resources are a crucial underpinning of the de- velopment and use of climate models, because they provide the required degree of flexibility to support fast-turnaround model testing and innovative and risky model development activities, while providing the computing capabilities for institution or agency specific goals (such as simulations in support of assessments). This approach has proven extremely useful in the past, and this mode of operation and support needs to be maintained. These largely dedicated facilities must be maintained and refreshed on an ongoing basis. They represent a substantial national investment. For instance, the Committee estimates that maintaining a computing system of the class of Gaea (dedicated almost exclusively to GFDL climate modeling), or NCAR’s Yellow- stone system (for which climate modeling is one major priority), is in excess of $30 mil- lion per year, including purchase, maintenance, power, human support, and assuming a 3-year replacement time scale. However, as noted previously, this arrangement of dedicated climate computing assets does not currently provide the critical mass in computing for breakthrough, innova- tive modeling activities that require the largest possible computational capabilities. Examples of such activities include ultra-high-resolution climate model simulations for the study of regional climates and extremes, the use of eddy-resolving ocean models to study critical ocean issues such as the oceanic uptake of heat and carbon and their feedback on the climate system, and global cloud-resolving modeling to better un- derstand the interaction of atmospheric convection and climate. The machines associ- ated with individual institutions are well suited for their more targeted goals, but not necessarily for such breakthrough calculations. For climate models such as CESM, the most computationally intensive simulations are being performed on the largest super- computing systems (e.g., as maintained by DOE) that serve a much broader scientific community than climate modeling. This strategy is attractive because it leverages costly external national resources and allows the climate modeling community to experiment with a wider class of computer architectures than it could internally afford 233

OCR for page 223
A N AT I O N A L S T R AT E G Y F O R A D VA N C I N G C L I M AT E M O D E L I N G to maintain. However, access to these external systems can be unreliable, and they often have operating protocols that are not suited to the very long simulations often needed for climate models. In addition, the external centers often have very different priorities for allocating resources to particular proposed models and simulations than just their importance for furthering climate science. Despite its obvious drawbacks, this is a “resource of opportunity” that the climate modeling community should con- tinue to exploit for extreme-scale computing challenges. To effectively use both forthcoming climate-dedicated computers and more experi- mental systems of opportunity, the climate community needs to aggressively invest in research into how to design models that achieve maximum performance from such systems (HECRTF, 2004). This problem is not unique to climate science, but the complexity of climate model codes exacerbates this issue considerably, as noted in Chapter 7. The design challenge is complicated by the diverse landscape of possible architectures, but the basic issue is architecture independent—achieving much higher concurrency in climate model codes than is now realizable through code refactoring, compiler tools, new algorithms, etc. This investment leverages off the proposed na- tional software infrastructure, which would facilitate the transfer of software tools and methodologies developed using one model across to other climate models, allowing the community as a whole to navigate hardware transitions more nimbly. Should the United States Invest in a National Climate Computing Facility? The committee debated whether the current combination of institution-specific computing and use of external computer resources of opportunity was the best national strategy for climate computing. In particular, we envisioned a national facility dedicated to climate supercomputing (which we will refer to as the National Climate Computing Facility or NCCF) to enable Grand Challenge calculations that have the potential to provide breakthrough scientific results through simulations at spatial resolutions and/or with representations of processes not previously possible. An NCCF is not intended as a new U.S. climate modeling center; rather, it is envisioned to be a central cutting-edge climate computational resource for pioneering calculations that benefit the entire U.S. climate modeling community and explore the next generation of climate modeling capabilities. In this section, we list some advantages and disad- vantages of this approach. Achieving a large positive impact on climate modeling would require a substantial ad- ditional national investment in climate computing of $100 million per year or more, in addition to the resources needed to follow the committee’s other recommendations on software infrastructure and research into optimization of climate codes for ex- 234

OCR for page 223
Strategies for Optimizing U.S. Institutional Arrangements treme-scale computer architectures. Given the current pressures on human resources for model development, on making model output useful to a broad applications community, and on maintaining an adequate climate observing system, a consensus community-based process would be needed for weighing large additional invest- ments in computing against further investments in these other key links of the climate modeling enterprise. An NCCF must complement institutionally specific computational resources, not replace them. The NCCF would focus on the execution of cutting-edge models that are primarily developed at existing U.S. centers, but on problems exceeding their internal computational capabilities. Some types of simulations appropriate for an NCCF might include • the study of regional climate change and extreme weather events, including hurricanes, droughts, and floods, using atmospheric models with resolutions down to a few kilometers or less; • the study of the effects of small-scale processes in the ocean, including meso- scale eddies, on climate variability and change; • the study of biogeochemical cycles, including the carbon cycle and atmo- spheric chemical changes, at very high resolution to better represent ecosys- tem-scale effects and assess their future response to, and feedback on, climate change; • the study of projected changes in land-based ice sheets and their interaction with the ocean that will influence future sea-level change; and • the study of the interactions of ecosystems and climate change at very fine regional scales. These simulations might involve both global and regional modeling components. The cost of an NCCF would depend on its scope. To be transformational, it would have to offer a several-fold increase in the size or speed of computations that could be performed on institutional machines, and more useful, reliable, and stable access than is likely to be provided by national computing resources not specific to climate modeling. As discussed in federal plans for high-end computing platforms (HECRTF, 2004) and borne out over the past decade, a single leadership system is expensive, and typically costs in excess of $100 million per year to procure and operate. Advantages of an NCCF If the U.S. climate modeling community had stable access to such a hardware platform, it would be easier to customize or codesign software infrastructure to maximize effi- 235

OCR for page 223
A N AT I O N A L S T R AT E G Y F O R A D VA N C I N G C L I M AT E M O D E L I N G ciency on that hardware. A single high-end facility would allow higher-resolution sim- ulations to happen sooner, and it might speed up the inclusion of more Earth system components and larger ensembles. A single facility would provide a focal point for advancing the computational performance of U.S. climate models, which would have dividends for both scientific advances and the generation of climate information at the near-local scale that users desire. The dedication of such a facility solely to climate modeling might allow easier access to model output data and the development of data analysis tools for both model developers and model output users. The existence of such a single high-end facility could have significant advantages in economies of scale, such that it could be significantly more cost-effective to procure this additional computing resource through a single site rather than in a distributed fashion. An NCCF would leverage the investment in software infrastructure that has also been advocated in this report. The infrastructure would facilitate the efficient execution of models on the NCCF that were previously developed on different architectures at the various U.S. centers. Further, the existence of this high-end facility would provide in- centive for individual modeling institutions to adopt the same software infrastructure. Risks of an NCCF A dedicated leadership-class climate computer facility would entail large additional expense and potentially risky choices about architecture and management. In an en- vironment of constrained budgets, an NCCF would compete with institutional centers for computer resources and personnel, further fragmenting the climate modeling community into subcritical units. It might also be vulnerable to year-to-year budgetary instability. An NCCF would have to make choices about computer architecture that might place additional risks on the climate modeling community, associated with “pioneering” the use of untested computer architecture, programming environments, and performance optimization. These costs would be decreased by using better-tested architectures, but that might also reduce the potential payoff in transformational capabilities. The management of an NCCF so as to complement the capabilities of other insti- tutional and external computing resources would be an important challenge. Clear community-governed mechanisms would need to be set up to select the models and problems on which the facility focused. There would need to be close communication and feedback between the computational scientists involved with the operations of the facility and the climate scientists guiding the overall mission. Ultimately, the scien- 236

OCR for page 223
Strategies for Optimizing U.S. Institutional Arrangements tific objectives and imperatives of the overall U.S. climate modeling enterprise would need to drive the operational details of any such facility. Overall, an NCCF would be most attractive and least risky in an environment of sus- tained budget growth for climate science and modeling, which would allow it to be pursued in parallel with the other critical investments in climate modeling recom- mended in this report. Why Not a Single U.S. Climate Modeling Center? The approaches outlined throughout this report build on the current distributed system for U.S. climate modeling. They attempt to overcome the obstacles associated with a distributed system through frequent communication at U.S. modeling forums and the adoption of a common software infrastructure to support interlinked model development, execution, and analysis. We discussed a National Climate Computing Facility as a possible way to accelerate research into computational frontiers of climate science. Given this approach, a logical question to ask is: Why not simply move toward a single U.S. modeling center that could achieve these benefits under a “single roof,” replacing all the current climate modeling centers? The committee believes such a move is undesirable at this time for several reasons: • Current modeling institutions have a variety of missions supporting the needs of their sponsoring agencies, including operational prediction and data as- similation. It would be difficult to carry out those differing missions in a single, monolithic new institution without sacrificing the necessary focus. • There is a recognized benefit to fostering multiple approaches to address critical topics. The downside of this approach is the potential for duplication of efforts, although the other efforts recommended in this report should reduce such duplication, e.g., the efforts to foster communication and the use of com- mon infrastructure. • It could be hugely disruptive, at least in the near term. Unless there were an extraordinary and sustained national interagency commitment to the process, the new center would not supplant the current centers, and further dilution of effort and resources might ensue. The committee believes that a more distributed strategy embraces the philosophy of maintaining scientific diversity where appropriate while maximizing computational resource efficiency. This efficiency comes through the evolution to a common infra- structure, and the existence of a distributed computational capability including both 237

OCR for page 223
A N AT I O N A L S T R AT E G Y F O R A D VA N C I N G C L I M AT E M O D E L I N G institutionally dedicated resources and the NCCF. The hierarchy of models needed for climate modeling (discussed in Chapter 3) is mirrored by the hierarchy of computa- tional capabilities necessary to take full advantage of those models. Finding 13.5: The committee believes that the potential benefits of a move to a single U.S. climate modeling center are currently outweighed by the risks. Although it is difficult to objectively assess how many modeling efforts are now optimal in the United States, it is likely that adoption of the strategies recommended by the committee could make U.S. climate modeling efforts more integrated and transparent. These actions should lead to convergence among some modeling com- ponents that are most mature, while maintaining diversity and competitive innovation among those key components that have the greatest scientific uncertainty. With U.S. climate modeling efforts more tightly integrated, different centers may begin to col- laborate by specializing on different aspects of the climate modeling problem, acting as a distributed network that ultimately is stronger and more robust than an individ- ual climate modeling center could be. Recommendation 13.1: To promote communication and collaboration across the climate modeling enterprise, annual U.S. climate modeling forums should be organized to bring together scientists from the global and regional modeling ef- forts across the United States, scientists from other institutions that are involved in model development and analysis, and model users. Recommendation 13.2: Model intercomparison activities are key to advancing climate models, and one activity at the climate modeling forum should be dis- cussion and planning of carefully designed suites of simulations to compare the behavior of U.S. climate models with each other and with observational bench- marks. Regional climate models are a particularly pressing focus for this activity. Such simulations could take advantage of a shared software infrastructure to facilitate comparisons, including on a component basis. Recommendation 13.3: In order to advance climate modeling in the United States in the next 10-20 years, the United States should invest in initiatives that enable the climate modeling community to exploit extreme-scale computing capabilities through the development of new and common software architec- tures that can be shared across modeling centers and thus spur a national effort to push the computational frontiers of climate science. 238