Improving the Scientific Foundation for Atmosphere-Land-Ocean Simulations: Report of a Workshop (2005)

Several impediments to developing better representations of physical processes in models were identified by workshop participants. Most of these are described as “cultural,” meaning the impediments, to some degree, are entrenched in the customary way of doing research and education in the atmospheric and oceanic sciences and in the structures of universities, laboratories, and funding agencies as they have evolved through history. The impediments identified by the participants are summarized in this chapter. Subsequently, several ideas from the workshop participants for overcoming these impediments and for improving the progress of model development are presented.

First, and perhaps foremost, workshop participants expressed concern that the atmospheric and oceanic sciences do not attract the best and brightest graduate students. It was suggested that prospective graduate students tend to see meteorology as the strictly applied endeavor of making better weather and climate forecasts, which they often perceive as the subjective interpretation of satellite images and the application of advanced computer graphics. Participants hypothesized that this perception probably arises because the field is projected to the general public primarily through media weather forecasts. Likewise, climate science likely is perceived largely as an effort to assess and predict global warming. There is a vague impression that computers are used more for substantive reasons, but the idea that there are problems of large inherent

intellectual interest, requiring backgrounds in advanced mathematics and physics—and that are equally as challenging as these fields in and of themselves—generally is lost. The field seems to be more successful in recruiting students attracted to the idea of weather forecasting or policy-oriented climate research than in recruiting students with a fundamental mathematical and physical curiosity about the atmosphere and ocean. This state of affairs may result in part from the way the atmospheric sciences and climate science promote research to funding agencies by promising improved forecasts.

Once enrolled at a university, the contemporary graduate student often finds himself or herself beholden to the goals of his or her professor’s research grant, goals that are increasingly burdened by the necessity of producing short-term deliverables or of demonstrating broader impacts on society. This system blurs the line between student and employee. The student with a bright idea or who is singularly motivated to pursue a particular problem, however important that problem may be, may not be able to fit his or her interests into the available funding slots, even if the faculty are sympathetic to the idea.

Workshop participants noted that another challenge of engaging new people in the field is that the development and refinement of model parameterizations is not seen as an attractive career path by most young researchers. In the United States, most large modeling centers maintain groups dedicated to developing and improving parameterizations, but these groups tend to be small and, consequently, overtaxed. Moreover, these positions often are funded as support scientist or software engineer positions.

The current funding environment is subtly influencing the choices that researchers make, in ways that may inhibit progress in understanding and simulating climate and weather. For example, the development of instrumentation and the execution of field measurement programs, which are crucial for progress in a number of areas discussed at the workshop, require relatively long time horizons. Increasing emphasis on deliverables in some funding agencies together with tenure cycles at universities acts to discourage such labor-intensive and time-consuming endeavors in favor of activities such as simulation experiments with existing numerical models, where deliverable results often can be obtained in a short time.

In many areas the development of improved representations of physical processes requires knowledge of a broad variety of physics. For

example, improved simulation of boundary layers and convection over land depends on accurate representation of land-surface processes coupled strongly with convection and precipitation and with clouds, which strongly affect radiation and therefore surface fluxes. But studies of these processes often are compartmentalized within laboratories, universities, and funding agencies. In the aforementioned example, soil properties are usually studied in departments of hydrology, whereas clouds and convection are studied in atmospheric science departments. Parameterization of the soil components cannot proceed in isolation from representation of boundary-layer and convective fluxes, or from parameterization of clouds, yet the structure of research organizations discourages the cross-fertilization that is required to make progress. Compartmentalization also compromises the efficiency of observational programs. Workshop participants cited several examples in which expensive field measurement programs were conducted at the instigation of one group of scientists, but certain key measurements that could have helped another group were not made, even though the added cost would have been marginal.

Another cultural impediment to scientific progress that was discussed by participants is the reluctance of different federal agencies, or even different divisions of the same laboratory, to cooperate for the greater good. Although endorsements of such cooperation are easy to elicit, experience shows that actual cooperation is rare. The problem likely stems from competition for resources, fear of loss of control, and/or diminished credit for the results.

Workshop participants generated several ideas for addressing education-based impediments to progress in model development. Although addressed with enthusiasm and lively debate during the workshop, these are presented here as ideas rather than recommendations. As such, they are not thoroughly detailed, so they would need to be explored further should an institution wish to implement them. These ideas include agency-sponsored fellowships for students, instruction in model development, summer schools in which people with varying areas of expertise are brought to bear on a shared problem, and the need to better convey the intellectual excitement of the atmospheric sciences.

As noted, the current funding of graduate students through research grants produces a set of incentives and disincentives that do not always work in the best interests of scientific advance. An alternative approach would be to fund students with fellowships granted directly by the agencies. This already is done in a few cases, but it could be implemented more broadly. A wider use of fellowships in place of assistantships would empower students; that is, it would increase the students’ freedom to choose their schools and advisors and to base their thesis work on research projects of their own conception. It also might serve to ease pressure on proposal writing, management, and review as well as mitigate the disincentives that discourage conducting research with long time horizons.

Education naturally proceeds from the simple, basic, and well known to the elaborate, subtle, and uncertain, especially in the sciences. At present, university curricula in atmospheres and oceans do a good job of teaching about computational mathematics, geophysical fluid dynamical theory, and the observational record, but they do not do well enough in preparing students for postgraduate research involving the practices of complex computational simulation modeling and modern measurement techniques. Although in most universities the time constraints on expanding the curriculum are severe, this deficiency potentially could be ameliorated by such things as shifts in course contents (e.g., perhaps doing more in the style of case studies), expansion of apprentice research opportunities with professional modelers and instrumentalists, and the creation of inter-institution, special-topic programs for motivated students. Regarding the latter, interaction with multiple institutions (e.g., universities, research laboratories, private companies) may be prudent if a student wishes to conduct research that, for example, requires a class not taught at their university or requires collaboration with an instrumentalist. In other words, rather than adhere to the usual departmental requirements, such motivated students may design a special, unique program with his or her advisor combined with scientists from other institutions. Such special programs are best suited to the student with a

firm, fundamental knowledge base; thus, they likely are best restricted to someone at the Ph.D. or postdoctoral level.

In a multidisciplinary field such as climate, it is rare to find any one university that can provide students with the necessary training in all the important subjects, which range from fluid mechanics to radiation to remote sensing to chemistry. Summer schools run by universities or national centers are an effective way to supplement the education of graduate students, postdocs, and young scientists. They bring the workers on the forefront together, promote interactions, and give students hands-on experience in cutting-edge research. Notable examples of successful summer schools include those frequently conducted by the Woods Hole Oceanographic Institution and Cambridge University. More of these kinds of opportunities could potentially better educate and inspire the younger generation.

As mentioned above, there was a concern among many of the workshop participants that the sciences of the atmosphere and ocean suffer from misperceptions about their nature, and this affects the kinds of students attracted to the work. Several ideas for combating this were discussed at the workshop. At the heart of all these ideas is the notion that it is important to convey to prospective students the sense of intellectual excitement that pervades the field, in contrast to the image of mundane, subjective work portrayed by the media and others. One idea for accomplishing this might be to hold a series of summer schools aimed at undergraduate students in fields other than atmospheric science and oceanography. These might complement the graduate summer schools proposed above. Another idea is to encourage experienced professionals in our field to communicate to a broader audience by publishing in broadly-based scientific journals (e.g., Scientific American, Discover) and by publishing books aimed at the general, scientifically literate reader. The field also could benefit from an eloquent and prolific

spokesperson comparable to Richard Dawkins (Dawkins, 1990), who speaks for evolutionary biology.

As models age, they become refined, but they also lose their flexibility. Thus, workshop participants discussed the notion that continual renewal of large modeling efforts improves their effectiveness. Although designing and building a new model from scratch can be an incredibly large and daunting task, it can also serve as an invigorating renewal. Fresh-start modeling projects represent opportunities to implement the most current and novel ideas and simultaneously to involve young, fresh scientists in the model-building enterprise. An excellent example is the European Centre for Medium-Range Weather Forecasts (ECMWF) operational weather prediction model, the first version of which was created in the late 1970s by a group of young, talented, hard-working scientists. ECMWF brought forth many important modeling advances, and their success is due, in part, to the fact that the first ECMWF model was developed from scratch, making use of the strongest ideas available at the time. Today, 25 years later, the ECMWF model is evolving more slowly, but it continues to be one of the world’s best. ECMWF’s continuing success is favored by their frequent and institutionalized interactions with other scientists from around the world, through seminars, workshops, and extended visits. Moreover, with its practice of hiring young scientists for five-year contracts, ECMWF serves as a good example of making the development and improvement of parameterizations an appealing career choice. Workshop participants thought that the modeling enterprise in the United States might be improved if its modeling centers similarly increased incentives and rewards and elevated the intellectual excitement for such work.

The concept of the aforementioned Climate Process and Modeling Teams (CPTs)a collaboration with responsibility for investigating a particular physical process (e.g., cumulus convection) in a comprehensive way from theory, measurements, and fine-scale computational simulations to the design, implementation, and evaluation of its parameterization in a complex system modelseems a very attractive approach to closing the gap between fundamental science and Earth-system simu-

lation. During the experimental phase, the three pilot CPTs have been constrained not to propose new measurements. If these experiments work well, then expanding the scope of the CPTs could be beneficial in addressing additional processes and carrying out new field measurement programs in which this comprehensive responsibility is inherent in their conception and design.

Workshop participants also discussed that, as part of the process of continually improving models, efforts need to be made to remove physical representations that have been shown to be inferior or defective. It is important that the justified desire for multi-model ensembles not result in the retention of parameterizations that are known to be deficient.

A premise of Earth-system simulation models is that dynamical control inheres in the planetary and synoptic scales with necessary, but essentially simple, parameterized representation of the effects of processes on finer scales. Although this premise has worked well in practice for modern simulations of weather and climate, it must ultimately fail to be correct. As long as simulation models span a limited range of scales, the importance of cross-scale dynamical coupling cannot be investigated. Even the occasional modern use of an embedded finer-scale subdomain within a coarser large-domain grid (e.g., a nested-grid hurricane forecast) only partly encompasses the coupling possibilities as long as the information flow is only down-scale, from large to small. The more ambitious approach to cross-scale coupling is to routinely embed either more fundamentally based submodels in place of existing parameterization schemes. This new approach, called multi-scale modeling, involves embedding models that explicitly resolve the smaller scales as components of global circulation models. The embedded high-resolution models take the place of conventional macro-parameterizations. Prospective benefits include improved simulations of the global circulation, and mathematical convergence of the multi-scale model to a global high-resolution model, in the appropriate limit. Drawbacks include very large (but possibly feasible) computational requirements.

Given both the variety of possible feedbacks among component processes and subsystems and the spatial and temporal non-smoothness of atmospheric and oceanic fields near the grid cutoff scale of Earth-system simulation models, it is unlikely from the perspective of computational mathematics that the model solutions are very accurate with respect to the underlying partial differential equations (or even worse types, regarding smoothness) on which the simulation model is based. To the extent that potential solution non-smoothness is dealt with by artificial computational damping, there is further divergence from the true solution of the underlying equations. As a practical matter, present modeling methods are largely an empirically determined compromise between desired accuracy and smoothness of its solutions. Studies of simulation convergence with increasing grid resolution are rare and often ambiguous in their results. Evidence for non-modularity of complex simulation modelsin which swapping one component algorithm for another that in theory is comparably reliable and accurate leads to quite different model answerssuggests that they have a rough fitness landscape (see Appendix C) that is rarely considered in evaluating their answers (apart from the common practice of parameter tuning a given model’s answer to a more appealing outcome). It therefore seems likely that the choice of computational algorithms is often influential, if not determinative, in model-simulated behaviors. If so, then much more exploration of simulation sensitivity to computational methods would be useful to better understand the model solution’s validity.

Physical parameterizations and discretization methods are often regarded as very distinct aspects of a model’s formulation. This is a misconception. The implementation of a parameterization inevitably raises discretization issues, which can be quite challenging and are typically linked to the discretization methods used to represent the resolved-scale motions. Therefore, several workshop participants suggested it would be best if the physical and numerical aspects of a model’s design be considered as closely coupled problems. Furthermore, it may be that new approaches to the challenges of numerical integration could be valuable, such as using exact discrete representation to formulate parameterizations.

There are no fundamental differences in formulation between the global models used for climate simulation and those used for numerical weather prediction (NWP). Models used in operational NWP are evaluated daily for their ability to simulate the evolution of particular weather systems. The statistics of such forecasts have been analyzed to identify systematic errors that emerge in the first few days of simulation. In many cases these systematic forecast errors are closely related to the errors that the same models produce in simulations of the present climate. These considerations led the ECMWF (1999) and the World Climate Research Program (2001) to advocate the use of NWP as a means of evaluating climate models. The idea is to use climate models to produce weather forecasts for real cases, identify the forecast errors, and trace those errors back to weaknesses in the models’ formulations—in other words, to do what NWP centers do as a matter of course.

There are precedents for this. The United Kingdom’s Met Office has been using a single modeling system for both climate simulation and NWP for years now. In addition, the Max Planck Institute for Meteorology in Hamburg, Germany, developed its climate model by starting from a version of ECMWF’s NWP model. Until recently, however, the United States lagged behind in this area. This has now been corrected through a program called CAPT, the CCPP-ARM (Atmospheric Radiation Measurements) Parameterization Testbed, where CCPP stands for the Climate Change Prediction Program funded by the U.S. Department of Energy. CAPT is using analyses of global weather from NWP centers, in conjunction with field observations such as those provided by ARM, to evaluate parameterizations of subgrid-scale processes in global climate models (Williamson et al., 2005; Boyle et al., 2005). CAPT’s methods were first applied to the community atmosphere model (CAM) and are gradually being used with a wider variety of models, including the GFDL’s Atmospheric Model 2 (AM2). Workshop participants thought that much could be gained if these methods were widely adopted by the U.S. climate modeling community.