1
Workshop Context
At the dawning of the age of numerical simulation in the 1960s, few imagined the influence it would have on the atmospheric and oceanic sciences. Today, two generations later, numerical models of the atmosphere and the ocean are central to weather prediction, research, and education. The size and speed of computers and their computational and data-handling techniques have improved enormously; today we can do fast, inexpensive computations of a breathtaking range of fluid-mechanical phenomena.
Great strides have been made over the past few decades in our understanding of the atmosphere and ocean, in our modeling capabilities, and in numerical simulations of the atmosphere, ocean, and land surface. Yet current representations of unresolved processes in the models tend not to adequately represent our knowledge of the underlying physics. Moreover, there is evidence that further progress in numerical simulations is being impeded by the slow pace of improvement in the representation of certain key physical processes in the models. In the arena of weather forecasting, this is manifest in the discovery that there is somewhat greater divergence among the various different operational numerical prediction models than there is among ensemble members of the same model created with different initial conditions (Mylne et al., 2002). At the same time, climate modelers must deal with substantial disagreement among models in their prediction of such fundamental effects as cloud feedbacks.
There is an emerging perception that the physics of the growing family of geophysical flow models is not receiving the continuing attention needed to make these tools more useful and accurate. In some cases the models have stagnated, changing little since their inception. Some of the reasons for this stagnation lie in the underlying science. For example, despite decades of intensive efforts by several science communities, adequate models of geophysical turbulence still do not exist. But some of the problem seems cultural; today’s modelers and users of model output seem less engaged with improving model physics than with the increasingly sophisticated numerics, graphics, and architecture of the model system and with using models rather than observations to study geophysical flows. As education and training in the atmospheric and oceanic sciences turn away from model physics, students increasingly see model outputs as truth without the healthy skepticism that should be inherent with these tools.
No model is or ever will be perfect. The atmosphere and ocean are inherently nonlinear, and their chaotic physical processes occur over a vast range of scales, making it impossible to simulate every physical process at every scale. But because these models are often used to predict future events, which can have immediate to long-range policy implications, it is imperative that their underlying physical processes be represented as robustly as possible. To ensure success in this regard, it is vital that the science of geophysical modeling garners the attention and support necessary to continue its improvement.
The National Academies’ Board on Atmospheric Sciences and Climate organized its 2004 summer workshop to explore and evaluate current efforts to model physical processes of coupled atmosphere-land-ocean (A-L-O) models (see Appendix A for the complete Statement of Task). Specifically, the parameterization of physical processes in A-L-O models was addressed, including associated errors, testing, and efforts to improve the use of parameterizations. During the workshop discussions, participants examined some intellectual and scientific challenges in modeling and highlighted the proposition that some of the key impediments to progress in representing physical processes are primarily cultural in nature. For reasons that may broadly have to do with the incentives and disincentives that exist in certain parts of the atmospheric and oceanic sciences, scientists in the field may be slipping into a mode of conduct in which the arduous and often unrewarding task of developing and rigorously testing new parameterizations is avoided in favor of
tuning existing schemeswhich can result in compensating errorsor developing new schemes without subjecting them to evaluations of their effects on the weather or climate system or without rigorous tests against observations. The advent of multi-model ensembles and stochastic parameterizations, although no doubt contributing to improved forecasts and better quantification of uncertainty, may have the unintended effect of reducing incentives to improve model physics. In the field of education, the increasing ease with which large models may be obtained and run tempts students away from basic understanding and toward mere simulation; likewise, the long gestation period for instrument development and field project design and execution can discourage students and young researchers from these important endeavors.
The dichotomy of simulation versus understanding was a theme that connected many areas discussed at the workshop, beginning with the opening talk on this subject by Isaac Held of the National Oceanic and Atmospheric Administration’s (NOAA) Geophysical Fluid Dynamics Laboratory and Princeton University (see Appendix C for a paper by Held on this topic). As discussed during the workshop, meteorology, in particular, has benefited financially from the widespread perception that it is an applied science geared mostly toward weather forecasting and, more recently, climate prediction. Much research in the field is supported by funding agencies on the premise that it will improve forecasts, and the media often project an image of researchers beavering away at better forecasts. But the same perception that is successful in helping policy makers see the value of research in atmospheres and oceans may turn away talented young scientists who, steeped in the culture of physics and mathematics and driven primarily by curiosity, are best suited to tackling the difficult, often fundamental, scientific problems that our field faces.
The cultural issues come to a head in the field of climate research and prediction. As summarized in the concluding presentation by Dennis Hartmann of the University of Washington (see Appendix D for the workshop agenda), the culture of climate modeling has emphasized model intercomparison and the testing of model output against global observations. As Dr. Hartmann discussed, physical parameterizations are often viewed as blackbox subcomponents whose knobs, in the form of largely unobservable parameters, can be adjusted at will to obtain some desired result. Physical parameterizations, often with large numbers of unconstrained or loosely constrained parameters, are inserted into
models and judged largely on the merits of their perceived sophistication and their effect on model performance. Ideally, the development and testing of parameterizations would include (1) evaluations of the effect of the physical process on the weather and climate system to assess its sensitivity and demonstrate the importance of the process, (2) comparisons of the parameterized process against observations (or high-resolution simulations of it), and (3) evaluations of the effects of the parameterization on the structure and evolution of the larger-scale flow. Rigorous offline tests of parameterizations against observations often are lacking, and some researchers believe that such tests are flawed because they omit the feedbacks that the process in question would interact with as part of a large model. The lack of rigor in the development and offline testing of some key physical parameterizations was viewed by some workshop participants as being due to a combination of two influences: (1) the isolation of climate modelers from observationalists and from those involved in numerical weather prediction and (2) the pressure on climate science to produce useful predictions of global warming, which is reducing to subcritical levels the skilled man-hours dedicated to developing and implementing improved physical parameterizations. A-L-O models are a mixture of scientific exploration and application, and it is important to protect the health of both activities without letting the latter eclipse the former.
The workshop during which these and a myriad of other issues were examined was a menagerie of discussions organized around six topic areas capped by an afternoon of synthesizing discussion (see Appendix D). Because of the interactive nature of the discussions, no attempts were made to attribute specific ideas to particular people. By the National Academies’ policy for this type of workshop, there are no findings or recommendations in this report. Instead it summarizes the major discussion items that arose during the workshop; it is not an edited workshop transcript.
The workshop also sought to identify a limited number of key physical processes whose representation in models is regarded as problematic. Chapter 2 reviews the discussion of some parameterizations that are in need of improvement. Chapter 3 explores how parameterizations are developed and tested. Chapter 4 summarizes the workshop participants’ thoughts on cultural impediments to improving model parameterizations and ideas for alleviating these problems and improving simula-
