August 10, 1998

Dr. Aristedes Patrinos

Department of Energy

Associate Director

Biological and Environmental Research ER-70

19901 Germantown Road

Germantown, MD 20874-1290

Dear Ari:

This letter is in response to your request that JASON review plans for the Advanced Climate Prediction Initiative. Specifically, your request was that

JASON undertake a study to determine whether the pace of computational technology development (computers, networks, software, etc) will be sufficient to perform century long simulations of climate change with weather-scale resolution. Additionally, JASON should provide advice on how best to utilize this technology to link the scientific resources of the DOE laboratories and the US Global Change Research Program to produce unprecedented climate change projections, including measures of climate variability, by the middle of the next decade when greenhouse gas stabilization policies are enacted.
To respond to this request, we and the USGCRP jointly convened a workshop at SIO from July 1-3, 1998. Speakers included representatives from the climate modeling community, the high-performance computing community (including the ASCI program), and several users of climate models. Beyond this input, we also reviewed extensive written material and drew upon our own prior experience with climate and high-performance computing, as well as the management of large-scale technical enterprises.

The "executive summary" of our response is captured by the following two observations:

    1. Substantial increases in the computational power available to US researchers are well-warranted and can contribute to a better understanding of the climate system.



    3. Computational power alone will not greatly improve our abilities to predict climate. Linked observational programs and process studies are essential for a balanced global change effort.

We elaborate on these points in turn.

The rationale for increased computational power

Understanding the earth?s climate is one of today?s most challenging scientific problems. Multiple components are involved, including the atmosphere, the ocean, the cryosphere, the land surface, and the biosphere. These vary and interact with one another on diverse spatial and temporal scales.

Models of the climate embody and test our understanding of this system. They also allow us to assimilate observational data ("fill in the gaps"). Ideally, they allow us to forecast changes in the climate system (both due to natural variability and to anthropogenic response) and the likely effects of various mitigation strategies.

Fundamental physical considerations set the spatial and temporal resolutions needed for a realistic climate model. These, together with the necessity for century-to-millennium scale runs to quantify the response to anthropogenic forcing or the natural variability imply, at a minimum, terascale computing resources. ACPI is a step toward meeting that need.

Successful climate modeling will find utility in a number of different modes. On short timescales, one may expect some improvement in numerical weather prediction (NWP), although this is quite a distinct problem from climate (because of the very different timescales involved) and most of the improvement is expected from better observational data. Over somewhat longer intervals, seasonal to interannual climate prediction will likely be a real benefit. [The improved agricultural response of Peru to El Niño forecasts during the last decade clearly demonstrates the benefits possible.] Accurate descriptions on decade-to-century scales are necessary to evaluate the likely impacts of anthropogenic activities and the likely efficacy of various responses. And simulations involving many millennia are required to describe paleoclimates. These diverse needs imply the need for flexible simulation capability that can be used in new and imaginative ways.

Many other nations have realized the importance of sustained and focused investment in climate modeling, the European Center for Medium-range Weather Forecasting (ECMWF) perhaps being the outstanding example. These organizations benefit from sustained funding, state-of-the-art technology, and strong intellectual leadership. Surprisingly, although the US has led world in climate observations and in pursuing the science embodied in the models, it lags significantly in the exploitation of modeling capabilities. Greater computing power will help to address this deficiency. Among other benefits, it will enable objective testing of simulations with diverse assumptions and process descriptions.

Organization, management, and implementation of the ACPI

In considering how additional computing power might best be applied to climate modeling, it is important to realize that the science is as yet uncertain, and controversial in parts. Further, any successful modeling effort will need to closely integrate diverse disciplines, including computer and computational science, applied mathematics, fluid dynamics, chemistry, weather, biology, and climatology. There is also the obvious need to harness the creative energy and talents of the university community.

The basic question in organizing the ACPI is whether to simply supply more computational capability to the diverse US modeling community, or to create a more centralized structure for a coherent attack on climate simulation. Our inclination is toward the latter. Strong scientific leadership must be identified early in the project and charged with managing the computational and scientific resources. Science must be at the center of the enterprise, with technology as a means to an end, not an end in itself.

However ACPI is implemented, we urge a close coupling with NWP. The necessity of continually confronting detailed observations can only be a bracing experience for the modeling community, as has been shown by the ECMWF experience. A USCMWF or organization similar to NCAR but focused on climate are notions that capture many attractive organizational aspects.

We believe that the case is strong for the ACPI proceeding expeditiously. It will address serious deficiencies in climate computing and will allow climate modelers to efficiently refine and exploit their models. However, as the benefits of terascale climate computing are some years off, there are some short term steps that could be considered, including a reallocation of existing computing resources or the purchase of a 100-1000 Gflop capacity.

ACPI is only a part of the Global Change program

As noted above, ACPI in and of itself will not greatly improve our abilities to predict climate. Success will require that modeling be integrated with observational programs and process studies. Indeed, it is essential that all three of these elements work together. Such coordination is rare within the global change research program, the DOE?s on-going ARM program being a conspicuous exception.

It is encouraging that ACPI includes a substantial outreach component to make modeling results more widely available. However, even before ACPI begins, it is important to educate the public and policy makers about the nature of climate science and the real scientific uncertainties and deficiencies the field is trying to address (in part through efforts like ACPI). Among these are:

I hope that these thoughts will be of use to you and others as you plan and implement the ACPI. Please don?t hesitate to contact me if you would like any clarification or amplification of these points.


Dr. Steven E. Koonin

Study Contributors:

H. Abarbanel, K. Case, F. Dyson, S. Flatte, N. Fortson, E. Frieman, M. Gregg, M. Goldgerger, W. Happer, R. Jeanloz, W. Munk, W. Nierenberg, O. Rothaus, M. Ruderman, R. Schwitters, P. Weinberger, F. Zachariasen