Group Leader:  Rob Armstrong

Objectives: To create regional data sets, including physical, chemical, and biological data, to be used for evaluating ecosystem models. The regions of interest initially include the U.S. JGOFS time-series sites (HOT and BATS) and process-study sites (NABE, EqPac; Arabian Sea; Southern Ocean), but could be expanded to other sites as well. The approach will be incremental, starting with the Equatorial Pacific and expanding as we gain more experience.

Background: There are two motivations for the creation of regional testbeds:

•  it is desirable to have a forum in which different models and modeling approaches can be compared to common data sets, so that the strengths and weaknesses of these approaches can be compared; and

• it is only by making such comparisons on a regional basis that we will be able to decide whether it is possible to create a single model structure, with parameter values that are generated regionally in response to environmental forcings, or whether an approach based on dividing the world a priori into biogeochemical provinces is the only way to handle the different characteristics of different regions.

This effort builds on efforts at model intercomparison initiated at the International JGOFS workshop in Toulouse, France, in November 1995. At that workshop, several ecosystem models were compared in relation to a common data set from the subarctic Pacific. The structure we envision for U.S. JGOFS Regional Testbeds differs fundamentally from this "workshop" approach in that the evaluation of models will be distributed in time and space.

Individual investigators will be able to access a common data base and common physical model, and will be able to tune their models to these data.  Comparisons between the performances of different models will initially be made on an informal basis among participants, though as the project progresses it may be desirable to create a more formal forum for model intercomparison, perhaps in conjunction with a yearly JGOFS SMP summer meeting. The informality of the process, and the ability of investigators to experiment with their models, should relieve the idea that this project is a shootout among existing models, the winner of which will be crowned an official JGOFS-sanctioned model.

Tasks:  For the Regional Testbeds project to be successful, the assembling of these regional testbeds must not fall as an onerous task on any individual or group of individuals. For this reason, our plan is to rely on individuals who have worked on coupled physical-biological models of various regions, asking them to share with the community the data sets and physical models they have used in developing their ecosystem models.

For the first regional testbed, on the Equatorial Pacific, Fei Chai, Mark Verschell, and Mike Roman have agreed to collaborate to design the testbed, both by assembling appropriate data sets and by establishing a common physical model (the plan is to have both 1D and 3D physical platforms).

They will provide information for two different longitudes, since a model that requires vastly different parameter values to perform well even within the same region is probably not a very good model. As we gain experience with the process, other regions will be added; Raleigh Hood, for example, expressed interest in developing an Arabian Sea testbed; we hope others will join the effort as it progresses.

Membership:
Rob Armstrong
Nick Bates
Adrian Burd
Fei Chai
Dick Dugdale
Raleigh Hood
George Jackson
Ed Kearns
Mike Roman
Mark Verschell

Group Leaders: Rob Armstrong

SUMMARY:

The concept of regional testbeds has been developed over the past three years at SMP workshops. This effort is of fundamental importance to the goal of predictive modeling of the ocean carbon cycle. On one side, there is the idea that, for modeling, the ocean must be divided into roughly 100 "biogeochemical provinces," each of which must be characterized in terms of its physics and biology. At the other extreme lie current biogeochemical simulations, starting with Fasham et al. (1993), in which a single model, representing a single ecosystem structure, is used everywhere on the globe. There is no consensus on which approach is better. From a modeling perspective, using a single ecosystem structure has an appealing economy and simplicity; but it has not been convincingly demonstrated that this approach can capture the diversity of the real world with sufficient accuracy for biogeochemical prediction. On the other hand, the need to estimate parameter values for tens or hundreds of regions, and for seasons within each region, would seem to be an almost impossible task, given the difficulties of estimating optimal parameter values for single ecosystems.

We add to this conceptual problem a second uncomfortable fact: most data-based models that have been developed for biogeochemical simulation have been developed for application in single regions. Extrapolating these model structures to the globe, even given that parameter values could be supplied regionally, is therefore quite speculative. There is an urgent need to ascertain whether certain ecosystem models structures are more robust than others, in the sense that the same model structure will be able to perform well in a variety of regions and physical settings (albeit with parameter values suitable for each region). If such structures can be identified, and the reasons for their success understood, we will have come a long way toward solving the problem of ocean biogeochemical prediction.

To address this problem, it was proposed that a set of "regional testbeds" be created to allow intercomparison of different biogeochemical models. The testbed for any particular region would include both a time series of biogeochemical data, distributed with depth, that must be predicted, and a set of physical forcings that are appropriate for the time frame of the biogeochemical data. Individual investigators could then try out their ecosystem models in these different regions. From this comparison we will learn which ecosystem structures are more robust, and can investigate and debate the reasons for their success.

This program was first initiated at the SMP PI meeting in Boulder in June 1998, but little was done until the Keystone meeting in July 1999. The major stumbling block at the time was that the computer resources required for a full 3D spatial realization in each region were beyond anyone's imagination. At the Keystone meeting, this roadblock was overcome by the suggestion that investigators in different regions, working with 3D GCM's, could distill the 3D results at selected grid points into 1D representations of those points, including parameterized upwelling/downwelling and vertical velocities, and that the biogeochemistry models could be run within this 1D framework.

At the Woods Hole SMP PI meeting in July 2000, Raleigh Hood presented a prototype for implementing this plan in the Arabian Sea. Our problem then was how to extend the range of the program to represent many different oceanic regimes. New stumbling blocks swiftly arose.

1. There was a perception that for the program to succeed, all major ecosystems must be included from the beginning.
2. This requirement was fueled in part by the belief the "ultimate goal" of the program must be a formal evaluation or "shootout" among models, leading to a "community model."
3. It was posited that the results of such a comparison be available in the OCMIP time frame, so that this "consensus" ecosystem model could be included in evaluating differences among physical models.

With all these desiderata, the amount of effort and coordination required to achieve success mushroomed, the amount of money required from NSF would have been too large, and the effort appeared doomed.

Finally, into the picture stepped a "white knight": Marjy Friedrichs of Old Dominion has volunteered to write a proposal to make the program happen. The secret is that we propose to take a much more limited approach. We believe that by limiting ourselves to the following goals, we can finally begin to make progress. First, the set of investigators who have expressed interest already represent most of the approaches that are currently being taken, so that a limited comparison within this study group will probably yield most of the insights that might come from a more inclusive group of investigators; we can add other investigators or regions as deemed necessary. Second and Third, a "shootout" seems as unwarranted at this time as it was at the time of the Toulouse workshop, which many of us attended. Despite the desire expressed by OCMIP participants for a consensus ecosystem, the physics in GCM's is really not that much advanced (if at all) over the biogeochemistry, and the biogeochemistry needs time and effort to mature at its own pace. We will support "shootouts" if, and only if, such an effort can be supported intellectually based on biogeochemistry. Otherwise, we will rely on the informal and formal comparison of results within the study group, and let OCMIP make its own decisions (informed, of course, by results from Regional Testbeds) on what biogeochemistry to include in their intercomparisons. TEST.

Participants:

Group Leaders: Marjy Friedrichs

SUMMARY:

Marjy Friedrichs is now spearheading these efforts in concert with her SMP funding (see mention of this in 2000 report) to conduct intercomparisons of the "broad suite of models designed to simulate biogeochemical cycling at the various process study sites ... in order to critically examine which ecosystem structures and formulations are most robust, and to investigate and explore the reasons for their success. Furthermore, we will critically evaluate the feasibility of developing biogeochemical models that are applicable over a wide range of diverse ecosystems."

A summary of this working group meeting should be available soon.