stability observed as food web complexity increases. A deep insight from these models is that weakly interacting species in general damp oscillations. Although the paper was not specifically about marine communities, it is worth noting that communities with food webs rather than simple food chains, with omnivores rather than food specialists, with intraguild predation (predation on a competitor, particularly its young), and with allochthonous food supply are particularly stabilized. The majority of these characteristics apply to the majority of marine food webs. Berlow (1999) amplified this advance by demonstrating the efficacy of weak interactions in generating spatial and interannual variation in community structure with a simple mussel-barnacle-whelk system.
This shift in perspective is remarkable. When the goal was identification of strong interactors, they were hard to guess. When the goal instead is to find weak interactors with potentially large damping or amplifying effects, suspects jump to mind. It is easy to predict a cottage industry among ocean ecologists in the manipulation of omnivores, for example. Observations of strong interactions, particularly when the same species interact strongly in one place and not in another (e.g., the starfish-barnacle-mussel triad in Washington State and southeast Alaska; cf. Paine, 1980), translate into questions about what stabilizing species were missing in the former. Anchovies and sardines in this view exhibit such dramatic oscillations (Bakun, 1997) because they live in simplified food chains.
I cannot help but comment on how pleased I think that Gordon Riley would be to see ocean ecologists employ these new equations and a deeper understanding of the physical and chemical processes of the sea to write quantitative descriptions and predictions of marine ecosystem processes. He sought to combine understanding of the chemistry and physics of the oceans with the Lotka-Volterra equations systematically to dissect the workings of marine ecosystems. He supported use of the term "biological oceanography" to get away from the "grab-bag of semi-defined concepts [that he perceived to dominate the ecology of his day] to clear, step-wise analytical approaches to variation in nature" (Mills, 1995, p. 39). OEUVRE coined the term "ocean ecology" in this spirit but also to encompass the pressing need to obtain at least recent paleo-information about the workings of marine ecosystems.
• How then can one understand the multiple-scale and pervasive human impacts on the sea in the face of the confounding effects of weather and climate change? Resolving and understanding anthropogenic and natural sources of variability and change on coastal to basin scales is arguably the greatest challenge to oceanographic science for the foreseeable future.
Anthropogenic effects from injection of nutrients and pollutants and from removal of predators certainly are pervasive. The challenge, given that a fully "natural" community free of anthropogenic effects appears to be a purely theoretical construct in the pejorative sense, is to avoid the trap of scurrying to understand the magnitudes of anthropogenic impacts without making the effort to understand their mechanisms. In the period since the OEUVRE workshop, I have not discovered any comparably perspective-shifting contributions to the ones mentioned under the other headings. This situation makes me continue to support the position taken by the OEUVRE group, that mankind is doing many manipulations without understanding their consequences and that a greater effort needs to be devoted to taking advantage of these manipulations to uncover the consequences and their mechanisms as the consequences arise rather than afterward. The surest method is to predict these consequences and then learn from the errors in the predictions as the perturbation proceeds.
It is possible to look back at the other sections of this paper, however, and to note that time-series analysis has played a central role in gaining understanding that the CNP ecosystem is dynamically stable, and not statically stable, and that time series similarly have played a large role in dissecting food web interactions. It is not too soon to think about the consequences of increasing atmospheric inputs to the CNP from the industrialization of Asia and to ask how to resolve them from more natural variation. Time series again seem to be prominent in the answers.
One kind of time-series analysis, called "intervention analysis" (Box et al., 1994), appears to hold particular promise for characterizing some anthropogenic effects because it was developed to do so in the context of atmospheric pollutants. The procedure is to collect a long time series that includes the period before a change in policy or other anthropogenic perturbation whose timing is known. For example, one can look for changes in the record of lead deposition after the switch to lead-free fuels. The procedure is to fit an explicit time-series model to the pre-change data set. The model is then used to forecast the post-change data set, and the residuals from this forecast contain the treatment effect. Statistical power of this method depends on the length and simplicity of the pre-change time series (i.e., the ability to fit an explicit time-series model before the perturbation). Many ocean ecologists will be reluctant to trade traditional replication in space for replication in time, but if the whole system has been altered, then replication in space is elusive.
Ocean ecologists also must become as creative in extending time series backward as they are in extending them forward. Geochemists have made great contributions in this regard with chemical proxies for temperature and nutrients