Climate Modeling for Ecosystems Studies: Ensemble Approaches

An ensemble approach is recommended for projecting marine ecosystem responses to climate change.  It is impossible to predict the exact trajectory of the climate system, especially with regards to phasing, i.e., the timing of multi-year fluctuations.  Errors of this sort are compounded by the flaws in the mechanistic models, whether of the empirical or dynamical type, that are used to link the physics to the biology.  A practical way to deal with these inevitable errors, and to quantitatively estimate uncertainties, is through the use of model ensembles.  There are a variety of methods that have been employed, and each involves trade-offs.  There are certainly benefits to being able to draw from a relatively large number and diverse set of individual projections; these benefits may often exceed those associated with the use of more sophisticated and in principle more realistic models, if only a small number of runs is then feasible.  On the other hand, simple representations of the mechanisms linking the physics to the biology have their own problems.  They are unlikely to be effective if there are future changes in the “rules” controlling the aspects of the ecosystem of interest.  Perhaps models of intermediate complexity, which include interactions between key processes, but can still be run multiple times under various climate forcing scenarios, represent an attractive option.  Assuming multiple model projections are available, there are choices that must be made.  For example, ratings can be assigned based on past performance or hindcasts, and these ratings can then be used to select only the better models, and perhaps to weight the contributions from individual simulations in the construction of ensemble means.  While these ratings can be done objectively, there is still subjectivity in picking the criteria.  Many of the issues mentioned above have been encountered in the development of long-range weather forecasts, and there may be some useful lessons.