Discussion Session 1: What model approaches are available that could help us scale up from data on individual responses to impacts at the level of populations, communities and ecosystems?

Chair: Jerry Blackford. Rapporteur: Kenny Schneider.

Whilst models and modelling techniques are fundamental for addressing climate science and have for several years now been explicitly part of ocean research programmes, there still exists a conceptual gap between the experimental and modelling communities. There are tensions between the general synthetic ambitions and simplistic process descriptions of models and the necessarily reductionist and more detailed approach of experiments. Further models are often seen as black boxes and the diversity and appropriateness of various model approaches not well understood by non-modellers.

What are the models for? Broadly models can be thought of as having two distinct purposes, firstly to test a conceptual understanding of how a system works and to understand the inherent dynamics of a system, secondly to make forecasts often aimed at policy makers and society in general. In reality this distinction is somewhat blurred and a particular model can be used to both ends.

One of the key issues in using model outcomes to communicate to society is the treatment of uncertainty. Two primary techniques are available; firstly although forecasts cannot be ground-truthed, hindcasts can and should be evaluated by objective, quantifiable criteria. Secondly uncertainty in parameter values and process response can be addressed by sensitivity analysis using monte-carlo or deterministic scenario techniques. These have the drawback of being computationally intensive.

What kinds of model approaches exist? It is not always apparent that a diverse range of model approaches exist and that these model approaches are not always well integrated. A generic point of agreement is that no one model type can fully address an issue as complex as climate change, the ideal being a suite of communicating models. Directly coupling different model systems is possible, but limited by computational resource and error propagation. The following list is by far from exhaustive, but serves to illustrate the scope of modelling approaches.

Model Class

Typical characteristics


Earth System models

Combine terrestrial, atmosphere and ocean systems, biology simplistic or omitted completely. Coarse physical resolution

Carbon budgets, mean physical properties, climate.

Global Ocean Models

Simplistic biogeochemistry, single or few functional types, moderate physical resolution, tend not to resolve coastal and shelf sea processes

Carbon budgets, physical properties

Biogeochemical / Ecosystem models

Detailed physics, Intermediate biogeochemistry – several functional types, tending to lower trophic levels. Can be regional and resolve coastal and shelf systems. Some plasticity.

Carbon and nutrient cycles, productivity, regional physical heterogeneity

End to End models

Usually regional merging physical, microbial and higher trophic levels, generally not elaborate in terms of process descriptions but have many functional types.

Productivity from lower to higher trophic levels

Physiological models

Including Trait based models / Optimal allocation / Dynamic Energy budget models, these model physiological trade-offs within individuals or functional groups providing a far more plastic (and realistic) response to environmental drivers.

Productivity, carbon and nutrient cycling

Evolutionary models

Closely related to physiological models, allow for adaptation or evolutionary process, generally restricted to fewer functional groups, but may have many instances of each.

Productivity, carbon and nutrient cycling, evolutionary outcomes

Population models

Related to biogeochemical / ecosystem models but may deal with single species and the suite of processes that affect them.

Biomass, abundance

Food web / Ecological models

Models with detailed food web dynamics but tending not to be spatially or even temporally dynamic.

Biomass, abundance.

In general neighbouring classes of models may overlap significantly. Typically there is a trade-off between geographic completeness and physical resolution, and then physical resolution and biological realism.

As an illustration of usage one might use a regional biogeochemical model, forced by an earth system climate model to estimate the range of environmental conditions that a particular region might be exposed to over the next century. These environmental conditions could them inform the appropriate choice of scenarios for say temperature and CO2 experimental manipulations. The experimental outcomes could be combined with others to inform a biogeochemical response framework, or more specifically to parameterise or evaluate physiological or population models. Forecasting specific policy relevant outcomes certainly requires understanding of both the physical and chemical environmental drivers and the ecological response. Added complexity exists if the ecological response feeds back to the environmental drivers, commonly the case in OA studies as the biological production – respiration balance significantly impacts the systems carbonate chemistry.

How can we improve the bidirectional flow of information between models and experiments? In recent years the ocean community have moved from employing models as a synthesis tool a posteriori (for example the JGOFS programme synthesis phase) to running model programs concurrently with experimental phases (for example, the EU EPOCA programme and the UK UKOA programme). There is a general desire for experiments to inform models and vice-versa but to date this has been restricted. Two reasons can be put forward, one organisational, one conceptual. The optimal program for model-experiment interaction should consist of an initial design phase to ensure that the modellers understand the scope of data likely to emerge and that the experiments can be designed to maximise the utility of emerging data for models and secondly sufficient time exists, post experiments, to allow models to integrate the emerging data. Such an interaction is difficult to achieve within the confines of a three to four year programme. The second reason is that often insufficient resources are devoted to transferring the specific outputs of an experiment to more generic conceptual understanding that can be incorporated into models. We, as scientists and optimists, have a tendency to plan an experiment to test a hypothesis and expect a nice unambiguous result. The reality, especially in OA related experiments, is that numerous experiments have produced apparently contradictory results and that the modeller is left confused and nervous about committing any given response to code. What is missing is an intermediate step, one could term this the development of a conceptual framework, which has the tendency to fall between the experimental and modelling camps, but, in general, requires both to achieve. This “missing-link” activity may well be facilitated by targeted cross-programme activities.

Finally it is worth giving a modellers perspective on data types. Often model development (or evaluation) is hampered by a lack of spatially or temporally explicit data sets and a lack of meta data (i.e. what are the physical conditions, is the biology replete or stressed?). Most valuable are long term time series or regional transect surveys; often traditional cruise based snapshot data is of limited use.