This week’s edition of Nature featured a comment by Drew Purves and colleagues titled “Ecosystems: Time to model all life on Earth”, in which they propose that we need a new model type, general ecosystem models (GEM), that describe the interplay of all major organism types on earth. They say that GEMs
[…] could capture the broad-scale structure and function of any ecosystem in the world by simulating processes — including feeding, reproduction and death — that drive the distribution and abundance of organisms within that ecosystem. Ecologists could apply a GEM to African savannas, for instance, to model the total biomass of all the plants, the grazers that feed on the plants, the carnivores that feed on the grazers and so on. Over time, the flows of energy and nutrients could be mapped between them. All of the organisms would be grouped not by species, but according to a few key traits such as whether they are plants, birds or mammals, cold blooded or warm blooded, diurnal or nocturnal. By encoding processes such as migration and predation into simple mathematical and computational forms, ecologists could model what happens to the various groups over time.
On the technical side, it seemed that Purves et al. envision a GEM a bit like a Dynamic Global Vegetation model (DGVM), with functional grouping, individuals in cohorts etc., only that animals are added explicitly. I say explicitly, because implicitly DGVMs do include pest outbreaks, soil organisms etc. in their processes for mortality or decomposition.
The issue of model structure, however, is anyway discussed rather shortly before the article moves to a different topic, namely data: Purves et al. state that the
biggest stumbling block to constructing GEMs […] is obtaining the data to parameterize and validate them.
I would add: and also how to use these data to parameterize the model. After all, it is likely that the suggested data campaign would end up with a collection of heterogeneous, non-completely independent data types that need to be combined for model parameterizations and selection. We discuss these issues in our recent review on bringing together DGVMs with data.
Nevertheless, I’m a bit skeptical whether data, or to extend this, connecting data to the models, is really our sole worry – my personal feeling is that a major advance in theory is needed as well to create models that can really satisfyingly address the specifications laid out in the article. One topic I’m thinking of is how to deal with the different spatial scales, i.e. how to model these networks of organisms operating on different scales in a spatially heterogeneous environment, and scale all this up to the global scale. Sure, you can assume a well-mixed community, and with a lot of data you can always parameterize this model, but then you are implicitly including all the spatial structure in the model parameters, and the question is whether your model will then really extrapolate far better than a purely correlative approach (a point we also make here).
Still, one can’t make progress by not trying things out, so I’m curious to see more results on this in the future.