Darwin marvelled at the diversity of life on Earth. He wrote, in the “Origin of Species”: “Thus, from the war of nature, from famine and death, the most exalted object which we are capable of conceiving, namely, the production of the higher animals, directly follows. There is grandeur in this view of life, with its several powers, having been originally breathed into a few forms or into one; and that, whilst this planet has gone cycling on according to the fixed law of gravity, from so simple a beginning endless forms most beautiful and most wonderful have been, and are being, evolved.”.
However, at the very end of “Origin”, Darwin hinted at something more… he had written a masterwork, the most relevant book to ever be written (yes not that one!), explaining the mechanisms generating biodiversity. But there was something more, not only these species were there in all their diverse magnificence, but all interacted in a complex network: “It is interesting to contemplate a tangled bank, clothed with many plants of many kinds, with birds singing on the bushes, with various insects flitting about, and with worms crawling through the damp earth, and to reflect that these elaborately constructed forms, so different from each other, and dependent upon each other in so complex a manner, have all been produced by laws acting around us.”
In recent years biogeographers have been turning their attention to these “tangled bank” or ecological networks, which natural forces shape them and how they are affected by human disturbance to ecosystems. Biogeography of ecological networks has raised so much attention because food webs in particular depict the flow of energy through ecosystems (Moore & de Ruiter, 2012) and are a perfect framework to evaluate the impacts of environmental changes on ecosystem services (Montoya et al. 2003).
However, to address this issue we need data, good (enough) data. What kind of data? Well, we need information on the biotic interactions of communities at large spatial scales, continental to global. Where to get this information from? From two main sources: i) databases of local empirical networks and ii) inference of local networks from co-occurrence, species trait data and/or diet. To be rigorous, the first is proper data, while the second is not, but let’s let this one pass. I’ll try to weigh the advantages and disadvantages of each approach.
Databases of Ecological Networks
There are several databases that can be used to get these local empirical networks: GLOBI, mangal and EcoBase are the ones that I think are more used (these are the ones I’ve used anyway). These (at least GLOBI and mangal) can be accessed from R with the packages rglobi and rmangal. Additionally, a few years ago, I created an R package called fw, available in GitHub, to download food webs from different databases with a common format (I haven’t maintained it for some time, so some errors might arise).
Advantages
The main advantage of ecological networks downloaded from these databases is that these represent empirically, “real”, verified biotic interactions. Contrary to inferred ecological networks, here we have real proper data.
Disadvantages
Uneven spatial distribution – much more sampling in Europe and North America (more studies there).
Multiple spatial scales – ecological networks can be local (e.g. one small study site) or continental/oceanic (e.g. North Atlantic).
Multiple temporal scales – Researchers aggregate the interactions and species present in their study site for a given period of time, which impacts network structure.
Multiple node resolution within (basal nodes are frequently aggregated) and between networks (e.g. species, functional or taxonomic groups). If we have a network in which each node is a species and another using trophic aggregation (aggregating all species with common predators and prey species) can we really compare both? Additionally, food webs tend to have more taxonomically aggregated nodes as we go to lower trophic levels.
Multiple definitions of “interaction” (this might seem strange, but what is an interaction?). What is the threshold to say “these species are interacting”?
Using different sampling strategies to identify nodes (e.g. camera traps, environmental DNA), and interactions (e.g. stable isotopes). This has an impact on which species and interactions are present in the network.
Some databases do not have metadata (e.g. information on the methods used, scale, sampling effort and date). This makes comparability between samples difficult or impossible, which is quite problematic.
Inferred networks
Let’s say we have a metaweb of biotic interactions across Europe (like the one by Maiorano et al 2020) and we know species occurrence in our study site (there are a lot of sources for this information). We can assume that, if in the metaweb a wolf eats rabbits, then whenever wolves and rabbits co-occur this interaction is present. This is admittedly a very simplistic description of the method, but is good enough to give an idea (but check Morales-Castilla et al. 2015).
Advantages
One of the main advantages of inferring local interactions using this approach is that it provides an even spatial cover (with no spatial bias characteristic of the databases).
Another great advantage is that, for the better and for the worse, all networks are created using exactly the same methodology (node presence and interaction presence are defined with the same set of rules).
Disadvantages
The main point against this approach is that it does not represent verified interactions. Additionally, these are not presented at the spatial scale that they occur (species generally interact at a much finer scale than the one in use here, for example, 10km2 grids).
The reliability of interaction inference is dependent upon the criteria used (is it a metaweb, species traits, species diet?). Some researchers challenge this approach, arguing that species co-occurrence (which is used to identify which species are present in a local network) is not enough to assume interactions (if the species are known to interact).
Finally…
We have discussed the issues related to resorting to databases of ecological networks elsewhere (Mestre et al. 2022a). Additionally, there is a good discussion (worth reading!) in the scientific literature about whether species co-occurrence (on which inferred networks are partially based) is a good proxy for interactions (some references below). I tend to agree with the argument that, at least, “Co-occurrence is a necessary but not sufficient condition for a biotic interaction to occur.” (Stephens et al. 2020a).
The number of advantages and disadvantages of each of the two approaches is not, in itself, relevant. For instance, for the databases of local ecological networks, I identified only one advantage… but it is a major one: the fact that these are empirically verified, not inferred, interactions. On the other hand, the fact that these empirical databases are obtained with a multiplicity of methods, and temporal, spatial and taxonomic aggregations make these really difficult to compare (trust me, I know!).
I have used in my own research datasets based on databases Mestre et al. (2022b), but I´m also currently using inferred networks in something else I’m working on. Choosing each of these approaches is dependent on the kind of research question, considering that, as we move from databases to inferred networks, there is a loss of empirically verified interactions but a gain of technical data quality (same node resolution, same spatial distribution). Which is better suited to each research question is a choice that must be defended by the researcher.
As we have written in Mestre et al. (2022a), to improve the usability of ecological networks from databases, data must be integrated into a database (such as mangal), providing adequate metadata information. The goal is to have comparable datasets in which we can either account for their differences or sub-select comparable samples thus lessening some of the problems with this approach.
However, neither source of information is perfect (perfect would be to have empirical ecological networks systematically sampled in multiple points with the same methods across a large enough region).
These are the sources of information that we have. And this is what we must work with!
References
Maiorano, L., Montemaggiori, A., Ficetola, G. F., O’connor, L., & Thuiller, W. (2020). TETRA‐EU 1.0: a species‐level trophic metaweb of European tetrapods. Global Ecology and Biogeography, 29(9), 1452-1457.
Montoya, J. M., Rodríguez, M. A., & Hawkins, B. A. (2003). Food web complexity and higher‐level ecosystem services. Ecology Letters, 6(7), 587-593.
Moore, J. C., & de Ruiter, P. C. (2012). Energetic food webs: an analysis of real and model ecosystems. Oxford Ecology and Evolution.
Morales-Castilla, I., Matias, M. G., Gravel, D., & Araújo, M. B. (2015). Inferring biotic interactions from proxies. Trends in Ecology & Evolution, 30(6), 347-356.
Mestre F, Gravel D, García-Callejas D, Pinto-Cruz C, Matias MG, & Araújo MB. (2022a) Disentangling food-web environment relationships: a review with guidelines. Basic and Applied Ecology, 61: 102-115.
Mestre F, Rozenfeld A & Araújo MB (2022b). Human disturbances affect the topology of food webs. Ecology Letters.
References on the use of co-occurrence
Blanchet, F. G., Cazelles, K., & Gravel, D. (2020). Co‐occurrence is not evidence of ecological interactions. Ecology Letters, 23(7), 1050-1063.
Cazelles, K., Araújo, M. B., Mouquet, N., & Gravel, D. (2016). A theory for species co-occurrence in interaction networks. Theoretical Ecology, 9, 39-48.
Holt, R. D. (2020). Some thoughts about the challenge of inferring ecological interactions from spatial data. Biodiversity Informatics, 15(1), 61-66.
Morueta‐Holme, N., Blonder, B., Sandel, B., McGill, B. J., Peet, R. K., Ott, J. E., Violle, C., Enquiest, B. J., Jorgensen, P. M. & Svenning, J. C. (2016). A network approach for inferring species associations from co‐occurrence data. Ecography, 39(12), 1139-1150.
Peterson, A. T., Soberón, J., Ramsey, J., & Osorio-Olvera, L. (2020). Co-occurrence networks do not support identification of biotic interactions. Biodiversity Informatics, 15(1), 1-10.
Stephens, C. R. (2020a). Can we infer species interactions from co-occurrence patterns? A Reply to Peterson et al.(2020). Biodiversity Informatics, 15(1), 57-60.
Stephens, C. R., González-Salazar, C., Villalobos, M., & Marquet, P. (2020b). Can ecological interactions be inferred from spatial data?. Biodiversity Informatics, 15(1), 11-54.