Uncertainty is of paramount importance when dealing with paleo-food webs. There is uncertainty associated with and inherent to the structures of many modern food web models, but there is always the opportunity of reducing or even eliminating much of that through empirical observation. Opportunities are more limited in the fossil record. Two different approaches to the problem have developed, namely restricting reconstruction to exceptionally well-preserved portions of the record, and modeling uncertain data and the impact of uncertainty. Both approaches are discussed in the following section, but since the former approach differs little from the reconstruction of modern communities, emphasis is placed on the latter.
The first step in constructing a food web network model of a paleocommunity is the assembly of a comprehensive taxonomic database of species known to have been members of that particular community. I will not discuss definitions of paleocommunities here. There are two sources of uncertainty in this step: missing species and community averaging. There are many taxa that have never been preserved as fossils, or their preservation prevents proper enumeration for network purposes, for example many microbial taxa. Solutions to these problems include a restriction to lagerstätte, or restricting one’s interpretations to comparative dynamics. The first option was adopted by Dunne et al. (2008) in their reconstruction of two Cambrian food webs. These marine food webs preserve several grades of organisms that are generally not present as fossils, namely soft-bodied organisms. The taxonomic completeness of the food webs are therefore expected to be greater than what is otherwise available for most of the fossil record. The other approach is suitable when two or more communities are being compared, for example the Permo-Triassic communities of the Karoo Basin compared in Angielczyk et al. (2006) and Roopnarine et al. (2007). This approach asserts that although the reconstructed food webs lack all the original species, consistent rules of reconstruction will result in comparable reconstructions. Any observed differences will therefore be the result of true ecological differences, and not be functions of the network model. The approach necessarily assumes reasonable taphonomic uniformity among the paleocommunities.
The impact of incompleteness has been investigated preliminarily by “fossilizing” two high quality modern marine food webs, the Greater Antillean coral reef discussed earlier, and the San Francisco Bay community. Fossilization was simulated by removing all genera, from the networks, that are not represented at the genus level in the Sepkoski genus-level database. Comparing the perturbation dynamics (upcoming post) of the modern and fossil networks shows that the fossilized and incomplete networks exhibit greater variance and lowered resistance to perturbation, but that the qualitative form of the results is unchanged.