A Spatio-temporal Bayesian Network Approach for Revealing Functional Ecological Networks in Fisheries

Ecosystems consist of complex dynamic interactions among species and the environment, the understanding of which has implications for predicting the environmental response to changes in climate and biodiversity. Machine learning techniques can allow such complex, spatially varying interactions to be recovered from collected field data. In this study, we apply structure learning techniques to identify functional relationships between trophic groups of species that vary across space and time. Specifically, Bayesian networks are created on a

window of data

for each of the 20

geographically

different and

temporally

varied sub-regions within an oceanic area. In addition, we explored the

spatial

and

temporal

variation of

pre-defined functions

(like predation, competition) that are generalisable by experts’ knowledge. We were able to discover meaningful ecological networks that were more precisely

spatially-specific

rather than temporally, as previously suggested for this region. To validate the discovered networks, we predict the biomass of the trophic groups by using dynamic Bayesian networks, and correcting for spatial autocorrelation by including a

spatial node

in our models.