Networks of species interactions can capture meaningful information on the structure and functioning of ecosystems. Yet the scarcity of existing data, and the difficulty associated with comprehensively sampling interactions between species, means that to describe the structure, variation, and change of ecological networks over time and space, we need to rely on modeling tools with the capacity to make accurate predictions about how species interact. Here we provide a proof-of-concept, where we show a simple neural-network model makes accurate predictions about species interactions, and use this model to reconstruct a metaweb of host-parasite interactions across space, and assess the challenges and opportunities associated with improving interaction predictions. We then provide a primer on the relevant method and tools that will guide the development and integration of these tools, and provide a road map forward toward integration of multiple sources of data and methodlogical approaches (including statistical, dynamical, and inferential models) to sketch the path forward for this research program.

Ecosystems are composed of networks of interacting species. These interactions allow communities of species to persist through time through both neutral and adaptive processes. Despite their importance, a robust understanding of (and ability to predict and forecast) interactions among species remains elusive. This knowledge-gap is largely driven by a shortfall of data—although species occurrence data has rapidly increased in the last decade, species interaction data has not kept pace, largely due to the effort required to sample interactions. This means there are many interactions between species that occur in nature, but we do not know these interactions occur because we have never observed them. These so-called “false-negatives” bias data and hinder inference about the structure and dynamics of interaction networks. Here, we demonstrate the realized rate of false-negatives in data can be quite high, even in thoroughly sampled systems, due to the intrinsic variation in abundances across species in a community. We illustrate how a null model of occurrence detection can be used to estimate the false-negative rate in a given dataset. We also show how to directly incorporate uncertainty due to observation error into model-based predictions of interaction probabilities between species. One hypothesis is that interactions between “rare” species are themselves rare because these species are less likely to encounter one-another than species of higher relative abundance, and that this can (in part) explain the common pattern of nestedness in bipartite interaction networks. However, we demonstrate that across several datasets of spatial or temporally replicated networks, there are positive associations between species co-occurrence and interactions, which suggests these interactions among “rare” species actually exist but simply are not observed. Finally, we assess how false negatives influence various models of network prediction, and recommend directly accounting for observation error in predictive models. We conclude by discussing how the understanding of false-negatives can inform how we design monitoring schemes for species interactions.

This is a conference paper presented at the ICLR 2023 "Machine Learning for Remote Sensing" workshop.Protecting and restoring ecological connectivity is essential to climate change adaptation, and necessary if species are to shift their geographic distributions to track their suitable climatic conditions over the coming century. Despite the increasing availability of near real-time and high resolution data for landcover change, current connectivity planning projects are hindered by the computational time required to run connectivity analyses at realistic geographic scales with realistic models of movement. This bottleneck precludes application of optimization algorithms to prioritize ecological restoration to maintain and improve connectivity. Here we propose we can make progress toward overcoming these challenges using machine-learning methods. Our proposed methods will enable rapid optimization of connectivity prioritization and extend its application to many more species than is currently possible. We conclude by illustrating how this project will contribute to efforts to apply connectivity conservation using an example of ongoing restoration in southern Québec.

No abstract

No abstract

No abstract