Graph-based Investigation of Insect Spread Dynamics

In this project, we plan to design methods to monitor native insect species whose occurrence has been, or is being, affected by the recent introduction of alien (i.e., exotic) invasive insect species. Biological invasions are considered a major threat to biodiversity, as invasive species can affect the structure of communities in the recipient environment through direct and indirect effects on native species. These effects can be subtle and are often neglected in the early stages of invasions, overshadowed by the more obvious effects of new invaders. As such, they may be overlooked until they have caused irreversible cascading effects that affect entire communities. In the management of alien species, early detection, supported by an effective high resolution monitoring system, is often critical to the effectiveness of eradication, containment or control measures, as many measures are most effective when insect densities are low.

On the one hand, by monitoring the presence of invasive species, we will estimate their spread and the extent of the threat they pose to native species. On the other hand, by monitoring affected native species, we will assess the actual risk they face. We will consider three exemplary complexes of native species -- coleopterans, lepidopteran leafminers and leafhoppers -- along with their alien invading counterpart, respectively: Popillia japonica, Aspilanta oinophylla, and Erasmoneura vulnerata. As many other alien insect species, the latter have entered Switzerland from its southern border within the last decade. In Ticino, the potential far-reaching effects on native biodiversity can and need to be investigated in order to alert and improve the preparedness of other Swiss regions that might be colonized later.

We plan to apply methods based on machine learning and, more specifically, on spatio-temporal graph neural networks. Designed methodologies and novel methods will rely on different data sources, such as images, sounds recorded in the soil, local weather and crop status data. Neural network will be used to detect and numerically assess the presence of the insect species, either as adults (images) or in the larval stage (sound in soil) over time; this information, along with local weather and crop status as well as the interactions among different species (e.g., alien and native), will be used to predict the presence and spread of these insect species over space and time.

We aim to obtain novel methods for detecting/predicting and assessing the presence of specific insect species in larval stages too, which we expect to provide a further boost in spread forecasting accuracy while lowering the monitoring effort; our multi-insect models will deal with a large amount of heterogeneous information and provide the unprecedented ability to predict the spread of alien and native insects. As a final result, our work will provide fundamental information to understand the systemic consequences of an increasing frequency of biological invasions, beyond the direct damage caused. Ideally, we will uncover patterns that allow early detection of critical situations that are likely to precede adverse impacts of alien invasions on native biodiversity (``early warning''), so that activities can be deployed to find solutions that mitigate those impacts.

The contribution described above is important in the context of the COST Action CA22129 "InsectAI - Using Image-based AI for Insect Monitoring & Conservation". Besides our contribution on machine learning-based methods for monitoring insects, we will also contribute with data collected during the project; at the same time we will benefit from the data collected by other participants to the COST Action.