Next generation pollinator monitoring in agroecosystems: eDNA meets camera traps and AI

Biodiversity monitoring is a fundamental tool allowing researchers and land managers to understand the trends of indicator populations, with extension to the state of entire ecosystems. Scientific evidence points to a precipitous decline of insect abundance in recent decades. Due to insects’ great functional diversity, significant difficulty lies in teasing out the exact drivers of insect population decline. In response to management actions, measurement of recovery rates will rely on standardized monitoring strategies in the future, underpinned by sound baseline data and analytical approaches. Systematic documentation of insect traits should further improve data collection and guide effective management practices that can then be applied elsewhere, ideally allowing recovery of insect species diversity. Such breakthroughs will result in improvement of biodiversity at scale.

Our current contribution focuses on potential solutions for monitoring pollinator communities in agricultural landscapes. Strategies include implementation of modern and experimental technology-based insect monitoring techniques used as stand-alone approaches or in combination. We present recent findings from research trials on monitoring insect biodiversity using environmental DNA (eDNA) collection, camera trapping (CT), and artificial intelligence (AI) algorithms for identification, focusing on their use-cases, potential benefits, and current limitations. Specifically, we study the use of eDNA-derived community composition as contextual information to inform and constrain AI-based classification and quantification of insects, effectively providing site-specific ecological priors for visual inference. This multimodal-based approach should reduce misclassification for non-native taxa and enhance abundance estimates compared to image data alone.

We further discuss methodological combinations that more comprehensively document the insect community at individual sites - with an objective of mitigating method-specific detection biases. Trait-based digital repositories of insect species should accelerate the access to data on interactions of insects in the environment, improving knowledge on conserving them. As documentation requires appropriate deposition of voucher materials, we consider how input of trait-based data into the Global Repository of Insect Traits (GRIT) can support conservation efforts. We present preliminary findings as a proof-of-concept of a workflow combining insect eDNA collection with CT/AI.