Abstract: Understanding how biological communities respond to environmental changes is a key challenge in ecology and the apparent decline of insect populations calls for more and better monitoring data. Recent advances in computer vision and deep learning have greatly advanced image-based methods for species identification. These methods offer a range of transformative solutions for ecological monitoring. Insect studies have long been challenged by the insurmountable workloads associated with the sorting, identifying, and counting individual specimens from bulk samples such as pan trap samples. In turn, this affects the scale of efforts to map and monitor insect diversity altogether. Image-based approaches to insect sample sorting, specimen identification and biomass estimation are poised to alleviate this constraint. Here, we present a fully automated workflow incorporating robotics to handle and image each specimen in a bulk insect sample individually. The pipeline includes the separation, localization, and pick up of each individual using custom-designed tools fitted to a general purpose collaborative robot. The robot introduces each specimen to our recently described BIODISCOVER system, which generates images and calculates size and structural features. We further show how these data can be used to accurately estimate biomass of a wide range of taxa. We apply the system to analyse arthropod samples collected across two growing seasons from emergence traps located in 48 plots across six agricultural habitat types including ploughed and unploughed fields. Our results demonstrate clear negative effects of ploughing on the size structure and total biomass of the most common taxa of insects and spiders. The detailed data generated by the system increases our understanding of changes in the abundance, composition and diversity of invertebrate community and is globally scalable to tackle many pressing questions related to global trends in populations of insects and other invertebrates. Our results add to a recent body of literature showing the huge potential of image-based methods and deep learning for specimen-based research, and furthermore offers a perspective to future automatized approaches for deriving ecological data from bulk arthropod samples.
