An AI‐driven biomass estimation tool enabling population‐ and specimen‐level variation analyses. A case study on parasitoid flies

Insects contribute substantially to global animal biomass, yet quantifying how biomass is distributed among taxa and changes through space and time remains challenging. Traditional body-size measurements are laborious and often rely on regression models that cannot fully capture inter-and intraspecific morphological variation, limiting the accuracy of biomass estimation. Here, we present the first ecological application of InsectMorphoAI, an automated framework that estimates insect body volume and curvilinear body length from standardised two-dimensional images. The method combines instance segmentation with biologically interpretable geometric reconstruction to generate specimen-level body-size measurements while preserving voucher specimens for subsequent analyses. Using bristle flies (Diptera: Tachinidae) as a model system, we quantified species composition, abundance, body size, body proportions, and biomass proxies in assemblages sampled during two standardised Malaise-trap surveys conducted 24 years apart in a Mediterranean forest. Automated body-volume estimates enabled biomass to be partitioned among species and revealed how differences in body shape influence the relationship between abundance and biomass. Comparison between body volume and curvilinear body length further demonstrated that these proxies, although broadly consistent at the assemblage level, may diverge for individual species because of differences in body robustness. A specimen-level bootstrap analysis showed that reliable species-level biomass estimates can generally be obtained by imaging relatively few specimens, providing practical guidance for future biodiversity studies. More broadly, our results demonstrate that automated morphometrics can transform standardised specimen images into quantitative ecological data, enabling biomass-informed analyses that complement traditional abundance-based approaches and expand the use of natural history collections in long-term biodiversity monitoring.