Environmental DNA and AI-based automated image recognition for soil fauna monitoring in Switzerland: Case studies, challenges, and perspectives

Soil biodiversity remains one of the least documented components of terrestrial ecosystems, largely due to limited taxonomic expertise, major knowledge gaps, and the slow and costly nature of traditional identification methods, compounded by the declining availability of specialists. To address this gap, we conducted two studies across Switzerland to evaluate how high-throughput methods - environmental DNA (eDNA) metabarcoding and artificial intelligence (AI) image recognition - can support the development of soil biodiversity indicators.

We collected soil samples from 320 sites across all biogeographic regions of Switzerland from different established aboveground biodiversity monitoring networks, targeting arthropods for both morphological and eDNA-based analyses. We used metabarcoding to detect and identify the most abundant microarthropods: springtails (Collembola) and mites (Oribatida and Gamasina). Additionally, we prepared a collection of mounted specimens and established a corresponding reference database of long DNA barcodes. In the second study, we digitized the soil arthropod sample collection using a macrophotography imaging system. High-resolution images were used to train deep-learning image recognition models at multiple taxonomic levels and to classify specimens according to morphological traits linked to their degree of adaptation to soil, thereby enabling the derivation of the Soil Biological Quality index (QBS-ar). To further enhance classification performance, we combined machine learning approaches with deep-learning models to optimize image annotation workflows and implemented active learning strategies to efficiently refine training datasets.

We compare the outputs, uncertainties, and methodological constraints of both approaches, emphasizing how these factors influence biodiversity indicators such as richness estimates, community composition, and habitat-specific assemblage patterns. We further assess the advantages and limitations of each method across different operational contexts, ranging from large-scale monitoring schemes to targeted ecological assessments. In addition, we discuss the implications of our findings for future national and international biodiversity monitoring programs, underscoring the discrepancies observed between widely used aboveground indicators - such as vascular plant diversity - and soil arthropod diversity. Our results highlight the importance of systematically integrating soil fauna into ecosystem monitoring frameworks. Finally, we outline key research priorities to improve data integration, strengthen reference databases, and advance the broader implementation of scalable, high-throughput tools for soil biodiversity assessment.