Automated imaging and taxonomy-guided AI for accurate and scalable soil biodiversity diagnosis

Soil biodiversity is crucial for our functional biosphere and 95% of our food relies on healthy soil. Yet over 70% of earth’s soil is degraded, highlighting the urgency to restore soil health, a goal emphasized by the recent European Soil Monitoring directive. Soil-born nematodes exist within all trophic levels of the soil food web and represent a universal bioindicator of soil biodiversity, even in degraded soils. However, this indicator is not widely used and requires nematologist and soil ecology experts as well as significant labor-intensive manual analyses. With the support of EIC and EIT, we developed an automated end-to-end diagnosis tool, comprised of an automated soil sample imaging system (NEMASCOPE ^TM) and a multi-level, taxonomy-guided computer vision AI for nematode species identification. Our technology provides quantitative soil biodiversity parameters based on the existing scientific framework of Nematode-based Indices (NBIs), assessing soil health, immunity, fertility, soil-based plant parasites, carbon cycling, pollution, and organic degradation pathway, among other NBIs for soil assessment. Validated by research phytopathogenic laboratories, the tool demonstrated to be in average more accurate (>90%) and over 20-times faster (<15 min) in end-to-end biodiversity analysis compared to manual analysis. The system’s nematode identification performance we evaluated on Root-knot nematode (RKN) species level identification accuracy across Meloidogyne species that are among the most economically damaging plant-parasitic nematodes, using naturally infested field samples containing M. chitwoodi, which are challenging to distinguish from other Meloidogyne species due to their morphological similarities. Compared with manual identification, the AI-based approach achieved an accuracy of ~95% in identifying RKN genera with species-level prediction accuracy for M. chitwoodi with ~96%, essentially matching manual expert performance. Our platform demonstrates expert-level accuracy for nematode identification down to the species level particularly necessary for plant-parasite index (PPI) assessment. The technology allows scalable, industry-ready diagnostics addressing the global shortage of nematologist expertise with the potential to become a new standard in commercial and research sectors, aiding in the global efforts to manage and restore soil health.