The Southern Ocean, characterized by high levels of endemism and ecological diversity, is particularly vulnerable to climate-driven environmental change. Pioneer benthic species represent effective indicators of ecosystem responses, as they are among the first organisms to react to changes in seasonality, trophic availability, and physical conditions. In recent years, artificial intelligence-based photoanalysis tools have improved the efficiency of quantitative benthic image analysis; however, most existing models focus on macro- and megabenthic communities from tropical and temperate environments and are rarely used to characterize polar communities. Automated semantic segmentation models for pioneer benthic species colonising Autonomous Reef Monitoring Structures (ARMS) deployed in the Ross Sea (Antarctica) were developed and validated using TagLab, an AI-based semantic segmentation platform. Three independent models were trained on the dominant bryozoan taxa, Micropora sp., Camptoplites spp., and Beania erecta Waters, 1904, collectively accounting for over 80% of the total annotated surface area. Models achieved high segmentation performance (accuracy: 0.939-0.982; mIoU: 0.890-0.965) and reduced analysis time by approximately 90% compared to manual segmentation workflows. Percent cover estimates obtained from automated and manual segmentation showed strong agreement across all taxa and immersion intervals (Spearman ρ > 0.98).
Estimates of colony abundance, based on automated individual colony detection and counting, proved less robust, with discrepancies attributable to species-specific morphological artefacts affecting automated delineation. These findings represent a first validation of AI-assisted semantic segmentation for quantitative monitoring of pioneer benthic biodiversity in polar environments, yielding promising results that support its broader applicability to additional taxa, ARMS deployments and sites across the Southern Ocean.