What drives cultural ecosystem services in mountain protected areas? An AI-assisted answer using social media data

Understanding how environmental and social factors shape the distribution of Cultural Ecosystem Services (CES) is essential for balancing biodiversity conservation and human well-being in protected areas (PAs). In this study, we integrate crowdsourced social media data, complementary AI-based image classification approaches (deep learning and large language models), and ecological niche models (ENMs) to map and compare the supply and demand of CES in mountain PAs across southern Europe. Using geotagged photographs classified into eight CES types and retaining only consensus-based labels, we assessed model performance, compared generalized and park-specific ENM approaches, identified key environmental and social predictors, explored bundles, and evaluated the influence of protection categories on CES suitability. Our models performed well across most CES types, confirming that social media data, when combined with AI and ENMs, provide a reliable and scalable approach for mapping human-nature interactions. The analysis revealed that while no universal set of predictors explains CES supply, some variables consistently influenced multiple CES types. By contrasting generalized and localized models, we show that the transferability depends on the type of service considered, with some CES exhibiting stable drivers across regions and others being strongly context-dependent. CES clustered into two dominant bundles: one associated with cultural and heritage values and another linked to nature-based experiences. Across most CES types and protected areas, potential supply substantially exceeded observed demand, with usage ratios typically below 5%, revealing a large proportion of underutilized areas. PAs showed higher overall CES suitability than unprotected zones, with Biosphere Reserves exhibiting particularly balanced outcomes between conservation and human use. Overall, this study shows how integrative and comparative AI-assisted approaches can support context-sensitive management and the sustainable enjoyment of cultural benefits in mountain PAs.