Applying deep learning on social media to investigate cultural ecosystem services in protected areas worldwide

Abstract

AbstractProtected areas (PAs) are the cornerstone of conservation efforts. Although they provide many benefits to humanity, the variability in the provision of cultural ecosystem services (CES) among global PAs remains unknown. To investigate this, we combined Convolutional Neural Networks with hierarchical clustering to categorize photos from Flickr taken in PAs worldwide. A final sample of 87,090 photos in 2813 PAs within 207 countries was obtained. Distinct global patterns of CES activities emerged. Such activities had three main interaction types: human-nature (abiotic), human-nature (biotic) and human–human. Human-nature (abiotic) interactions dominated in mountain ranges. Human-nature (biotic) photos were more common in equatorial countries, and human–human photos occurred mainly in Europe. To determine the extent of the influence of biome type of PAs on CES, mixed-effects models were subsequently run. These models additionally included the country of PAs as a random effect. Despite differences in physical environments, PAs within each country generally shared similar CES types. Moreover, the effect of biome differences was marginal, thereby demonstrating that country-level management of PAs likely has a more important role in influencing CES activities in PAs. To conclude, we suggest that our results demonstrate the utility of social media data for understanding visitor activities in PAs.