An open‐source image classifier for characterizing recreational activities across landscapes

Environmental management increasingly relies on information about ecosystem services for decision-making. Compared with regulating and provisioning services, cultural (CES) are particularly challenging to characterize measure at management-relevant spatial scales, which has hindered their consideration in practice. Social media one source of spatially explicit data where environments support various types CES, including physical activity. As tools automating social content analysis artificial intelligence (AI) become more commonplace, studies promoting the potential AI provide new insights into CES. Few studies, however, have evaluated what biases inherent this approach whether it is truly reproducible. This study introduces applies a novel open-source convolutional neural network model that uses computer vision recognize recreational activities photographs shared as media. We train 12 common map aspect recreation national forest Washington, USA, based images uploaded Flickr. The image classifier performs well, overall, but varies by activity type. model, trained from region, nearly well region same forest, suggesting broadly applicable across similar public lands. By comparing results our CNN an on-site survey, we find there apparent visitors choose photograph post After considering issues underlying models, diversity natural features (such rivers, lakes higher elevations) some built infrastructure (campgrounds, trails, roads) greater region. make training weights available software, facilitate reproducibility further development researchers who seek understand values scales—and example how build, test apply other CESs. Read free Plain Language Summary article Journal blog.