An Open‐Source Image Classifier for Characterizing Recreational Activities Across Landscapes

Citation

Winder, S. G., Lee, H., Seo, B., Lia, E. H., & Wood, S. A. (2022). An open‐source image classifier for characterizing recreational activities across landscapes. People and Nature, 4(5), 1249-1262. doi.org/10.1002/pan3.10382

Link to file

Environmental managers are increasingly using information about ecosystem services to guide their decisions. But not all ecosystem services are equally easy to measure. Compared to things like clean water or crop pollination (i.e., regulating and provisioning services), cultural ecosystem services (CES) are especially tough to pin down. These are the non-material benefits people get from nature, like inspiration, spiritual connection, or recreation. Because they’re hard to measure at the scales that matter for management, they often get left out of the decision-making process.

Social media offers one promising way to fill that gap. Photos and posts tagged with location data can show where and how people are using natural spaces for activities like hiking, biking, or fishing. As artificial intelligence tools for analyzing this kind of content become more common, researchers have been excited about their potential to reveal new insights about cultural ecosystem services. But few studies have stopped to ask: what biases come with this approach? And can the results be reproduced?

This study set out to answer those questions. The researchers built a new, open-source AI model, a type called a convolutional neural network, that uses computer vision to recognize recreational activities in photos shared on social media. They trained it to identify 12 common activities, then used it to map recreation across a national forest in Washington State, based on images uploaded to Flickr.

The model performed well overall, though accuracy varied by activity type. Importantly, a model trained on photos from one part of the forest worked nearly as well in a different part of the same forest, suggesting it could be useful across similar public lands.

But the researchers also uncovered a key bias. By comparing the AI’s results with an on-site survey of visitors, they found that not all activities are equally likely to be photographed and posted online. Some activities are overrepresented in social media data, while others are underrepresented. This means the picture social media paints is incomplete.

After accounting for these data and model limitations, the team mapped where recreational activity was most diverse. They found that natural features like rivers, lakes, and higher elevations — along with built infrastructure like campgrounds, trails, and roads — supported the widest variety of activities in the region.

The researchers made their model available as open-source software. Their goal is to help other scientists build on this work; improve reproducibility; and ultimately give land managers better tools for understanding the recreational value of public lands, and for bringing cultural ecosystem services into the decisions that shape them.

Read the plain language summary provided by the researchers.

Abstract

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