This study developed an accurate and scalable machine learning approach to predict and quantify CES flows across Florida’s natural and working lands at the Census County Subdivision scale from 2014 to 2019.

The Qwen2.5-VL-OSL model was used to predict 12 CES classes, including 10 widely recognized outdoor recreational activities, as well as wildlife viewing and landscape aesthetics.

CES Flows

• Qwen2.5-VL-OSL is a robust one-shot learning model that provide 92% overall accuracy in CES classes.
• The 12 CES classes include: Biking, Boating, Camping, Fishing, Hiking, Horseback Riding, Hunting, Shelling, Surfing, Swimming, Wildlife Viewing, and Landscape Aesthetics.
• CES flow units are defined as the density of Photo-User-Days (PUDs), calculated by dividing the total PUDs by the number of unique users and the six-year period.

How to Cite This Map

If you use this map in your work, please cite the following:

Hao-Yu Liao, Chang Zhao, Jiayi Song, and Wei Shao. 2025. Mapping Cultural Ecosystem Services Using One-Shot In-Context Learning with Multimodal Large Language Models. In The 33rd ACM International Conference on Advances in Geographic Information Systems (SIGSPATIAL ’25), November 3–6, 2025, Minneapolis, MN, USA. ACM, New York, NY, USA, 4 pages. https://doi.org/10.1145/3748636.3764178