FL-CES-FLOWS-CLIP

Mapping Cultural Ecosystem Service Flows from Social Media Imagery with Vision–Language Models: A Zero-Shot CLIP Framework

---

Overview

This repository provides code and reproducible workflows for mapping Cultural Ecosystem Services (CES) from Flickr images using the open-source Contrastive Language–Image Pretraining (CLIP) model that integrates:

• CLIP's visual embeddings-based Binary Random Forests (BRF) for open-set image filtering
• Zero-shot CLIP for CES image classification
• CES flow mapping as indicated by average annual Photo-User-Days (PUD) and average annual PUD/user at 1 km resolution

The workflow leverages Flickr imagery (2014–2019) across the state of Florida's natural and working lands to support scalable, interpretable, and reproducible CES analysis.

---

Interactive Web Application

https://es-geoai.rc.ufl.edu/FL-CES-Flows-CLIP/

Installation & Quick Start

A fully reproducible Conda environment is provided via environment.yml.

1. Install Conda (Miniconda recommended)

• https://docs.conda.io/en/latest/miniconda.html

2. Create the Conda environment

conda env create -f environment.yml

3. Activate the environment

conda activate FL-CES-Flows-CLIP

4. Launch Jupyter Lab

jupyter lab

5. Open the main workflow notebook

notebooks/CLIP-BRF-ZS/Main.ipynb

Repository Structure

FL-CES-FLOWS-CLIP/
├── notebooks/          CLIP-BRF-ZS and CES analysis workflows
├── model-outputs/      Aggregated CES and PUD outputs
├── model/              CLIP visual embedding–based Binary Random Forest models
├── figures/            Figures outputs
├── environment.yml     Conda environment specification
└── README.md

Large intermediate outputs (e.g., full-resolution CSV or GeoTIFF files) are generated during notebook execution and are not required for installation.

Citation

If you use this repository, please cite one or more of the following works. Citation files (RIS format) — copy and import into Zotero / EndNote / Mendeley.

TY  - PREPRINT
TI  - Mapping Cultural Ecosystem Service Flows from Social Media Imagery with Vision–Language Models: A Zero-Shot CLIP Framework
AU  - Liao, Hao-Yu
AU  - Zhao, Chang
AU  - Koylu, Caglar
AU  - Cao, Haojie
AU  - Qiu, JiangXiao
AU  - Callaghan, Corey T.
AU  - Song, Jiayi
AU  - Shao, Wei
PY  - 2025
DO  - 10.32942/X29S8C
UR  - https://doi.org/10.32942/X29S8C
JO  - EcoEvoRxiv
ER  -

TY  - CONF
TI  - Mapping Cultural Ecosystem Services Using One-Shot In-Context Learning with Multimodal Large Language Models
AU  - Liao, Hao-Yu
AU  - Zhao, Chang
AU  - Song, Jiayi
AU  - Shao, Wei
PY  - 2025
DO  - 10.1145/3748636.3764178
UR  - https://doi.org/10.1145/3748636.3764178
JO  - Proceedings of the 33rd ACM International Conference on Advances in Geographic Information Systems
VL  - 33
SP  - 1
EP  - 4
PB  - Association for Computing Machinery
CY  - Minneapolis, MN, USA
ER  -