Cultural Ecosystem Services Image Classification and Spatial Analysis Scripts

Three scripts establish a complete technical workflow for analyzing Cultural Ecosystem Services (CES) from social media images. The workflow includes a CNN classification model using EfficientNetV2-S, parameter optimization for a Geographical Random Forest model, and an integrated RF-SHAP-GRF analysis for driver interpretation. The scripts were authored by 楠 刘 and last updated in April 2026.

Use Cases

• Classify social media images into Cultural Ecosystem Services categories using the CNN script with EfficientNetV2-S.
• Optimize GRF model parameters like bandwidth, ntree, and mtry using grid-level CES diversity and environmental variables.
• Interpret global variable importance and spatial non-stationarity of drivers using the integrated RF-SHAP-GRF analysis script.
• Generate visual outputs including RF importance plots, SHAP summary plots, and GRF local importance maps from the analysis pipeline.

Strengths

• Provides a complete, reproducible three-stage analytical pipeline documented in scripts.
• Implements advanced techniques like two-stage training, focal loss, and test-time augmentation for the CNN model.
• Systematically searches key GRF parameters (bandwidth, ntree, mtry) and evaluates performance with OOB MSE.

Limitations

• The dataset is very small at 55.5 KB, containing only code scripts without the underlying image or geospatial data.
• The performance and applicability of the workflow depend on the unspecified source and quality of the original social media images.
• No information is provided on the sample size, geographic coverage, or time range of the data used to develop the workflow.

Provenance

Source

figshare

Collection Method

Workflow scripts for processing social media images and environmental variables.

Freshness

Last updated in April 2026.