Training, Validation, and Testing Labels, and Scripts for Semantic Segmentation of Tea (Camellia sinensis) Plantations in Kenya

In this folder, we host the label data and script we used to build models for our paper: "Do Geospatial Foundation Models Outperform Sentinel-2 Composites? A Counterexample from Semantic Segmentation of Tea Plantations in Western Kenya," submitted to the journal ENVIRONMENTAL RESEARCH: Food Systems. label_data ==> Contains training, validation, and testing label GeoPackages. Each has a column called label, which indicates whether it is a tea plantation (1), a non-tea area (0), or unknown (-1). You can use these for training, testing, and validating any supervised machine learning model of your choice freely. We digitized these polygons in 2024 using the Google Hybrid base map in QGIS. Refer to the manuscript for detailed procedure on label creation. scripts ==> Contains the codes we used for downloading input images (Sentinel-2, AlphaEarth, and TESSERA). You can use them to access the data we used or adapt them for your own use freely. If you use either the data or the script in your work, consider citing our paper accordingly: Oluoch, W. A., Masayi, N. N. (XXXX). Do Geospatial Foundation Models Outperform Sentinel-2 Composites? A Counterexample from Semantic Segmentation of Tea Plantations in Western Kenya. Journal XX(yy) aaa-bbb. doi:10.....