Oceanic Front Identification Dataset: Sentinel‑2, AlphaEarth Embeddings & Multi‑scale Features

This dataset supports the research article "Intelligent Identification of Oceanic Fronts by Integrating AlphaEarth Embeddings and Multi-scale Spatial Features". It contains all labelled samples, multi-scale feature extractions, AlphaEarth embeddings, and analysis scripts required to reproduce the figures and tables in the paper.

Content overview:- 1,055 manually labelled sample points (estuary front / water) from five estuaries (Yangtze, Pearl, Qiantang, Mississippi, Colorado), with Sentinel‑2 L2A surface reflectances (11 bands), 64‑dimensional AlphaEarth embeddings, and multi‑scale spatial features (mean and standard deviation for window sizes 3×3 to 15×15).- Random forest feature importance scores .- Canonical correlation analysis (CCA) results.- Leave‑one‑estuary cross‑validation input dataset (with estuary labels).-Sentinel‑2 satellite images of the Yangtze, Qiantang, and Colorado River estuaries, which can be used for oceanic front identification and prediction (due to the large file size of the images, they are stored in six separate folders).- Google Earth Engine (JavaScript) scripts for feature extraction, the two‑step random forest + red‑band constraint method, and optimal red‑band threshold calculation (Youden index).- Python scripts for leave‑one‑estuary cross‑validation and CCA + Pearson correlation analysis, together with a requirements.txt file specifying exact package versions.

For further details, please see the README.md file included in this repository.