Code and Results for the GRL Paper "Foundation-scale Satellite Embeddings Reframe Hydrological Generalization as a Representation Problem"
收藏资源简介：
This repository contains the code, trained model weights, and results associated with the paper "Foundation-scale Satellite Embeddings Reframe Hydrological Generalization as a Representation Problem" (Ou & Zheng, 2026, Geophysical Research Letters). Contents:- code_dataloader/ Data loading utilities- code_model/ Model architecture, training, and evaluation scripts- cross_validation/ 5-fold cross-validation basin splits (cv1–cv5)- data_CAMELS_AUS_v2/ Hydrometeorological forcings and catchment attributes (Fowler et al., 2025, CC-BY 4.0)- data_ROIAlign/ Single-basin example (basin 216002) for ROIAlign processing- models/ Trained model weights for all 4 experiments (96 files total)- results/ Model predictions and evaluation metrics (generated on 2026-03-26)- ROIAlign.ipynb Mask-aware ROIAlign processing demo- test_*.ipynb Four one-click executable notebooks that reproduce all results reported in the paper Note on satellite embeddings (data_embeddings/):The processed 32×32×64 embeddings derived from the Google AlphaEarth Foundations dataset are not included due to data redistribution restrictions. To obtain the processed embeddings, please contact the author at 12431320@mail.sustech.edu.cn. Alternatively, users may generate the embeddings independently using the provided ROIAlign.ipynb with data accessed via Google Earth Engine (https://developers.google.com/earth-engine/datasets/catalog/GOOGLE_SATELLITE_EMBEDDING_V1_ANNUAL).
