azzenn4/Timor_ML4FLood

Dataset for domain shift test in Sen1Floods11 training set, this dataset is manually curated for flood and cloud occlusions test set from WorldFloodsv2 sourced from Pleiades-1A/1B (1), PlanetScope (3) and Sentinel-2 (1). Based on the paper Cross-Resolution Domain Shift: From U-Net Encoder–Decoder to the Prithvi Hybrid Model Gunawan, A. A. S., Kam, R. M., & Andrew, H.

Each samples varies in HxW pixel count, refering to process_patches.py / sliding_window_crop(). Given an image tensor of shape (C, H, W), the function scans across the spatial dimensions in steps of stride=224, extracting a 224×224 window at each position.

Edge handling : When a window would extend beyond the image boundary, the endpoint is clamped to the image edge (y_end = min(y + 224, H)), and the start is recalculated backwards from that clamped end (y_start = max(y_end - 224, 0)). This means the last patch in any row or column overlaps with the second-to-last patch rather than being smaller, every patch is guaranteed to be exactly 224×224.

Reproducibility : Each AOI (Area of interest) were hand-picked to give the best possible analysis of flood segmentation and cloud robustness, refer to EMSR507_[AOI] / include.txt that includes the geotiff suffix for testing set.

---

Sentinel-2 — AOI05

Follows the standard WorldFloods 12-band format, resampled to 10–20 m resolution. The SWIR bands (B11, B12) are the most critical for flood mapping — water strongly absorbs SWIR energy, making them the primary tool for separating open water from cloud shadows and dark surfaces.

Band	Name	Resolution
B1	Ultra Blue	60 m
B2	Blue	10 m
B3	Green	10 m
B4	Red	10 m
B5	Red Edge (VRE 1)	20 m
B6	Red Edge (VRE 2)	20 m
B7	Red Edge (VRE 3)	20 m
B8	NIR	10 m
B8A	Narrow NIR	20 m
B9	Water Vapour	60 m
B11	SWIR 1	20 m
B12	SWIR 2	20 m

---

PlanetScope — AOI02, AOI03, AOI07

A compact 4-band sensor optimized for high temporal frequency and sharp spatial detail. Without SWIR coverage, water-shadow discrimination is more limited compared to Sentinel-2.

Band	Name	Resolution
Band 1	Blue	3 m
Band 2	Green	3 m
Band 3	Red	3 m
Band 4	Near-Infrared (NIR)	3 m

---

Pleiades-1A / 1B — AOI01

A 4-band sensor delivering very high spatial resolution, frequently pan-sharpened to 0.5 m. Fine-grained features like building footprints and narrow waterways become resolvable, though no SWIR bands are available.

Band	Name	Resolution
Band 1	Blue	2 m (0.5 m pan-sharpened)
Band 2	Green	2 m (0.5 m pan-sharpened)
Band 3	Red	2 m (0.5 m pan-sharpened)
Band 4	Near-Infrared (NIR)	2 m (0.5 m pan-sharpened)

---

Citations

@article{portales-julia_global_2023,
    title = {Global flood extent segmentation in optical satellite images},
    volume = {13},
    issn = {2045-2322},
    doi = {10.1038/s41598-023-47595-7},
    number = {1},
    urldate = {2023-11-30},
    journal = {Scientific Reports},
    author = {Portalés-Julià, Enrique and Mateo-García, Gonzalo and Purcell, Cormac and Gómez-Chova, Luis},
    month = nov,
    year = {2023},
    pages = {20316},
}
@article{mateo-garcia_inorbit_2023,
    title = {In-orbit demonstration of a re-trainable machine learning payload for processing optical imagery},
    volume = {13},
    issn = {2045-2322},
    url = {https://www.nature.com/articles/s41598-023-34436-w},
    doi = {10.1038/s41598-023-34436-w},
    number = {1},
    urldate = {2023-06-27},
    journal = {Scientific Reports},
    author = {Mateo-Garcia, Gonzalo and Veitch-Michaelis, Josh and Purcell, Cormac and Longepe, Nicolas and Reid, Simon and Anlind, Alice and Bruhn, Fredrik and Parr, James and Mathieu, Pierre Philippe},
    month = jun,
    year = {2023},
    pages = {10391},
}

@article{mateo-garcia_towards_2021,
    title = {Towards global flood mapping onboard low cost satellites with machine learning},
    volume = {11},
    issn = {2045-2322},
    doi = {10.1038/s41598-021-86650-z},
    number = {1},
    urldate = {2021-04-01},
    journal = {Scientific Reports},
    author = {Mateo-Garcia, Gonzalo and Veitch-Michaelis, Joshua and Smith, Lewis and Oprea, Silviu Vlad and Schumann, Guy and Gal, Yarin and Baydin, Atılım Güneş and Backes, Dietmar},
    month = mar,
    year = {2021},
    pages = {7249},
}