SARLAND

SARLAND = SARLO-80 SAR crops + ESA WorldCover 10 m land cover, aligned to the SAR frame.

This dataset reuses the exact SAR samples from ONERA/SARLO-80 (complex UMBRA SAR crops at ~80 cm + SICD metadata + captions) and adds, for every sample, a 10 m land cover layer, reprojected so that it overlays pixel-for-pixel on the SAR crop using the optical→SAR transform derived from the SICD.

---

Land cover classes

Every land cover pixel uses one of the 11 ESA WorldCover class codes below. The same colors are used throughout the example figures.

Code	Class	RGB
10	Tree cover	(0, 100, 0)
20	Shrubland	(255, 187, 34)
30	Grassland	(255, 255, 76)
40	Cropland	(240, 150, 255)
50	Built-up	(250, 0, 0)
60	Bare / sparse vegetation	(180, 180, 180)
70	Snow and ice	(240, 240, 240)
80	Permanent water bodies	(0, 100, 200)
90	Herbaceous wetland	(0, 150, 160)
95	Mangroves	(0, 207, 117)
100	Moss and lichen	(250, 230, 160)

---

Examples

Each example shows the SAR amplitude (left) and the aligned land cover (right), with the classes present in that crop listed below the figure.

Example 1 — sample 00000062
Classes: Tree cover, Shrubland, Grassland, Cropland, Built-up, Bare / sparse vegetation, Permanent water bodies, Herbaceous wetland

Example 2 — sample 00000047
Classes: Tree cover, Shrubland, Grassland, Cropland, Built-up, Bare / sparse vegetation, Permanent water bodies, Herbaceous wetland

Example 3 — sample 00000017
Classes: Tree cover, Shrubland, Grassland, Cropland, Built-up, Bare / sparse vegetation, Permanent water bodies, Herbaceous wetland

Example 4 — sample 00000056
Classes: Tree cover, Shrubland, Grassland, Cropland, Built-up, Bare / sparse vegetation, Permanent water bodies, Herbaceous wetland

Example 5 — sample 00000092
Classes: Tree cover, Shrubland, Grassland, Cropland, Built-up, Bare / sparse vegetation, Permanent water bodies, Herbaceous wetland

Example 6 — sample 00000158
Classes: Tree cover, Shrubland, Grassland, Cropland, Built-up, Bare / sparse vegetation, Permanent water bodies, Herbaceous wetland

Example 7 — sample 00000013
Classes: Tree cover, Shrubland, Grassland, Built-up, Bare / sparse vegetation, Permanent water bodies, Herbaceous wetland

---

Structure

WebDataset format, organized into chunks and shards:

train/
  chunk_000/
    shard-00000.tar
    shard-00001.tar
    ...
  chunk_001/
    ...

Each sample (shared key {id}, an 8-digit index) contains:

Member	Content
{id}.sar.npy	Complex SAR (SLC), identical to SARLO-80
{id}.sar.png	SAR amplitude (slant range), uint8
{id}.sicd.xml	SICD metadata (acquisition geometry)
{id}.meta.json	Metadata (crop, projection, captions, incidence angles…)
{id}.landcover.npy	New. (2, H, W) uint8 = [label, mask], aligned to the SAR frame
{id}.landcover.png	New. Colorized land cover, RGBA (alpha = validity mask)

• label: per-pixel ESA WorldCover class code (see legend above), on the SAR crop pixel grid (~80 cm).
• mask: 1 = valid land cover pixel, 0 = invalid / outside footprint.
• The land cover is resampled onto the SAR grid via an affine warp (INTER_NEAREST), which keeps crisp class boundaries.

---

Loading

import io, json
import numpy as np
from PIL import Image
from huggingface_hub import hf_hub_download
import webdataset as wds

local_tar = hf_hub_download(
    repo_id="SoleneDEBUYSERE/SARLAND",
    repo_type="dataset",
    filename="train/chunk_000/shard-00000.tar",
)

ds = wds.WebDataset(local_tar, shardshuffle=False)
sample = next(iter(ds))

sar_complex = np.load(io.BytesIO(sample["sar.npy"]), allow_pickle=False)
sar_amp     = np.asarray(Image.open(io.BytesIO(sample["sar.png"])).convert("L"))
meta        = json.loads(sample["meta.json"].decode("utf-8"))

# Land cover aligned to the SAR frame
lc = np.load(io.BytesIO(sample["landcover.npy"]), allow_pickle=False)  # (2, H, W)
label, mask = lc[0], lc[1]
lc_rgba = np.asarray(Image.open(io.BytesIO(sample["landcover.png"])).convert("RGBA"))

print("SAR amplitude:", sar_amp.shape)
print("Land cover   :", label.shape, "classes:", np.unique(label[mask > 0]))

Plotting SAR + land cover side by side

import numpy as np
import matplotlib.pyplot as plt
from matplotlib.patches import Patch

WORLDCOVER_LABELS = {
    10: "Tree cover", 20: "Shrubland", 30: "Grassland", 40: "Cropland",
    50: "Built-up", 60: "Bare / sparse vegetation", 70: "Snow and ice",
    80: "Permanent water bodies", 90: "Herbaceous wetland",
    95: "Mangroves", 100: "Moss and lichen",
}
WORLDCOVER_RGB = {
    10: (0, 100, 0), 20: (255, 187, 34), 30: (255, 255, 76), 40: (240, 150, 255),
    50: (250, 0, 0), 60: (180, 180, 180), 70: (240, 240, 240), 80: (0, 100, 200),
    90: (0, 150, 160), 95: (0, 207, 117), 100: (250, 230, 160),
}

amp = sar_amp.astype(np.float32)
p1, p99 = np.percentile(amp, [1, 99])
amp = np.clip((amp - p1) / (p99 - p1 + 1e-6), 0, 1)
valid = mask > 0

fig, axes = plt.subplots(1, 2, figsize=(11, 6))
axes[0].imshow(amp, cmap="gray");  axes[0].set_title("SAR amplitude")
axes[1].imshow(lc_rgba);           axes[1].set_title("Land cover (ESA WorldCover)")
for ax in axes: ax.axis("off")

present = [c for c in WORLDCOVER_LABELS if np.any(label[valid] == c)]
handles = [Patch(facecolor=np.array(WORLDCOVER_RGB[c]) / 255.0, edgecolor="k",
                 label=f"{c} — {WORLDCOVER_LABELS[c]}") for c in present]
fig.legend(handles=handles, loc="lower center", ncol=min(4, len(handles)),
           bbox_to_anchor=(0.5, -0.02))
fig.subplots_adjust(bottom=0.16)
fig.savefig("sarland_example.png", dpi=200, bbox_inches="tight")

---

Licenses and attribution

Dataset — components have distinct licenses:

• Land cover (ESA WorldCover 10 m, 2020 v100): © ESA WorldCover project, produced by a consortium led by VITO. Distributed under CC BY 4.0. Please cite the ESA WorldCover product.
• SAR: imagettes derived from UMBRA data; refer to the UMBRA terms of use and to the source dataset ONERA/SARLO-80.

If you use this dataset, please cite the ESA WorldCover and UMBRA sources, as well as this dataset. If you use this dataset, please cite the ESA WorldCover and UMBRA sources, as well as the source dataset ONERA/SARLO-80.

Until the dedicated SARLAND publication is available, please cite [ONERA/SARLO-80](https://arxiv.org/abs/2606.20523) as the reference dataset for the SAR imagery.

Acknowledgments

The authors gratefully acknowledge DEMR-ONERA for making available the computational resources required for the generation of this dataset. We especially thank Nicolas Trouvé and his team for their support.