links-ads/insar-regional-snow-mapping

---
license: mit
task_categories:
- image-segmentation
tags:
- remote-sensing
- InSAR
- snow
- SAR
- Sentinel-1
- geospatial
- regression
- Italian-Alps
pretty_name: InSAR Regional Snow Mapping (Dataset_icip2026)
size_categories:
- 1K<n<10K
---

# InSAR Regional Snow Mapping Dataset

<!-- Dataset associated with the paper: -->

<!-- > **[Paper title]**
> Luca Barco, Lorenzo Innocenti, Bianca Bartoli, Edoardo Arnaudo, Claudio Rossi, Paolo Garza
> *[Venue, Year]* -->

The dataset pairs Sentinel-1 InSAR stacks with IT-SNOW snow products and ERA5-derived weather
variables over the Italian Alps, for the purpose of pixel-wise regression of Snow Depth (HS) and
Snow Water Equivalent (SWE).

---

## Dataset overview

| Property | Value |
|---|---|
| Area | Italian Alps (Trentino region) |
| Bounding box | 9.80°E – 10.87°E, 45.93°N – 46.40°N |
| CRS | EPSG:4326 (WGS84) |
| Raster size | 93 × 211 pixels |
| Spatial resolution | ~500 m (≈ 0.005°) |
| Time windows | 64 (12-day Sentinel-1 repeat cycle) |
| Date range | January 2021 – March 2023 |
| InSAR bands per window | 7 |
| Weather bands per window | 2 |
| Targets | HS (snow depth, cm) · SWE (snow water equivalent) |

---

## Archive structure

The dataset is distributed as a single archive, `data.tar.gz`:

```
Dataset_icip2026/
├── INSAR/ # 64 × 7 = 448 GeoTIFF files (+ filtered raw intensity)
│ ├── YYYY-MM-DD_YYYY-MM-DD_coh_IW1_VV_*.tif
│ ├── YYYY-MM-DD_YYYY-MM-DD_displacement_VV.tif
│ ├── YYYY-MM-DD_YYYY-MM-DD_elevation_VV.tif
│ ├── YYYY-MM-DD_YYYY-MM-DD_Intensity_ifg_IW1_VV_*_db.tif
│ ├── YYYY-MM-DD_YYYY-MM-DD_localIncidenceAngle.tif
│ ├── YYYY-MM-DD_YYYY-MM-DD_Phase_ifg_IW1_VV_*.tif
│ └── YYYY-MM-DD_YYYY-MM-DD_Unw_Phase_ifg_*_VV.tif
├── ITSNOW/ # IT-SNOW snow products
│ ├── HS/ # 64 GeoTIFFs – YYYY-MM-DD_to_YYYY-MM-DD_hs.tif
│ └── SWE/ # 64 GeoTIFFs – YYYY-MM-DD_to_YYYY-MM-DD_swe.tif
├── S3/ # Weather rasters
│ ├── YYYY-MM-DD_YYYY-MM-DD_meanTemp.tif (mean air temperature, °C)
│ └── YYYY-MM-DD_YYYY-MM-DD_pdd.tif (positive degree days)
└── master_mask.tif # Binary valid-area mask (1 = valid, 0 = excluded)
```

---

## Data fields

### INSAR/

Seven Sentinel-1 IW-mode, VV-polarisation interferometric products, one raster per band per 12-day
window. All rasters are float32, co-registered to the same 93×211 WGS84 grid. Values outside the
valid area and known no-data sentinels (≤ −9000, 0) are NaN.

| Band | Description |
|---|---|
| `coh_IW1_VV` | Interferometric coherence (0–1) |
| `displacement_VV` | Line-of-sight surface displacement (m) |
| `elevation_VV` | Digital elevation model co-registered to SAR geometry (m) |
| `Intensity_ifg_IW1_VV_*_db` | Interferometric intensity in decibels (dB) |
| `localIncidenceAngle` | Local incidence angle (degrees) |
| `Phase_ifg_IW1_VV` | Wrapped interferometric phase (rad, −π to π) |
| `Unw_Phase_ifg_*_VV` | Unwrapped interferometric phase (rad) |

> The raw linear-scale intensity file (`Intensity_ifg_IW1_VV_*.tif`, without `_db`) is included
> in the archive but is excluded from training by the data-loading code.

### ITSNOW/

Interpolated IT-SNOW products resampled to the same 93×211 grid.

| Variable | Unit | Description |
|---|---|---|
| HS | cm | Snow depth |
| SWE | cm | Snow water equivalent |

### S3/

ERA5-derived weather aggregates over each 12-day InSAR window, resampled to the same grid.

| Variable | Description |
|---|---|
| `meanTemp` | Mean air temperature over the window (°C) |
| `pdd` | Positive degree days accumulated over the window |

### master_mask.tif

uint8 raster (93×211). Pixel value 1 = within the study area; 0 = excluded (glacier boundaries,
cloud shadow, or no-data regions). Normalization statistics and evaluation metrics are computed
only over masked pixels.

---

## Splits

The 64 time windows are sorted chronologically and split as follows (fixed, not randomised):

| Split | Windows | Notes |
|---|---|---|
| Train + validation | 57 | Rolling 5-train / 1-val pattern |
| Test | 7 (last) | 2022-12-30 → 2023-03-24 |

---

## Usage

Download and extract:

```bash
mkdir -p data
wget -O data/Dataset_icip2026.tar.gz \
https://huggingface.co/datasets/links-ads/insar-regional-snow-mapping/resolve/main/Dataset_icip2026.tar.gz
tar -xzf data/Dataset_icip2026.tar.gz -C data/
```

Load a single sample with the accompanying code repository:

```python
from pathlib import Path
from rsm.io import match_insar_to_targets, stack_insar_bands, stack_weather_bands, load_mask

root = Path("data/Dataset_icip2026")
samples = match_insar_to_targets(
root / "INSAR", root / "ITSNOW", root / "S3"
)
mask = load_mask(root / "master_mask.tif")

sample = samples[0]
print(sample["key"]) # e.g. '2021-01-09_to_2021-01-21'

insar, _ = stack_insar_bands(sample["band_files"]) # (93, 211, 7)
weather, _ = stack_weather_bands(sample["weather_files"]) # (93, 211, 2)
```

See the [code repository](https://github.com/links-ads/insar-regional-snow-mapping) for full
training and evaluation scripts.

---

## Source data

| Source | Description |
|---|---|
| [Copernicus Sentinel-1](https://sentinel.esa.int/web/sentinel/missions/sentinel-1) | C-band SAR, IW mode, VV polarisation; processed with SNAP and SNAPHU |
| [IT-SNOW](https://www.isprambiente.gov.it/it/progetti/suolo-e-territorio-1/it-snow) | National daily snow depth and SWE product for Italy (ISPRA) |
| ERA5 | ECMWF reanalysis used to derive mean temperature and positive degree days |

---

<!-- ## Citation

If you use this dataset, please cite the paper:

```bibtex
@inproceedings{barco2026insar,
title = {[Paper title]},
author = {Barco, Luca and Innocenti, Lorenzo and Bartoli, Bianca and
Arnaudo, Edoardo and Rossi, Claudio and Garza, Paolo},
booktitle = {[Venue]},
year = {2026}
}
``` -->

---

## License

This dataset is released under the [MIT License](LICENSE). Sentinel-1 data © ESA / Copernicus
Programme. IT-SNOW data © ISPRA. ERA5 data © ECMWF (Copernicus Climate Change Service).