|
|
--- |
|
|
license: mit |
|
|
language: |
|
|
- en |
|
|
tags: |
|
|
- agricolture |
|
|
- computer-vision |
|
|
pretty_name: sen12vts |
|
|
--- |
|
|
# SEN12VTS: Sentinel 1 and 2 Vegetation Time-Series Dataset |
|
|
|
|
|
## Overview |
|
|
The **SEN12VTS** (Sentinel-1 & Sentinel-2 Vegetation Time-Series) dataset has been created to support research on time-series analysis for vegetation indices, specifically targeting **NDVI (Normalized Difference Vegetation Index)** regression tasks. Recognizing the lack of datasets catering to this specific temporal and spatial need, SEN12VTS was developed to fill the gap with a high-quality, Europe-focused time-series dataset. |
|
|
|
|
|
## Motivation |
|
|
This dataset is part of the **GUARDIANS project**, aiming to build models for vegetation monitoring across Europe. The focus is on: |
|
|
- **Highly vegetated areas**. |
|
|
- Sampling over time to enable the reconstruction of coherent time-series for selected zones. |
|
|
|
|
|
## Dataset Description |
|
|
### Spatial and Temporal Extent |
|
|
- **Region:** Europe (approx. bounded by EPSG:4326 `(-10.5, 34.5, 31.6, 71.5)`). |
|
|
- **Temporal Coverage:** 2022–2023. |
|
|
- **Spatial Resolution:** 10 meters. |
|
|
|
|
|
### Data Sources |
|
|
1. **ESA WorldCover 2021**: Used to identify highly vegetated areas. |
|
|
2. **Sentinel-1 RTC**: Radiometrically Terrain Corrected radar imagery for ascending and descending orbits, VV and VH polarization. |
|
|
3. **Sentinel-2 L2A**: Atmospherically corrected optical imagery with 12 spectral bands and Scene Classification (SCL). |
|
|
|
|
|
### Sampling Methodology |
|
|
- Bounding boxes (bboxes) of size 512x512 pixels were sampled where **90% of pixels** corresponded to vegetation categories (WorldCover values: 10, 20, 30, 40, 90, 95). |
|
|
- Non-overlapping bboxes were selected within the Europe bounds. |
|
|
- Sentinel-1 and Sentinel-2 data were downloaded for 1,166 bboxes across the two years. |
|
|
- Final cropped tiles: **256x256 pixels**. |
|
|
|
|
|
### Dataset Statistics |
|
|
- **BBoxes (2022 only):** 36 |
|
|
- **BBoxes (2023 only):** 454 |
|
|
- **BBoxes (both years):** 676 |
|
|
- **Total Dataset Size:** ~824.89 GB |
|
|
|
|
|
## Data Structure |
|
|
Each bbox folder contains the following subfolders and files: |
|
|
``` |
|
|
bbox_number/ |
|
|
├── s1_rtc/ |
|
|
│ ├── ascending/ |
|
|
│ │ ├── s1_rtc_YYYYMMDDTHHMMSS.tif |
|
|
│ │ ├── s1_rtc_YYYYMMDDTHHMMSS.tif |
|
|
│ │ ├── ... |
|
|
│ ├── descending/ |
|
|
│ ├── s1_rtc_YYYYMMDDTHHMMSS.tif |
|
|
│ ├── s1_rtc_YYYYMMDDTHHMMSS.tif |
|
|
│ ├── ... |
|
|
├── s2/ |
|
|
│ ├── s2_YYYYMMDDTHHMMSS.tif |
|
|
│ ├── s2_YYYYMMDDTHHMMSS.tif |
|
|
│ ├── ... |
|
|
├── worldcover/ |
|
|
├── worldcover.tif |
|
|
``` |
|
|
### File Formats |
|
|
- **TIFF (Tagged Image File Format):** Used for storing all raster data with metadata and lossless compression. |
|
|
|
|
|
## Sentinel Data Details |
|
|
### Sentinel-1 RTC |
|
|
- Pre-processed for radiometric terrain correction and orthorectified to UTM zones. |
|
|
- Data stored in ascending and descending orbits to allow analysis of orbit-specific effects. |
|
|
- Each file has two bands, in this order: VV, VH |
|
|
|
|
|
### Sentinel-2 L2A |
|
|
- 12 spectral bands selected, along with a scene classification map ([SCL](https://custom-scripts.sentinel-hub.com/custom-scripts/sentinel-2/scene-classification/)). |
|
|
- Resolutions vary from **10m to 60m** (resampled to 10m). |
|
|
|
|
|
### [WorldCover](https://esa-worldcover.org/en) |
|
|
- Extracted and resampled to match bbox projections for consistency. |
|
|
|
|
|
## Example Visualization |
|
|
The dataset provides: |
|
|
- RGB composites from Sentinel-2 (red, green, and blue bands). |
|
|
- NDVI derived from Sentinel-2 data. |
|
|
- Sentinel-1 radar images for both polarizations (ascending and descending). |
|
|
- WorldCover classification maps. |
|
|
|
|
|
 |
|
|
|
|
|
|
|
|
## Citation |
|
|
If you use SEN12VTS in your research, please cite this repository and the corresponding publication. |
|
|
|
|
|
--- |
|
|
|
|
|
## Usage |
|
|
The dataset can be used for various applications, including but not limited to: |
|
|
- NDVI regression. |
|
|
- Vegetation monitoring and analysis. |
|
|
- Time-series forecasting. |
|
|
- Multi-sensor data fusion. |
|
|
|
|
|
## Licensing |
|
|
The dataset is distributed under the MIT, and its use is subject to compliance with the Sentinel and ESA WorldCover data usage policies. |
|
|
|
|
|
--- |
|
|
|
|
|
### Contact |
|
|
For questions, issues, or contributions, please open an issue on this repository or contact federico.oldani@gmail.com. |
