SanuthK/AgroGate_SAR2NDVI_With_TIMESERIES

Dataset Card — AgroGate_SAR2NDVI_With_TIMESERIES

Dataset Summary

AgroGate_SAR2NDVI_With_TIMESERIES is a geospatial dataset containing paired SAR and NDVI satellite images of paddy fields collected across multiple growth stages within a single cultivation season. Each SAR–NDVI pair is captured as close in time as possible (±7 days) to ensure accurate spectral correspondence, and several such pairs are included per paddy lifecycle (typically 3–4 months).

The dataset supports time-series analysis of crop dynamics, SAR-optical fusion, and digital twin applications in precision agriculture.

Developed as part of the AgroGate Project, which builds a digital twin of Sri Lanka’s agricultural landscape using AI and remote sensing.

---

Dataset Structure

Root directory:

AgroGate_SAR2NDVI_With_TIMESERIES/
 └── 2023/
      ├── NDVI/
      │    ├── 2023_yala_76_stage1_NDVI.tif
      │    ├── 2023_yala_76_stage2_NDVI.tif
      │    ├── 2023_yala_76_stage3_NDVI.tif
      │    ├── 2023_yala_76_stage4_NDVI.tif
      │    └── 2023_yala_76_stage5_NDVI.tif
      │
      ├── SAR/
      │    ├── 2023_yala_76_stage1_SAR.tif
      │    ├── 2023_yala_76_stage2_SAR.tif
      │    ├── 2023_yala_76_stage3_SAR.tif
      │    ├── 2023_yala_76_stage4_SAR.tif
      │    └── 2023_yala_76_stage5_SAR.tif
      │
      ├── Plots/
      │    ├── 2023_yala_76_stage1_composite.png
      │    ├── 2023_yala_76_stage2_composite.png
      │    ├── ...
      │
      └── metadata.csv

---

Data Description

Field	Type	Description
field_id	string	Identifier for the paddy field (e.g., 76)
season	string	Season name (e.g., Yala 2023)
stage	string	Growth stage name (stage1–stage5)
sar_image	GeoTIFF	Sentinel-1 SAR image of the field
ndvi_image	GeoTIFF	Sentinel-2 NDVI image of the same field
plot_image	PNG	Optional visualization combining both
acquisition_date_sar	date	Date of SAR acquisition
acquisition_date_ndvi	date	Date of NDVI acquisition
time_diff_days	integer	Temporal gap (days) between SAR and NDVI
roi_center	coordinates	Latitude/longitude of field center
resolution	int	Image resolution (10 m)
tile_size_m	float	Tile width/height (≈1280 m)

---

Example Metadata Entry

field_id,season,stage,sar_image,ndvi_image,acquisition_date_sar,acquisition_date_ndvi,time_diff_days
76,Yala_2023,stage4,2023/ SAR/2023_yala_76_stage4_SAR.tif,2023/ NDVI/2023_yala_76_stage4_NDVI.tif,2023-08-12,2023-08-10,2

---

Data Splits

Split	Description
train	70% of fields
validation	15% of fields
test	15% of fields

---

Usage Example

from datasets import load_dataset
import rasterio

dataset = load_dataset("SanuthK/AgroGate_SAR2NDVI_With_TIMESERIES")

sample = dataset["train"][0]
sar_path = sample["sar_image"]
ndvi_path = sample["ndvi_image"]

with rasterio.open(sar_path) as src:
    sar = src.read(1)

with rasterio.open(ndvi_path) as src:
    ndvi = src.read(1)

print(sample["stage"], sample["time_diff_days"])

---

Dataset Creation

🛰 Source Data

• SAR: Sentinel-1 GRD (VV/VH polarization)

• NDVI: Derived from Sentinel-2 Level-2A (10 m bands)

  • NDVI = (B8 - B4) / (B8 + B4)
  • Cloud-masked using Sentinel-2 Scene Classification Layer (SCL)

🌾 Temporal Pairing

• SAR and NDVI images paired within ±7 days

• Each field includes 4–5 such pairs, spaced ~3–4 weeks apart

• Each pair represents a growth stage:

  • Stage 1: Land preparation
  • Stage 2: Early vegetative
  • Stage 3: Reproductive
  • Stage 4: Ripening
  • Stage 5: Harvest

🗺 Spatial Tile Extraction

• Centered on identified paddy field locations
• Circular buffer of 640 m radius (≈1.28 km tile)
• Clipped to bounding box and exported at 10 m scale

---

License

CC-BY 4.0 — You may share and adapt with attribution.

---

Citation

@dataset{agrogate_sar2ndvi_with_timeseries_2025,
  title = {AgroGate_SAR2NDVI_With_TIMESERIES},
  author = {Wanniarachchi S. K. and Weerakoon D.S.K},
  year = {2025},
  publisher = {Hugging Face Datasets},
  license = {CC-BY 4.0},
  url = {https://huggingface.co/datasets/SanuthK/AgroGate_SAR2NDVI_With_TIMESERIES}
}

---

Acknowledgements

Developed under the AgroGate Project, an initiative by AgroGate Team to create a no-code, AI-powered digital twin platform for Sri Lankan agriculture. Built using Sentinel-1 and Sentinel-2 imagery, this dataset forms the foundation for modeling paddy growth stages and supporting sustainable farming decisions.

---