Model Overview
Satellite Data: Sentinel-2
Optical satellite imagery is an important resource for agricultural monitoring. However, cloud cover can limit the availability of usable optical data and interrupt temporal monitoring for applications such as crop health monitoring, crop classification, and yield estimation, particularly across tropical and temperate regions.
Synthetic Aperture Radar (SAR) provides complementary information because it can observe the Earth regardless of atmospheric conditions.
The Cloud Gap Fill NDVI model combines contextual information from optical and SAR data to reconstruct cloud-affected NDVI values. It uses a deep learning architecture to fuse Sentinel-2 optical data with Sentinel-1 SAR data and reconstruct missing NDVI values using historical temporal dependencies.
The reconstructed NDVI enables more continuous assessment of vegetation status and can provide information related to vegetation health, phenological development, and land cover dynamics.
Model Inputs
The following input is required:
| Input Name | Description |
|---|---|
| Raster | Sentinel-2 raster image. |
Model Output
The model reconstructs NDVI values at cloud-masked locations within the selected Area of Interest and generates the resulting raster file.
Output Files
| S.No. | Output | Description | Download output as |
|---|---|---|---|
| 1 | Reconstructed NDVI | Reconstructed NDVI raster with values filled at cloud-masked locations in the AOI using fused Sentinel-1 and Sentinel-2 temporal information. | GeoTIFF |
Estimated Model Run Time
| AOI Size Range (Sq KMs) | Estimated Model Run Time* |
|---|---|
| 5 to 100 | 45 minutes |
| 100 to 250 | 2 hours |
| 250 to 500 | 4 hours 30 minutes |
| 500 to 1000 | 9 hours |
| 1000 to 2000 | 11 hours 15 minutes |
| 2000 to 3000 | 21 hours |
| 3000 to 5000 | 36 hours |
*Actual time may vary.
Additional Details
| Parameter | Details |
|---|---|
| Minimum AOI Size | 5 sq km |
| Maximum AOI Size | 5000 sq km |
| Geographies Supported | All geographies |
| Sensors Supported | Sentinel-2 |
Limitations
Sentinel-1 data should be available for the selected date or within ±2 days of the selected date.
The model currently works for the NDVI index.
Model Accuracy
The model has been tested across different geographical locations in the UK, USA, Africa, and India.
The average R² value is 0.75, which varies depending on the temporal extent of cloud cover.