Hi everyone,
I'm conducting a crop monitoring project in southeastern Turkey and comparing NDVI values across three platforms (AWS, GEE, and Sentinel Hub) using the exact same Sentinel-2 tiles and date.
During this process, I observed consistent NDVI differences across platforms, even when using the same data product level (e.g., L1C).
Scene Metadata
- Date: 24 August 2024
- Tiles: 37SDA, 37SDB, 37SEA, 37SEB
- AOI: Same polygon used on all platforms
- Bands: B04 (Red) & B08 (NIR), scaled to o0–1] as needed
NDVI Comparison (Tile 37SDA Example)
Platform | Product | NDVI Mean | NDVI Min | NDVI Max |
AWS | L1C | 0.5206 | 0.4238 | 0.5582 |
GEE | L1C | 0.52198 | 0.4475 | 0.5582 |
GEE | L2A | 0.6085 | 0.5173 | 0.6458 |
Sentinel Hub | L1C (EO Browser) | 0.74 | 0.56 | 0.80 |
Sentinel Hub | L2A (EO Browser) | 0.85 | 0.61 | 0.93 |
Key Observations
- NDVI from AWS and GEE (L1C) are almost identical (~0.52).
- Sentinel Hub L1C NDVI (via EO Browser) is ~0.74 → a large jump from other L1C sources.
- L2A consistently produces higher NDVI, but Sentinel Hub L2A is even higher than GEE-L2A.
This suggests there may be platform-specific adjustments or visual enhancement applied, especially in Sentinel Hub’s EO Browser.
Questions
- Are NDVI values in EO Browser computed from raw reflectance bands, or are they visually enhanced (e.g., stretched, clipped)?
- Could internal processing steps (e.g., histogram equalization, scaling) cause these differences?
- Is it safe to compare NDVI values across platforms (AWS, GEE, Sentinel Hub) for analysis and classification purposes?
Thanks,