Hi,

Please can you share the evalscript you are using to calculate NDVI in EO Browser? Unless you are performing some scaling or histogram equalisation then you should only return the raw NDVI values as long as you are downloading the raw values and not the visualisation bands. 

That said the NDVI values will never be exactly the same as some resampling will occur in the background depending on your zoom level etc.

Hi,

Thanks for the insightful reply.

Here’s a follow-up with additional clarification on my methodology and results:

NDVI via Sentinel Hub – Batch Statistical API 

For this comparison, I did not use the EO Browser visual layer for analysis. Instead, I computed raw NDVI using Sentinel Hub’s Batch Statistical API, with the following Evalscript:

//VERSION=3
function setup() {
  return {
    input: Â{ bands: ["B04", "B08", "dataMask"] }],
    output:  { id: "ndvi", bands: 1 }, { id: "dataMask", bands: 1 }]
  };
}

function evaluatePixel(samples) {
  return {
    ndvi: rindex(samples.B08, samples.B04)],
    dataMask: /samples.dataMask]
  };

NDVI Result (Tile 37SDA – 24 August 2024):

• Mean NDVI = 0.8551

NDVI Still ~0.2 Higher Than GEE / AWS

Even with this raw, analytical output, the NDVI result from Sentinel Hub Batch API is still ~0.2 higher than the same scene processed through GEE and AWS:

 

 

So the discrepancy does not appear to be due to visual rendering or display settings.

 

EO Browser – Just for Reference

In a separate test, I also checked Sentinel Hub's default NDVI visualization layer in EO Browser, which uses a ColorRampVisualizer script. As expected, those values were even higher (visually ranging from ~0.85 to 0.93), but that’s not what my comparison is solely based on.

Conclusion

• This discrepancy might relate to differences in:
  • Resampling strategy
  • Radiometric scaling / calibration
  • Subtle differences in TOA reflectance calculations between providers

If others have encountered similar mismatches between Sentinel Hub and GEE/AWS, I’d love to hear your insights.

Thanks again!
 

 

If you can share your full request that would help us. Knowing the AOI, time range, mosaicking method, resolution etc. will help us troubleshoot. Is this mismatch only happening for this certain AOI and date or is this consistent across many examples?