Calculating NDVI values using Sentinel 3 OLCI

Question

I am trying to compute NDVI values for certain polygons using SEN3 dataset, but keeps getting negative values It seems I am getting wrong values for NIR and RED bands when trying to transform SENTINEL 3 OLCI values from radiance to surface reflectance. I am assuming that is the problem.

var sigma2VisParmSEN3 = {min: 0, max: 1, palette: 'red,orange,yellow,green'};

var SEN3 = ee.ImageCollection('COPERNICUS/S3/OLCI')
                  .filterDate('2019-01-01', '2019-02-01')

// Select bands for visualization and apply band-specific scale factors.
var SEN_corr = SEN3.select(['Oa17_radiance', 'Oa08_radiance'])
              .median()
              // Convert to radiance units.
              .multiply(ee.Image([0.00493004, 0.00876539]));

print(SEN_corr);
//calculate NDVI values per region - ***please change region to a shape you can use on your device***
var ndviSEN3 = SEN_corr
      .clipToCollection(region)
      .normalizedDifference(['Oa17_radiance', 'Oa08_radiance'])
      .rename('ndviSEN3');

Map.addLayer(ndviSEN3, sigma2VisParmSEN3, 'ndvi2SigmaSEN3');
var ndviFeatures = ee.FeatureCollection(ee.List(ndviSEN3
  .reduceRegion({
    reducer: ee.Reducer.toList(),
    geometry: region,
    scale: 10
  })
  .get('ndviSEN3'))
  .map(function (ndviSEN3) {
    return ee.Feature(null, {ndviSEN3: ndviSEN3})
  })
)
print(ndviFeatures.aggregate_array('ndviSEN3'))

I get these values for ndviFeatures:

(-0.06449873745441437, -0.3388196527957916.......)

Any suggestions?

Answer 1

I was very confused about this. I did a test comparing the NDVI of a single S2 and S3 scene, and got similar results as you - NDVI for S3 was too low. When I tried to divide with the radiance scales, instead of multiplying, I got the expected NDVI:

https://code.earthengine.google.com/074c58e2e82d5e919becee24e993d28d

The EE documentation is multiplying with the scales. Maybe this is wrong and this:

// Select bands for visualization and apply band-specific scale factors.
var rgb = dataset.select(['Oa08_radiance', 'Oa06_radiance', 'Oa04_radiance'])
    .median()
    // Convert to radiance units.
    .multiply(ee.Image([0.00876539, 0.0123538, 0.0115198]));

Should be:

// Select bands for visualization and apply band-specific scale factors.
var rgb = dataset.select(['Oa08_radiance', 'Oa06_radiance', 'Oa04_radiance'])
    .median()
    // Convert to radiance units.
    .divide(ee.Image([0.00876539, 0.0123538, 0.0115198]));

I'd be happy for some feedback on this.

Note that I used the radiance scales from the image properties, instead of the generic ones in the EE catalogue. That should give more accurate results. You could map over your collection and do the same.