Google Earth Engine: Difference in Area of Feature Collection when calculating using a reducer vs. when calculating using clipping

Question

I am wondering why the two approaches to calculate the area of a feature collection return different results and which one is "correct"?

1. I clip the feature collection on a white image and sum up over the multiplied area

    var whiteImage = ee.Image().unmask().eq(0).rename('test')
    var trueAreaImage = whiteImageClipped.multiply(ee.Image.pixelArea());
    var trueArea = trueAreaImage.reduceRegion({
      reducer: ee.Reducer.sum(),
      geometry: validationRegion,
      scale: 10
    });

2. The other time I am using a reducer to generate the raster data and sum up over the multiplied area

    var reducedImage = groundTruth.reduceToImage(['test'],'anyNonZero').rename('test')
    var trueAreaImage = reducedImage.multiply(ee.Image.pixelArea());
    var trueArea = trueAreaImage.reduceRegion({
      reducer: ee.Reducer.sum(),
      geometry: validationRegion,
      scale: 10
    });

The result of the two approaches is significantly different.

https://code.earthengine.google.com/8baa2a6c27490ca44d7904f7ab6208be

Answer 1

The conversion of vector to raster is the likely source of the difference. I'm concluding this based on three other ways of calculating image region area by reduction which produce equal results and none of them use vector to raster conversion.

Code Editor link

// ROI
var roi = ee.Geometry.Polygon(
  [[[-122.18794270981743, 37.238143748045786],
    [-122.18794270981743, 37.02574255804269],
    [-121.82814046372368, 37.02574255804269],
    [-121.82814046372368, 37.238143748045786]]], null, false);

// Get a pixel area image.
var areaImg = ee.Image.pixelArea();

// Calculate area of ROI by image reduction (NO CLIP).
var roiAreaNoClip = areaImg.reduceRegion({
  reducer: ee.Reducer.sum(),
  geometry: roi,
  scale: 10
});

// Calculate area of ROI by image reduction (CLIP to ROI).
var roiAreaClip = areaImg.clip(roi).reduceRegion({
  reducer: ee.Reducer.sum(),
  geometry: roi,
  scale: 10
});


// Multiply area image by constant 1 image (CLIP to ROI).
var roiAreaMultiplyClip = ee.Image.constant(1).clip(roi).multiply(areaImg)
  .reduceRegion({
    reducer: ee.Reducer.sum(),
    geometry: roi,
    scale: 10
  });

print('Compute area by reduction:', roiAreaNoClip.get('area'));
print('Clip image then compute area by reduction:', roiAreaClip.get('area'));
print('Multiply area image by constant 1 image (CLIP to ROI):', roiAreaMultiplyClip.get('constant'));

Compute area by reduction:
753633057.7863889

Clip image then compute area by reduction:
753633057.7863889

Multiply area image by constant 1 image (CLIP to ROI):
753633057.7863889

It seems like this is an unnecessary step in your example:

var trueAreaImage = reducedImage.multiply(ee.Image.pixelArea());

It could just be:

var area = ee.Image.pixelArea().reduceRegion({
  reducer: ee.Reducer.sum(),
  geometry: <THE COLLECTION USED TO CREATE THE reducedImage VAR>,
  scale: 10
});