Overestimation of Brazilian Amazon tree loss in Global Forest Change (Hansen et al.)

Question

I am trying to compute the Forest Loss by Year using the Global Forest Change (Hansen et al.). I have computed it using Engine Earth and Rasterstats (Python) and the estimates I obtain are a too high when compared with other sources (official statistics). Specifically, The Hansen dataset shows a large increase in tree loss during the year 2016. Also the loss during the last ten years 2011-2020 is estimated to be around 300 000 km2 in the Hansen dataset whereas it is around 100 000 in other sources. I include the code (Earth Engine) I am using to compute the estimates.

Can anyone provide any insights on this?

    var gfc2020 = ee.Image("UMD/hansen/global_forest_change_2020_v1_8"),
        brazil = ee.FeatureCollection("users/milfadic/brazil_muni");

    var lossYear = gfc2020.select(['lossyear']);
    var muni = brazil_simp.select(['fid']);
    var lossAreaImage = lossYear.multiply(ee.Image.pixelArea());

    var lossByYearArea = lossYear.addBands(lossYear).reduceRegions({
      collection: brazil,
      reducer: ee.Reducer.sum().group({
        groupField: 1
        }),
      scale: 30,
    });

    Export.table.toDrive({
      collection: lossByYearArea,
      description: 'forest_lossAreaImage _all_muni',
      folder: 'Brazil_muni',
      fileFormat: 'CSV'
    });   

Answer 1

The lossyear band is a number between 1 and 20 indicating in what year the treecover loss was detected. You should multiply a binary mask of loss by .pixelArea(), not a year of loss. Even though you are not using that further in your script, it just important to mention.

If you want to calculate loss per year, this can be a useful starting point: https://code.earthengine.google.com/27993efe7ce32bd591e81246b76199db

Also remember that the forest definition might be different in different countries (often we use 15% or 30% threshold for treecover).