How to create a list with the classes values in a raster?

Question

Let's say that I have a hypothetical study region and I just want to create a list object with all of the LULC classes within that region. How can a get it?

I got, for example, a MODIS LULC:

var ecoregions = ee.FeatureCollection('RESOLVE/ECOREGIONS/2017');
var caatinga = ecoregions.filterMetadata('ECO_NAME', 'equals', 'Caatinga')

var cover = ee.Image('MODIS/051/MCD12Q1/2012_01_01').select('Land_Cover_Type_1')
.clip(caatinga.geometry());

Map.addLayer(cover, {0: 15}, 'cover');
Map.centerObject(cover, 5.5)

var list = ee.List(cover.get('Land_Cover_Type_1'))

print(list)

But all that I get is a "null" information

I want to use this information further to calculate the area of each class within that region. But usually, I just create a sequence with all putative possible values like:

var classes = ee.List.sequence(1, 15)
print(classes)

An object like classes works just fine. However, not all classes between 1 and 15 are necessarily present and the outcome that I get is a table with empty columns for the inexistent classes. That wouldn't be a problem if I'm working with only a few empty classes, but in certain situations, I would like the outcome to have only the actually present classes without the burden to remove empty columns afterwards.

I feel like it should be something quite simple, but I couldn't figure it out yet. Any advice?

Answer 1

You can for instance reduce your region with a histogram reducer.

var histogram = ee.Array(
  cover
    .reduceRegion({
      reducer: ee.Reducer.fixedHistogram(1, 16, 15), 
      geometry: caatinga.geometry(), 
      scale: 100000, // Running this in a large scale to get some empty classes
      maxPixels: 1e13
    })
    .get('Land_Cover_Type_1')
)
var mask = histogram
  .slice(1, 1) // Extract second column - the counts
  .gt(0)
var classes = ee.List(
  histogram
    .mask(mask) // Remove rows where the count isn't > 0
    .slice(1, 0, 1) //  Extract the first column - the classes
    .project([0]) // Turn a 2D array into 1D
)

https://code.earthengine.google.com/0282d6af91be2ff05b20d9aa094c83b1

This is quite expensive at lower scales. If you have an option to use a static list of classes and add a step where you remove the empty "columns", it might be more efficient. But there's not enough information about the rest of your processing to judge if it makes sense or not.