Google Earth Engine – Sample Regions Excluding Masked Bands

attribute-joinsgoogle-earth-enginemaskingsamplepointsvalue-extraction

I have a set of points from around the world and I want to extract the pixel values from a set of images, giving information about the local environment, for each point. Then export the set of values as a .csv table.

My approach was to merge the images into a single multi-band image, then use SampleRegions to extract the values.

The problem is that if one of the images has a mask at the point location, the point is not included in the final results. So only points with no masked pixels are included in the output.
I would like to include all the points, even if some of the bands have masked pixels.

//1. Set points: 
var fc_points = ee.FeatureCollection([
  ee.Feature(ee.Geometry.Point([-17.5558, 65.8674]),    {ID: '1'}),
  ee.Feature(ee.Geometry.Point(-7.85187, 54.15745),     {ID: '2'}),   
  ee.Feature(ee.Geometry.Point(177.469394, -38.968075), {ID: '3'}),
  ee.Feature(ee.Geometry.Point(-78.93655, -2.2895),     {ID: '4'}),     
  ee.Feature(ee.Geometry.Point(6.11699, 61.51734),      {ID: '5'}), 
          ]);

//2. Make multiband image 
var koppen = ee.Image("users/fsn1995/Global_19862010_KG_5m")
koppen = koppen.updateMask(koppen.lte(30)); // Koppen-Geiger climate class
var mean_precipitation = ee.Image("WORLDCLIM/V1/BIO").select('bio12') // World Climate: mean annual precipitation 
var Cop_dataset = ee.Image("COPERNICUS/Landcover/100m/Proba-V-C3/Global/2019").select('discrete_classification'); //Copernicus Land cover 
var landCover = ee.Image('COPERNICUS/CORINE/V20/100m/2012').select('landcover'); //CORINE landcover

// rename bands
    var climate = koppen.rename('KG_climate').toFloat();
    var precipitation = mean_precipitation.rename('WC_mean_precipitation_mmyr');
    var landcover_Copernicus = Cop_dataset.rename('GLC_Copernicus').toFloat();
    var landcover_CORINE = landCover.rename('LC_CORINE').toFloat();

// combine into single image
var merged_image = climate
      .addBands(precipitation)
      .addBands(landcover_Copernicus)
      .addBands(landcover_CORINE)


// Sample pixel values for each point, using SampleRegions: (requires: Image, points)
var sampled = merged_image.sampleRegions({
  collection: fc_points, // is a feature collection
  properties: ['ID'], // List of column names of sample points inside feature collection   (not compulsory).
  scale:30  // spatial resolution of band or image
});

print(sampled,'Sample_Region');


Export.table.toDrive({
  collection: sampled,
  description: 'sample_example_pt',
  fileFormat: 'CSV'
});

I tried to convert masked values to -99 by adding .unmask(-99) to each of the original images. But that didn't help.

My next idea would be to sample each layer separately, then join the results based on the ID property.

Any suggestions?

Best Answer

I think that with 'sampleRegions' method is not possible to solve this issue. We need to use 'sample' method when mapping point feature collection as follows:

//1. Set points: 
var fc_points = ee.FeatureCollection([
  ee.Feature(ee.Geometry.Point([-17.5558, 65.8674]),    {ID: '1'}),
  ee.Feature(ee.Geometry.Point(-7.85187, 54.15745),     {ID: '2'}),   
  ee.Feature(ee.Geometry.Point(177.469394, -38.968075), {ID: '3'}),
  ee.Feature(ee.Geometry.Point(-78.93655, -2.2895),     {ID: '4'}),     
  ee.Feature(ee.Geometry.Point(6.11699, 61.51734),      {ID: '5'}), 
          ]);

//2. Make multiband image 
var koppen = ee.Image("users/fsn1995/Global_19862010_KG_5m")
koppen = koppen.updateMask(koppen.lte(30)); // Koppen-Geiger climate class
var mean_precipitation = ee.Image("WORLDCLIM/V1/BIO").select('bio12') // World Climate: mean annual precipitation 
var Cop_dataset = ee.Image("COPERNICUS/Landcover/100m/Proba-V-C3/Global/2019").select('discrete_classification'); //Copernicus Land cover 
var landCover = ee.Image('COPERNICUS/CORINE/V20/100m/2012').select('landcover'); //CORINE landcover

// rename bands
var climate = koppen.rename('KG_climate').toFloat();
var precipitation = mean_precipitation.rename('WC_mean_precipitation_mmyr');
var landcover_Copernicus = Cop_dataset.rename('GLC_Copernicus').toFloat();
var landcover_CORINE = landCover.rename('LC_CORINE').toFloat();

// combine into single image
var merged_image = climate
      .addBands(precipitation)
      .addBands(landcover_Copernicus)
      .addBands(landcover_CORINE)

// Sample pixel values for each point, using SampleRegions: (requires: Image, points)
var sampled = merged_image.sampleRegions({
  collection: fc_points, // is a feature collection
  properties: ['ID'], // List of column names of sample points inside feature collection   (not compulsory).
  scale:30  // spatial resolution of band or image
});

print("Sample_Region", sampled);

var sampled2 = fc_points.toList(fc_points.size()).map(function (feat) {
  
  return ee.Algorithms.If(ee.FeatureCollection(merged_image.sample(ee.Feature(feat).geometry(), 30)).first(),
                          ee.FeatureCollection(merged_image.sample(ee.Feature(feat).geometry(), 30)).first(), 
                          ee.Feature(feat).set({'GLC_Copernicus': 'null',
                                                'KG_climate': 'null',
                                                'LC_CORINE': 'null',
                                                'WC_mean_precipitation_mmyr': 'null'
                          }));
  
});

print("sampled2", sampled2);

Export.table.toDrive({
  collection: ee.FeatureCollection(sampled2),
  description: 'sample_example_pt',
  fileFormat: 'CSV'
});

After running above script in GEE code editor, the task produces a CSV file with these characteristics:

Masked pixels properties are printed as null but, you can use -99; as it was above proposed.

Related Solutions

[GIS] Google Earth Engine, Supervised Classification

I'd do something like this:

// Landsat 5 TM Raw Image Collection
var lt5 = ee.ImageCollection("LANDSAT/LT5_L1T");

// Filter by date and location (Landsat Path and Row Designation)
var lakeTiticaca_Early = lt5.filterDate('1985-01-01', '1986-01-01')
.filter(ee.Filter.eq('WRS_PATH', 2))
.filter(ee.Filter.eq('WRS_ROW', 71));

// Selecting Images with least cloud cover
var leastCloud = ee.Image(lakeTiticaca_Early.sort('CLOUD_COVER').first());

// Filter by date and location (Landsat Path and Row Designation)
var lakeTiticaca_Later = lt5.filterDate('2011-01-01', '2012-01-01')
.filter(ee.Filter.eq('WRS_PATH', 2))
.filter(ee.Filter.eq('WRS_ROW', 71));

// Selecting Images with least cloud cover
var leastCloud2 = ee.Image(lakeTiticaca_Later.sort('CLOUD_COVER').first());

// Applying MNDWI Function to Least Cloudy Images 
var mNDWI_Early = leastCloud.expression(
  '(Green - MIR) / (Green + MIR)', {
  'Green': leastCloud.select('B2'),
  'MIR': leastCloud.select('B5')
 });

var mNDWI_Later = leastCloud2.expression(
 '(Green - MIR) / (Green + MIR)', {
 'Green': leastCloud2.select('B2'),
 'MIR': leastCloud2.select('B5')
 });

// Creating an image collection from the MNDWI calculated Images 
var constant1 = ee.Image(mNDWI_Early).select([0],["mNDWI"]);
Map.addLayer(leastCloud,{bands:["B4","B5","B3"]},"before")
var constant2 = ee.Image(mNDWI_Later).select([0],["mNDWI"]);
Map.addLayer(leastCloud2,{bands:["B4","B5","B3"]},"after")

var dif = constant2.subtract(constant1)
Map.addLayer(dif,{},"dif")

var img1 = leastCloud.addBands(constant1)
var img2 = leastCloud2.addBands(constant2)

var final = ee.Image.cat(img1,img2)
Map.addLayer(final,{},"final")

//* Supervised Image Classification *//

var training = ee.FeatureCollection([water_nochange, water_gain, water_loss, land_nochange]).flatten()

var sample = final.sampleRegions(training, ["class"], 30)

var trained = ee.Classifier.cart().train(sample, "class")

var classified = final.classify(trained)
Map.addLayer(classified,{min:1, max:4, palette:["#b3b3ff","#0000e6", "#ff0000", "#009900"]},"classified")

You can access the feature collections in the following link: https://code.earthengine.google.com/3f4c2908fcf7c469357b80783f3ef26f

You should mask clouds, use another path/row to include the missing part of the lake, and make more training points. You can also try another classifier like svm, randomForest, etc.

labeling – Understanding the Difference Between sample, sampleRegions, and stratifiedSample in Google Earth Engine

sample() uses 1 region (either points or polygons) and does exhaustive sampling in that region (all pixels) unless you specify a smaller number of points. But the random sampling it does isn't optimal.

sampleRegions uses multiple regions (either points or polygons) and does exhaustive sampling in each region (all pixels). There are no options for randomness.

Unless you're just doing simple exhaustive sampling, you probably always want to use stratifiedSample as it has the most options and allows the most flexibility as you can always render your features into a "class band" even if it's just a sparse raster with a single value (using e.g.: ee.Image().paint(collection, 0))

Best Answer

Related Solutions

[GIS] Google Earth Engine, Supervised Classification

labeling – Understanding the Difference Between sample, sampleRegions, and stratifiedSample in Google Earth Engine

Related Question