Array Sorting and Reducing

Array sorting is useful for obtaining custom quality mosaics which involve reducing a subset of image bands according to the values in a different band. The following example sorts by NDVI, then gets the mean of a subset of observations in the collection with the highest NDVI values:

Code Editor (JavaScript)

// Define a function that scales and masks Landsat 8 surface reflectance images
// and adds an NDVI band.
function prepSrL8(image) {
  // Develop masks for unwanted pixels (fill, cloud, cloud shadow).
  var qaMask = image.select('QA_PIXEL').bitwiseAnd(parseInt('11111', 2)).eq(0);
  var saturationMask = image.select('QA_RADSAT').eq(0);

  // Apply the scaling factors to the appropriate bands.
  var opticalBands = image.select('SR_B.').multiply(0.0000275).add(-0.2);
  var thermalBands = image.select('ST_B.*').multiply(0.00341802).add(149.0);

  // Calculate NDVI.
  var ndvi = opticalBands.normalizedDifference(['SR_B5', 'SR_B4'])
      .rename('NDVI');

  // Replace original bands with scaled bands, add NDVI band, and apply masks.
  return image.addBands(opticalBands, null, true)
      .addBands(thermalBands, null, true)
      .addBands(ndvi)
      .updateMask(qaMask)
      .updateMask(saturationMask);
}

// Define an arbitrary region of interest as a point.
var roi = ee.Geometry.Point(-122.26032, 37.87187);

// Load a Landsat 8 surface reflectance collection.
var collection = ee.ImageCollection('LANDSAT/LC08/C02/T1_L2')
  // Filter to get only imagery at a point of interest.
  .filterBounds(roi)
  // Filter to get only six months of data.
  .filterDate('2021-01-01', '2021-07-01')
  // Prepare images by mapping the prepSrL8 function over the collection.
  .map(prepSrL8)
  // Select the bands of interest to avoid taking up unneeded memory.
  .select('SR_B.|NDVI');

// Convert the collection to an array.
var array = collection.toArray();

// Label of the axes.
var imageAxis = 0;
var bandAxis = 1;

// Get the NDVI slice and the bands of interest.
var bandNames = collection.first().bandNames();
var bands = array.arraySlice(bandAxis, 0, bandNames.length());
var ndvi = array.arraySlice(bandAxis, -1);

// Sort by descending NDVI.
var sorted = bands.arraySort(ndvi.multiply(-1));

// Get the highest 20% NDVI observations per pixel.
var numImages = sorted.arrayLength(imageAxis).multiply(0.2).int();
var highestNdvi = sorted.arraySlice(imageAxis, 0, numImages);

// Get the mean of the highest 20% NDVI observations by reducing
// along the image axis.
var mean = highestNdvi.arrayReduce({
  reducer: ee.Reducer.mean(),
  axes: [imageAxis]
});

// Turn the reduced array image into a multi-band image for display.
var meanImage = mean.arrayProject([bandAxis]).arrayFlatten([bandNames]);
Map.centerObject(roi, 12);
Map.addLayer(meanImage, {bands: ['SR_B6', 'SR_B5', 'SR_B4'], min: 0, max: 0.4});

Python setup

See the Python Environment page for information on the Python API and using geemap for interactive development.

import ee
import geemap.core as geemap

Colab (Python)

# Define a function that scales and masks Landsat 8 surface reflectance images
# and adds an NDVI band.
def prep_sr_l8(image):
  # Develop masks for unwanted pixels (fill, cloud, cloud shadow).
  qa_mask = image.select('QA_PIXEL').bitwiseAnd(int('11111', 2)).eq(0)
  saturation_mask = image.select('QA_RADSAT').eq(0)

  # Apply the scaling factors to the appropriate bands.
  optical_bands = image.select('SR_B.').multiply(0.0000275).add(-0.2)
  thermal_bands = image.select('ST_B.*').multiply(0.00341802).add(149.0)

  # Calculate NDVI.
  ndvi = optical_bands.normalizedDifference(['SR_B5', 'SR_B4']).rename('NDVI')

  # Replace the original bands with the scaled ones and apply the masks.
  return (
      image.addBands(optical_bands, None, True)
      .addBands(thermal_bands, None, True)
      .addBands(ndvi)
      .updateMask(qa_mask)
      .updateMask(saturation_mask)
  )


# Define an arbitrary region of interest as a point.
roi = ee.Geometry.Point(-122.26032, 37.87187)

# Load a Landsat 8 surface reflectance collection.
collection = (
    ee.ImageCollection('LANDSAT/LC08/C02/T1_L2')
    # Filter to get only imagery at a point of interest.
    .filterBounds(roi)
    # Filter to get only six months of data.
    .filterDate('2021-01-01', '2021-07-01')
    # Prepare images by mapping the prep_sr_l8 function over the collection.
    .map(prep_sr_l8)
    # Select the bands of interest to avoid taking up unneeded memory.
    .select('SR_B.|NDVI')
)

# Convert the collection to an array.
array = collection.toArray()

# Label of the axes.
image_axis = 0
band_axis = 1

# Get the NDVI slice and the bands of interest.
band_names = collection.first().bandNames()
bands = array.arraySlice(band_axis, 0, band_names.length())
ndvi = array.arraySlice(band_axis, -1)

# Sort by descending NDVI.
sorted = bands.arraySort(ndvi.multiply(-1))

# Get the highest 20% NDVI observations per pixel.
num_images = sorted.arrayLength(image_axis).multiply(0.2).int()
highest_ndvi = sorted.arraySlice(image_axis, 0, num_images)

# Get the mean of the highest 20% NDVI observations by reducing
# along the image axis.
mean = highest_ndvi.arrayReduce(reducer=ee.Reducer.mean(), axes=[image_axis])

# Turn the reduced array image into a multi-band image for display.
mean_image = mean.arrayProject([band_axis]).arrayFlatten([band_names])
m = geemap.Map()
m.center_object(roi, 12)
m.add_layer(
    mean_image, {'bands': ['SR_B6', 'SR_B5', 'SR_B4'], 'min': 0, 'max': 0.4}
)
m

As in the linear modeling example, separate the bands of interest from the sort index (NDVI) using arraySlice() along the band axis. Then sort the bands of interest by sort index using arraySort(). After the pixels have been sorted by descending NDVI, use arraySlice() along the imageAxis to get 20% of the highest NDVI pixels. Lastly, apply arrayReduce() along the imageAxis with a mean reducer to get the mean of the highest NDVI pixels. The final step converts the array image back to a multi-band image for display.