Raster to Vector Conversion

To convert from an Image (raster) to a FeatureCollection (vector) data type, use image.reduceToVectors(). This is the primary mechanism for vectorization in Earth Engine, and can be useful for generating regions for input to other types of reducer. The reduceToVectors() method creates polygon edges (optionally centroids or bounding boxes instead) at the boundary of homogeneous groups of connected pixels.

For example, consider a 2012 nightlights image of Japan. Let the nightlights digital number serve as a proxy for development intensity. Define zones using arbitrary thresholds on the nightlights, combine the zones into a single-band image, vectorize the zones using reduceToVectors():

Code Editor (JavaScript)

// Load a Japan boundary from the Large Scale International Boundary dataset.
var japan = ee.FeatureCollection('USDOS/LSIB_SIMPLE/2017')
  .filter(ee.Filter.eq('country_na', 'Japan'));

// Load a 2012 nightlights image, clipped to the Japan border.
var nl2012 = ee.Image('NOAA/DMSP-OLS/NIGHTTIME_LIGHTS/F182012')
  .select('stable_lights')
  .clipToCollection(japan);

// Define arbitrary thresholds on the 6-bit nightlights image.
var zones = nl2012.gt(30).add(nl2012.gt(55)).add(nl2012.gt(62));
zones = zones.updateMask(zones.neq(0));

// Convert the zones of the thresholded nightlights to vectors.
var vectors = zones.addBands(nl2012).reduceToVectors({
  geometry: japan,
  crs: nl2012.projection(),
  scale: 1000,
  geometryType: 'polygon',
  eightConnected: false,
  labelProperty: 'zone',
  reducer: ee.Reducer.mean()
});

// Display the thresholds.
Map.setCenter(139.6225, 35.712, 9);
Map.addLayer(zones, {min: 1, max: 3, palette: ['0000FF', '00FF00', 'FF0000']}, 'raster');

// Make a display image for the vectors, add it to the map.
var display = ee.Image(0).updateMask(0).paint(vectors, '000000', 3);
Map.addLayer(display, {palette: '000000'}, 'vectors');

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)

# Load a Japan boundary from the Large Scale International Boundary dataset.
japan = ee.FeatureCollection('USDOS/LSIB_SIMPLE/2017').filter(
    ee.Filter.eq('country_na', 'Japan')
)

# Load a 2012 nightlights image, clipped to the Japan border.
nl_2012 = (
    ee.Image('NOAA/DMSP-OLS/NIGHTTIME_LIGHTS/F182012')
    .select('stable_lights')
    .clipToCollection(japan)
)

# Define arbitrary thresholds on the 6-bit nightlights image.
zones = nl_2012.gt(30).add(nl_2012.gt(55)).add(nl_2012.gt(62))
zones = zones.updateMask(zones.neq(0))

# Convert the zones of the thresholded nightlights to vectors.
vectors = zones.addBands(nl_2012).reduceToVectors(
    geometry=japan,
    crs=nl_2012.projection(),
    scale=1000,
    geometryType='polygon',
    eightConnected=False,
    labelProperty='zone',
    reducer=ee.Reducer.mean(),
)

# Display the thresholds.
m = geemap.Map()
m.set_center(139.6225, 35.712, 9)
m.add_layer(
    zones,
    {'min': 1, 'max': 3, 'palette': ['0000FF', '00FF00', 'FF0000']},
    'raster',
)

# Make a display image for the vectors, add it to the map.
display_image = ee.Image(0).updateMask(0).paint(vectors, '000000', 3)
m.add_layer(display_image, {'palette': '000000'}, 'vectors')
m

Note that the first band in the input is used to identify homogeneous regions and the remaining bands are reduced according to the provided reducer, the output of which is added as a property to the resultant vectors. The geometry parameter specifies the extent over which the vectors should be created. In general, it is good practice to specify a minimal zone over which to create vectors. It is also good practice to specify the scale and crs to avoid ambiguity. The output type is ‘polygon’ where the polygons are formed from homogeneous zones of four-connected neighbors (i.e. eightConnected is false). The last two parameters, labelProperty and reducer, specify that the output polygons should receive a property with the zone label and the mean of the nightlights band(s), respectively.

The mapped result should look something like the Tokyo area shown in Figure 1. Inspection of the output polygons indicates that each polygon has a property storing the label of the zone ({1, 2, 3}) and the mean of the nightlights band, since the mean reducer is specified.

Figure 1. Zones of nightlights in the Tokyo, Japan area. Vector boundaries are displayed in black.