Vector to Raster Interpolation

Interpolation from vector to raster in Earth Engine creates an Image from a FeatureCollection. Specifically, Earth Engine uses numeric data stored in a property of the features to interpolate values at new locations outside of the features. The interpolation results in a continuous Image of interpolated values up to the distance specified.

Inverse Distance Weighted Interpolation

The inverse distance weighting (IDW) function in Earth Engine is based on the method described by Basso et al. (1999). An additional control parameter is added in the form of a decay factor (gamma) on the inverse distance. Other parameters include the mean and standard deviation of the property to interpolate and the maximum range distance over which to interpolate. The following example creates an interpolated surface of PM 2.5 concentrations from a FeatureCollection of points representing PM 2.5 measurement stations.

// Load a Fusion Table corresponding to mean annual PM2.5 concentrations at points.
var airQualityMeasurements = ee.FeatureCollection('ft:14BLob4jGA6au2MB1cx0GhxYDvZc-lVLAWhqBAZuN');
Map.addLayer(airQualityMeasurements, {}, 'Mean annual PM2.5 concentrations (micrograms/m^3)');

// This is the name of the property to interpolate.
var propertyToInterpolate = 'ArithmeticMean';

// Combine mean and SD reducers for efficiency.
var combinedReducer = ee.Reducer.mean().combine({
    reducer2: ee.Reducer.stdDev(),
    sharedInputs: true
});

// Estimate global mean and standard deviation (SD) from the points.
var stats = airQualityMeasurements.reduceColumns({
  reducer: combinedReducer,
  selectors: [propertyToInterpolate]
});

// Do the interpolation, valid to 50 kilometers.
var interpolatedPM25 = airQualityMeasurements.inverseDistance({
  range: 50 * 1000,
  propertyName: propertyToInterpolate,
  mean: stats.get('mean'),
  stdDev: stats.get('stdDev'),
  gamma: 0.5
});

// Visualize the resulting interpolated raster.
var vis = {min: 0, max: 15, palette: ['blue', 'green', 'red']};
Map.setCenter(-121.7944, 36.9235, 7);
Map.addLayer(interpolatedPM25, vis, 'Interpolated PM2.5 concentration');
    

Note that, as specified by the range parameter, the interpolation only exists up to 50 kilometers from the nearest measurement station.

Kriging

Kriging is an interpolation method that uses a modeled estimate of semi-variance to create an image of interpolated values that is an optimal combination of the values at known locations. The Kriging estimator requires parameters that describe the shape of a semi-variogram fit to the known data points. These parameters are illustrated by Figure 1.

Figure 1. The nugget, sill and range parameters illustrated on a idealized variogram function.

The following example samples a sea surface temperature (SST) image at random locations, then interpolates SST from the sample using Kriging:

// Load an image of sea surface temperature (SST).
var sst = ee.Image('NOAA/AVHRR_Pathfinder_V52_L3/20120802025048')
  .select('sea_surface_temperature')
  .rename('sst')
  .divide(100);

// Define a geometry in which to sample points
var geometry = ee.Geometry.Rectangle([-65.60, 31.75, -52.18, 43.12]);

// Sample the SST image at 1000 random locations.
var samples = sst.addBands(ee.Image.pixelLonLat())
  .sample({region: geometry, numPixels: 1000})
  .map(function(sample) {
    var lat = sample.get('latitude');
    var lon = sample.get('longitude');
    var sst = sample.get('sst');
    return ee.Feature(ee.Geometry.Point([lon, lat]), {sst: sst});
  });

// Interpolate SST from the sampled points.
var interpolated = samples.kriging({
  propertyName: 'sst',
  shape: 'exponential',
  range: 100 * 1000,
  sill: 1.0,
  nugget: 0.1,
  maxDistance: 100 * 1000,
  reducer: 'mean',
});

var colors = ['00007F', '0000FF', '0074FF',
              '0DFFEA', '8CFF41', 'FFDD00',
              'FF3700', 'C30000', '790000'];
var vis = {min:-3, max:40, palette: colors};

Map.setCenter(-60.029, 36.457, 5);
Map.addLayer(interpolated, vis, 'Interpolated');
Map.addLayer(sst, vis, 'Raw SST');
Map.addLayer(samples, {}, 'Samples', false);
    

The size of the neighborhood in which to perform the interpolation is specified by the maxDistance parameter. Larger sizes will result in smoother output but slower computations.