aazuspan / geesharp

Pan-sharpening in the Earth Engine code editor

geeSharp

Pan-sharpen multispectral imagery in the Google Earth Engine Code Editor with one line of code:

var sharp = geeSharp.sharpen(img.select(["B4", "B3", "B2"]), img.select("B8");

Usage

Pan-sharpening

To pan-sharpen an image, separate the lower resolution multispectral bands and the higher resolution panchromatic band into two images and pass them to the geeSharp.sharpen function. For example:

// Import the geeSharp module
var geeSharp = require("users/aazuspan/geeSharp:geeSharp");

// Load an example Landsat 8 TOA image to sharpen
var img = ee.Image("LANDSAT/LC08/C01/T1_TOA/LC08_047027_20160819");

// Select the 30 m spectral bands to sharpen
var ms = img.select(["B4", "B3", "B2"]);
// Select the 15 m panchromatic band
var pan = img.select(["B8"]);

// Pan-sharpen!
var sharpened = geeSharp.sharpen(ms, pan);

By default, pansharpening in geeSharp uses the Smoothing Filter-based Intensity Modulation (SFIM) algorithm because it is fast and produces consistent, high-quality results. However, you may want to experiment with other methods. You can do that by passing an algorithm name to the sharpen function.

var method = "brovey";
var sharpened = geeSharp.sharpen(ms, pan, method);

Most sharpening functions just require the unsharpened multispectral bands and the high-resolution panchromatic band as inputs, but some algorithms (like Gram-Schmidt) may accept other parameters. You can add those parameters after the method name when calling sharpen.

// The Gram-Schmidt algorithm may require additional parameters depending on the size of your image.
var method = "GS";
var geom = ee.Geometry.Point([-122.41676185101713, 47.26851080476613]).buffer(1000);
var scale = 30;
var maxPixels = 1e13;

var sharpened = geeSharp.sharpen(ms, pan, method, geom, scale, maxPixels);

Print geeSharp.methods for a full list of supported algorithms, and see the documentation for descriptions.

Image quality assessment

Image quality metrics measure the distortion between a reference image and an image that has been modified, such as a pan-sharpened image.

// Choose a metric
var metric = "RMSE";
// Reproject the unsharpened image to the sharpened resolution
var reproj = unsharpened.resample("bicubic").reproject(sharpened.projection());
// Calculate the metric
var quality = geeSharp.quality(reproj, sharpened, metric);

Warning
Metrics are affected by spatial resolution, so when comparing unsharpened and pan-sharpened images, always resample and reproject the unsharpened image to high resolution first to ensure an accurate comparison!

Most quality metrics just require an unmodified and a modified image and return a dictionary mapping band names to metric values, but some metrics require other parameters (e.g. ERGAS requires the high and low spectral resolution) and some return a single image-wise value (e.g. RASE and ERGAS). Print geeSharp.metrics for a full list of supported metrics and see the documentation for descriptions.

geesharp's Issues

Image.reduceRegion: Provide 'geometry' parameter when aggregating over an unbounded image

Reopening a script of some months ago, now I get the error "Layer error: Image.reduceRegion: Provide 'geometry' parameter when aggregating over an unbounded image"

Thanks in advance

SFIM

• Implement the Smoothing Filter-based Intensity Modulation (SFIM) sharpening algorithm from Liu, 2000.

Fix errors in QA formulas

• RASE and ERGAS must be calculated image-wise rather than band-wise according to Vaiopoulos 2011.
• PSNR should use the max possible value of the datatype rather than the max value within the image. This isn't practical with float data types, so this should probably be removed.