Is there a way to use fusion algorithms like STARFM in Google Earth Engine? - algorithm

Question

Landsat and MODIS products both have their advantages. Landsat with its high spatial resolution and MODIS with its high temporal resolution. I've read a lot about downloading the files and fuse them locally with algorithms like STARFM in Python for example. Is there a way to fuse both collections directly in Google Earth Engine to save computational time?

Answer 1

There is Google earth engine code in GitHub. The link is below:
https://github.com/MacGallagher/GEE_STARFM_FUSION
The code I copied is below:
/**** Start of imports. If edited, may not auto-convert in the playground. ****/
var l8 = ee.ImageCollection("LANDSAT/LC08/C01/T1_SR"),
modis = ee.ImageCollection("MODIS/006/MOD09GQ"),
geometry4 = /* color: #d63000 */ee.Geometry.Polygon(
[[[-116.004638671875, 42.809506838324204],
[-115.59539794921875, 42.78129125156276],
[-115.499267578125, 43.004647127794435],
[-115.499267578125, 43.28920196020127],
[-116.00189208984375, 43.35114690203121],
[-116.19415283203125, 43.07691312608711]]]),
geometry = /* color: #98ff00 */ee.Geometry.Polygon(
[[[-115.93391418457031, 43.09346209534859],
[-115.80276489257812, 43.095718426590174],
[-115.80276489257812, 43.161366298103566],
[-115.9332275390625, 43.16086543618915]]]);
/***** End of imports. If edited, may not auto-convert in the playground. *****/
// based on script by Faye Peters, University of Louisville
// modified by Megan Gallagher, Boise State University
// for code inquiries, or adjustments please email megangallagher#u.boisestate.edu
// code exists for:
// landsat 5,7, and 8 NDVI merging with MODIS Terra daily
// MODIS Terra and Sentinel-2
// Landsat pre tier and modis daily terra - works with base R code
// Landsat 8 and Sentinel-2
////////////////////////////////////////Variables//////////////////////////////////////////////
// import Landsat 8 and MODIS daily terra surface reflectance 250m
var region = geometry; //region you want to export
var bounds = geometry4 //outer bounds of image to catch boundary effect, make larger than region
var date_begin = '2016-03-02' //start date of data collection, must be a landsat image date
var date_end ='2016-10-13' // end date of data collection, must be the day AFTER last landsat image
var csv_title = '2016_Dates_BOP' //title of csv output
var ls_title = '2016_landsat' // title of landsat tif file
var mod_title = '2016_mod' // title of modis tif file
/////////////////////////////////////////Code////////////////////////////////////////////////////////////////////////
var starfm = function(modis, l8){
//Preliminary filtering of MODIS and Landsat
var filt_mod = modis.filterDate(date_begin, date_end);
var filt_l8 = l8.filterDate(date_begin, date_end)
.filterBounds(bounds);
//add bounds to subset area
var subset_bounds = bounds.transform('EPSG:4326', 30).bounds().getInfo();
// get rid of bad pixels but keep metadata date information
var removeBadObservations = function(image){
var valid_data_mask = ee.Image(image).select('pixel_qa').bitwiseAnd(2).neq(0);
var numberBandsHaveData =
image.mask().reduce(ee.Reducer.sum());
var allOrNoBandsHaveData =
numberBandsHaveData.eq(0).or(numberBandsHaveData.gte(9));
var allBandsHaveData = allOrNoBandsHaveData;
//Make sure no band is just under zero
var allBandsGT = image.reduce(ee.Reducer.min()).gt(-0.001);
var result = ee.Image(image).mask(image.mask().and(valid_data_mask).and(allBandsHaveData).and(allBandsGT));
return result.copyProperties(ee.Image(image),['system:time_start']);
};
//NDVI functions for Landsat and MODIS
var getNDVI_mod = function(image){
return image
.addBands(image.normalizedDifference(['sur_refl_b02','sur_refl_b01']).multiply(10000).rename('NDVI'));};
var getNDVI_l8= function(image){
var ndvi = ee.Image(image).normalizedDifference(['B5','B4']);
return ndvi.copyProperties(ee.Image(image),['system:time_start']);
};
var filt2_l8= filt_l8.filterBounds(subset_bounds).aside(print).map(removeBadObservations).map(getNDVI_l8);
// update mask for different areas
var subset_mask = ee.Image().byte().paint(geometry4, "id").add(1);
var filtered_modis = filt_mod.filterBounds(subset_bounds).aside(print).map(getNDVI_mod);
// Pull out the date:
var extract_modis_date = function(row) {
var d = ee.Date(row.get('system:time_start'));
var d2 = ee.Date.fromYMD(d.get('year'), d.get('month'), d.get('day'));
var result = ee.Feature(null, {'date': d2});
result = result.set({'date': d2});
return result;
};
var getQABits = function(image, start, end, newName) {
//// Compute the bits we need to extract.
var pattern = 0;
for (var i = start; i <= end; i++) {
pattern += Math.pow(2, i);
}
//// Return a single band image of the extracted QA bits, giving the band a new name.
return image.select([0], [newName])
.bitwiseAnd(pattern)
.rightShift(start); };
print(filtered_modis);
filtered_modis = filtered_modis.map(function(image){
var quality = getQABits(image.select(2), 4, 5, 'QAMask');
quality = quality.eq(3).not();
return image.clip(subset_bounds).mask(image.mask().multiply(subset_mask).multiply(quality));
});
filtered_modis = filtered_modis.select('NDVI'); //Take only NDVI band
var modis_multiband = ee.Image(filtered_modis.filterDate(date_begin, date_end).iterate( function(x, modis_multiband) {
return ee.Image(modis_multiband).addBands(ee.Image(x));
}, filtered_modis.first()));
var dates_modis = filtered_modis.map(extract_modis_date);
print(dates_modis.getInfo());
var filt2_l8_ndvi = filt2_l8.map(function(image) {
return ee.Image(image)
.clip(subset_bounds)
.mask(ee.Image(image).mask().multiply(subset_mask));
});
//use this to choose the range of days, and check for overlap
var day_expand = 1;
var reduceLandsatNDVI = function(MODISdate) {
MODISdate = ee.Date(MODISdate.get('date'));
var ndvi_subset = ee.ImageCollection(filt2_l8_ndvi).filterDate( MODISdate,
MODISdate.advance(day_expand, 'day') );
ndvi_subset = ndvi_subset.map(function (image) {
var diff = MODISdate.difference(ee.Date(ee.Image(image).get('system:time_start')), 'day').abs();
return ee.Image(image).set('diff', diff);
});
ndvi_subset = ndvi_subset.sort('diff');
var ndvi_first = ndvi_subset.reduce('first');
var ndvi_mean = ndvi_subset.reduce('mean');
return ee.Algorithms.If(
ndvi_first.bandNames(),
ndvi_first.eq(0).multiply(ndvi_mean).add(ndvi_first),
ee.Image(0)
);
};
var extract_landsat_date = function(MODISdate) {
MODISdate = ee.Date(MODISdate.get('date'));
var ndvi_subset =
ee.ImageCollection(filt2_l8_ndvi).filterDate( MODISdate,
MODISdate.advance(day_expand, 'day') );
ndvi_subset = ndvi_subset.map(function (image) {
var diff = MODISdate.difference(ee.Date(ee.Image(image).get('system:time_start')), 'day').abs();
return ee.Image(image).set('diff', diff);
});
ndvi_subset = ndvi_subset.sort('diff');
var d = ndvi_subset.aggregate_first('system:time_start');
var count = ndvi_subset.aggregate_count('system:time_start');
d = ee.Algorithms.If(
ee.Number(count).gt(0),
ee.Date(d),
ee.Date('1971-01-01')
);
d = ee.Date(d);
var d2 = ee.Date.fromYMD(d.get('year'), d.get('month'),
d.get('day'));
var result = ee.Feature(null, {'LSdate': d2, 'MODISdate':
MODISdate, 'CountLSScenes': count});
result = result.set({'LSdate': d2, 'MODISdate': MODISdate,
'CountLSScenes': count});
return result;
};
var ls_collection = dates_modis.map(reduceLandsatNDVI);
print(ls_collection,'ls collection');
var dates_landsat = dates_modis.map(extract_landsat_date);
Export.table(dates_landsat, csv_title);
var ls_multiband = ls_collection.iterate( function(x,
ls_multiband)
{ return ee.Image(ls_multiband).addBands(ee.Image(x));
}, ls_collection.first());
ls_multiband = ee.Image(ls_multiband).multiply(10000).int16();
ls_multiband = ls_multiband.mask(ls_multiband.mask().multiply(ls_multiband.neq(0)));
//remove repeated first layer
var ls_multiband2=ee.Image(ls_multiband.slice(1))
var modis_multiband2=ee.Image(modis_multiband.slice(1))
Export.image.toDrive({
image: modis_multiband2,
description: mod_title,
crs:'EPSG:4326 ',
region:region,
scale:30
});
Export.image.toDrive({
image: ls_multiband2,
description: ls_title,
crs:'EPSG:4326',
region:region,
scale:30
});
return 'Done!'
}
var running = starfm(modis,l8)
print(running)
///////////////////////////////////////////////////////////////////////////////////////////////////////////////////