Use image to apply unique filterDate range for each pixel in an imagecollection

Question

I have 2 images of timestamps (in system:time_start type format, ms since unix epoch) corresponding to a time series image collection; each pixel in the two images represents a date range, first image is the start date, second is the stop date.

Is there a way to reduce the image collection where each pixel in the image collection is filtered to its own unique time window?

Answer 1

There is the possibility to mask individual pixels of image of an image collection.

As you do not define any code, I created to sample image datestamps in milliseconds. The start dates are in 2017, the end dates in 2018, to make sure there is always a time window.

// image with DOYs in milliseconds
var date = ee.Date('2017-01-01').millis(); // get a date
var dayToMillis = 24*60*60*1000; //86400000
var randomStarts = ee.Image.random().multiply(365).multiply(dayToMillis)
                      .add(date).toLong().rename('startDate'); 
var randomEnds = ee.Image.random().multiply(365).multiply(dayToMillis)
                      .add(365*dayToMillis).add(date).toLong().rename('endDate');

Then we load a sample image collection, e.g. Landsat-8 and filter for the first image in 2018 to later present just one sample image on the map.

// load an image collection
var imageCollection = ee.ImageCollection("LANDSAT/LC08/C01/T1_TOA");
Map.centerObject(imageCollection.filterDate('2018','2019').first());

We can then map over that image collection, make a date image of each image timestamp and mask out every pixel with a date that is lt the start or gt the end.

// Mask each pixel based on the start and end images. 
imageCollection = imageCollection.map(function(image){
  var dateBand = ee.Image(image.date().millis());
  image = image.updateMask(dateBand.gte(randomStarts))
                .updateMask(dateBand.lte(randomEnds));
  return image.addBands(dateBand.rename('dateMillis'));
});

Add the result to the map and use the inspector tool to check that pixels outside the daterange are masked.

// Add sample image to the map (note that zoom will change the values as the random images 
// do not have a projection). 
Map.addLayer(randomStarts)
Map.addLayer(randomEnds)
Map.addLayer(imageCollection.filterDate('2018','2019').first().select(['B4','B3','B2','dateMillis']), 
              {bands: ['B4','B3','B2'], min: 0, max: 0.35}, 'sample image');

Link to reproducible example