I want to use supervised classification to classify a pattern that has a clear temporal pattern. For example, identifying stands of deciduous trees in a coniferous forest. NDVI would change overtime in the deciduous stands in a regular pattern that should be easily detectable. I assume there's an easy method to flatten the temporal dataset into a single image so that the bands in that image can be used in a classification algorithm. Maybe using .map(....)
?
Here's some code to build the answer from:
var startDate = '2016-05-01';
var endDate = '2016-09-01';
var lng = -122.3424; var lat = 37.9344; //SF
var region = ee.Geometry.Point(lng, lat);
//Image Import
var l8 = ee.ImageCollection('LANDSAT/LC08/C01/T1_TOA')
.filterBounds(region)
.filterDate(startDate,endDate);
// NDVI temporal
var ndvi = l8.map(function(image) {
var ndvi = image.normalizedDifference(['B5', 'B4']).rename("NDVI");
return ndvi;
});
Map.addLayer(ndvi,{},"NDVI Temporal"); // 8 images with 1 band
//NDVI FLATTENED??????? I want 1 image with 8 bands. The below code doesn't work...
var ndviFlat = ee.Image().addBands(ndvi.map(function(image){
var temp = image.select("NDVI");
return temp;
}));
From there, I will pass ndviFlat to .sampleRegions
, which only works with Images
not ImageCollections
:
//Classification Model:
var points = ee.FeatureCollection([trainingPointsPos,trainingPointsNeg]).flatten();
var training = ndviFlat.sampleRegions({
collection: points,
properties: ['class'],
scale: 30
});
var trained = ee.Classifier.randomForest(20).train(training, 'class', bands);
classified = regLayers.select(bands).classify(trained);