Google Earth Engine NDVI Time Series: A Sentinel-2 Script with Cloud Score+

An NDVI time series shows how vegetation in one place changes through the seasons and across years: a crop cycle, a drought, a forest recovering after fire. Earth Engine is the quickest way to build one, because the whole Sentinel-2 archive is already there. The weak point in most tutorials is cloud masking. They use the QA60 band, which misses thin cloud and shadow, so the chart fills with false dips. The script below uses Cloud Score+ instead.

Before you start

You need a Google account and an Earth Engine-enabled Google Cloud project; registration for non-commercial use is free at code.earthengine.google.com. Paste each block into the Code Editor in order. Replace the area with your own: draw a polygon with the geometry tools, or edit the coordinates.

Step 1: the area and the cloud-masked collection

// Area of interest: a 500 m circle around a point (farmland near Seville, Spain).
// Replace with your own point or a drawn polygon.
var aoi = ee.Geometry.Point([-5.97, 37.38]).buffer(500);

var s2 = ee.ImageCollection('COPERNICUS/S2_SR_HARMONIZED');
var csPlus = ee.ImageCollection('GOOGLE/CLOUD_SCORE_PLUS/V1/S2_HARMONIZED');
var QA_BAND = 'cs_cdf';
var CLEAR_THRESHOLD = 0.60; // 0.50-0.65 works for most areas

var ndvi = s2
  .filterBounds(aoi)
  .filterDate('2019-01-01', '2026-01-01')
  .linkCollection(csPlus, [QA_BAND])
  .map(function (img) {
    var clear = img.updateMask(img.select(QA_BAND).gte(CLEAR_THRESHOLD));
    return clear.normalizedDifference(['B8', 'B4']).rename('NDVI')
      .set('system:time_start', img.get('system:time_start'));
  });

COPERNICUS/S2_SR_HARMONIZED already removes the 2022 radiometric offset, so you can use the bands directly. linkCollection joins each Sentinel-2 image to its Cloud Score+ image, and pixels with a clear-sky score below 0.6 are masked.

Step 2: monthly composites

var start = ee.Date('2019-01-01');
var nMonths = 7 * 12;

var monthly = ee.ImageCollection.fromImages(
  ee.List.sequence(0, nMonths - 1).map(function (m) {
    var from = start.advance(m, 'month');
    var subset = ndvi.filterDate(from, from.advance(1, 'month'));
    return subset.median()
      .set('system:time_start', from.millis())
      .set('n', subset.size());
  })
).filter(ee.Filter.gt('n', 0)); // drop months with no images

Step 3: chart it

var chart = ui.Chart.image.series({
  imageCollection: monthly,
  region: aoi,
  reducer: ee.Reducer.mean(),
  scale: 10
}).setOptions({
  title: 'Monthly mean NDVI',
  vAxis: {title: 'NDVI', viewWindow: {min: 0, max: 1}},
  lineWidth: 1,
  pointSize: 3
});
print(chart);
Map.centerObject(aoi, 15);
Map.addLayer(monthly.filterDate('2025-06-01', '2025-07-01').first(),
  {min: 0, max: 0.9, palette: ['brown', 'yellow', 'green']}, 'NDVI June 2025');

Run the script. The chart appears in the Console. Click the arrow at its top right to open it full size, then Download CSV to take the numbers into Excel, R or Python.

Reading the chart

A regular wave each year is the growing season. A lower peak in one year can mean drought, a different crop or earlier harvest. A sudden, lasting drop suggests clearing, fire or building. Single sharp dips that recover the next month are often leftover cloud or haze; if you see many, raise the threshold to 0.65. For land under persistent cloud, Sentinel-1 radar is the alternative, because it sees through cloud.