Extracting NDVI values from Image using Google Earth Engine Python API

Question

I am using the Earth Engine Python API and attempting to extract date and NDVI values from an image collection. I have added an NDVI band to each image in the collection but I am having trouble figuring out how to extract the actual NDVI value from each band in order to produce a dataframe of an NDVI time series.

I am creating an NDVI band with the code below

    nir = img.select('B5')
    red = img.select('B4')
    ndvi = nir.subtract(red).divide(nir.add(red)).rename('NDVI')
    print('NDVI value:', ndvi.select('NDVI').getInfo())

and I am receiving info from the NDVI band in the form of

NDVI value: {'type': 'Image', 'bands': [{'id': 'NDVI', 'data_type': {'type': 'PixelType', 'precision': 'float'}, 'dimensions': [492, 492], 'origin': [2368, 3104], 'crs': 'EPSG:32612', 'crs_transform': [30, 0, 175185, 0, -30, 5063115]}]}

The documentation on accessing computed NDVI values is limited and I have been scanning tutorials and other exchange posts for ee api related solutions but have found nothing.

Answer 1

You currently have an image which has an NDVI value for each pixel in the original image. In order to get a single number, you need to reduce the image — to take the mean, median, maximum, minimum, or other such operation on pixels in the image, within a region of interest. Here is the documentation on reduceRegion, which is the way to do this.

Since you have an image collection and want a table, you also need to map over the image collection. Whenever you want a number or other value for each image in a collection, the way you do that is by adding a property to each image, using set(), or by producing new features that have only the data you want.

This example I wrote is in JavaScript since I wrote it in the Earth Engine Code Editor, but the differences with Python are fairly small so I hope you can follow along.

var meanNDVICollection = imgs.map(function (img) {
  // Your code.
  var nir = img.select('B5');
  var red = img.select('B4');
  var ndviImage = nir.subtract(red).divide(nir.add(red)).rename('NDVI');

  // Compute the mean of NDVI over the 'region'
  var ndviValue = ndviImage.reduceRegion({
    geometry: region,
    reducer: ee.Reducer.mean(),
  }).get('NDVI');  // result of reduceRegion is always a dictionary, so get the element we want

  // Add the value as a property to each image in the collection.
  // return img.set('NDVI', ndviValue);

  // Or, create a new feature with only the properties we want.
  return ee.Feature(null, {
    // Adding property we computed.
    'NDVI': ndviValue
  }).copyProperties(img, [
    // Picking properties from the original image.
    'system:time_start',
    'SUN_ELEVATION'
  ])
});

// Export, or you can use any other method for getting a table out of Earth Engine.
Export.table.toDrive({
  collection: meanNDVICollection
});

https://code.earthengine.google.com/580f2f06c9978caf43eac42b5dd0ff44

(Disclaimer: I do not have training in the scientific and statistical aspects of remote sensing. The choices of mean and scale: 20 in this script are for demonstration of Earth Engine only and should not be assumed to produce scientifically valid results.)

Answer 2

Kevin Reid provided a very helpful solution to my issue. Here is the python code that I ended up using

def meanNDVICollection (img):
  nir = img.select('B5');
  red = img.select('B4');
  ndviImage = nir.subtract(red).divide(nir.add(red)).rename('NDVI');

  # Compute the mean of NDVI over the 'region'
  ndviValue = ndviImage.reduceRegion(**{
    'geometry': hatCreekPolygon.getInfo(),
    'reducer': ee.Reducer.mean(),
  }).get('NDVI');  # result of reduceRegion is always a dictionary, so get the element we want

  newFeature = ee.Feature(None, {
      # Adding computed NDVI value
      'NDVI': ndviValue
  }).copyProperties(img, [
      # Picking properties from original image
      'system:time_start',
      'SUN_ELEVATION'
  ])

  return newFeature

This returned a Feature Collection which I was then able to loop through and access the date and ndvi values with get('NDVI') and get('system:time_start')