Spectral Indices

Quick win: our first NDVI map

Let's start with the classic. NDVI (Normalized Difference Vegetation Index) is the most widely used spectral index in remote sensing. We will compute it on a Landsat 8 image and see vegetation light up in green.

// NDVI example (Landsat 8 Surface Reflectance)
var img = ee.Image('LANDSAT/LC08/C02/T1_L2/LC08_044034_20210623')
  .multiply(0.0000275).add(-0.2); // apply scale/offset
var ndvi = img.normalizedDifference(['SR_B5', 'SR_B4']).rename('NDVI');
Map.centerObject(img, 9);
Map.addLayer(ndvi, {min: 0, max: 0.8, palette: ['brown', 'yellow', 'green']}, 'NDVI');
print('NDVI min/max', ndvi.reduceRegion({
  reducer: ee.Reducer.minMax(),
  geometry: img.geometry(),
  scale: 30,
  maxPixels: 1e9
}));

Common indices

• NDVI = (NIR - RED) / (NIR + RED) - vegetation vigor.
• NDWI = (NIR - SWIR) / (NIR + SWIR) - vegetation water content.
• MNDWI = (GREEN - SWIR) / (GREEN + SWIR) - open water detection.
• NBR = (NIR - SWIR2) / (NIR + SWIR2) - burn severity.
• NDBI = (SWIR - NIR) / (SWIR + NIR) - built-up areas.

Try another index: NDWI (Sentinel-2)

var s2 = ee.ImageCollection('COPERNICUS/S2_SR')
  .filterDate('2023-06-01', '2023-06-15')
  .filterBounds(ee.Geometry.Point([-82.3, 29.6]))
  .first();
var ndwi = s2.normalizedDifference(['B8', 'B11']).rename('NDWI');
Map.addLayer(ndwi, {min: -0.5, max: 0.8, palette: ['brown','beige','cyan']}, 'NDWI');

Beyond NDVI: EVI, SAVI, and NDRE

NDVI is great, but it has limits. In dense tropical forests, NDVI maxes out near 0.9 and stops distinguishing between "very green" and "extremely green." In arid landscapes, bare soil confuses the signal. And for precision agriculture, we sometimes need finer sensitivity to chlorophyll than NDVI can offer.

That is where these three advanced indices come in. Each one was designed to solve a specific weakness of NDVI. Let's meet them.

EVI (Enhanced Vegetation Index)

Think of EVI as "NDVI with a correction filter." It reduces atmospheric haze and soil background noise that can throw off NDVI readings. Most importantly, it does not saturate in dense vegetation, which makes it the go-to index for agricultural monitoring and tropical forests.

Formula: EVI = 2.5 × (NIR - RED) / (NIR + 6×RED - 7.5×BLUE + 1)

Notice the Blue band in the formula. That is the atmospheric correction term. It is also why EVI needs .expression() instead of .normalizedDifference().

// EVI on Sentinel-2
var s2 = ee.ImageCollection('COPERNICUS/S2_SR_HARMONIZED')
  .filterDate('2023-06-01', '2023-06-30')
  .filterBounds(ee.Geometry.Point([-93.5, 42.0]))
  .filter(ee.Filter.lt('CLOUDY_PIXEL_PERCENTAGE', 20))
  .median()
  .divide(10000); // scale to reflectance

var evi = s2.expression(
  '2.5 * ((NIR - RED) / (NIR + 6 * RED - 7.5 * BLUE + 1))',
  {
    NIR: s2.select('B8'),
    RED: s2.select('B4'),
    BLUE: s2.select('B2')
  }
).rename('EVI');

Map.addLayer(evi, {min: 0, max: 0.8, palette: ['brown','yellow','darkgreen']}, 'EVI');

SAVI (Soil-Adjusted Vegetation Index)

SAVI was designed for the places where bare soil shows through the vegetation canopy. Think of arid grasslands in West Texas, sparse shrubland in the Sahel, or a newly planted crop field. In these settings, soil reflectance contaminates NDVI readings, making sparse vegetation look even sparser than it is.

SAVI adds a soil brightness correction factor (L) to compensate. The standard value is L = 0.5, which works well for most intermediate vegetation cover.

Formula: SAVI = ((NIR - RED) / (NIR + RED + L)) × (1 + L), where L = 0.5

// SAVI on Landsat 8
var img = ee.Image('LANDSAT/LC08/C02/T1_L2/LC08_044034_20210623')
  .multiply(0.0000275).add(-0.2);

var savi = img.expression(
  '((NIR - RED) / (NIR + RED + 0.5)) * 1.5',
  {
    NIR: img.select('SR_B5'),
    RED: img.select('SR_B4')
  }
).rename('SAVI');

Map.addLayer(savi, {min: 0, max: 0.7, palette: ['brown','yellow','green']}, 'SAVI');

NDRE (Normalized Difference Red Edge)

NDRE is the precision tool. It uses Sentinel-2's red edge band (B5, centered at 705 nm) instead of the visible red band. Why does that matter? The red edge is more sensitive to chlorophyll content than the red band. This means NDRE can detect subtle vegetation stress that NDVI misses entirely.

Think of it this way: NDVI tells us "is there vegetation?" while NDRE tells us "how healthy is that vegetation?" Precision agriculture relies heavily on NDRE for exactly this reason.

Formula: NDRE = (NIR - Red Edge) / (NIR + Red Edge)

Important: NDRE only works with Sentinel-2 (and a few other sensors that have a red edge band). Landsat does not have a red edge band, so NDRE cannot be computed from Landsat imagery.

// NDRE on Sentinel-2 (only works with Sentinel-2, not Landsat)
var ndre = s2.normalizedDifference(['B8', 'B5']).rename('NDRE');
Map.addLayer(ndre, {min: 0, max: 0.6, palette: ['red','yellow','green']}, 'NDRE');

When to use which?

Here is a quick mental model:

• NDVI: The everyday workhorse. Use it when vegetation cover is moderate and soil is not a concern.
• EVI: The upgraded model. Use it for dense forests, cropland, or anywhere NDVI saturates above 0.85.
• SAVI: The desert specialist. Use it when more than 30% of the ground is bare soil.
• NDRE: The precision instrument. Use it for crop stress detection and precision agriculture (Sentinel-2 only).

Writing Custom Index Functions

Once we understand the formulas, the next step is to wrap them in reusable functions. This way, we can .map() our index over an entire collection with a single line of code. Let's see how.

// Reusable function to compute EVI for any Sentinel-2 image
function computeEVI(image) {
  var scaled = image.divide(10000); // scale to reflectance first
  return scaled.expression(
    '2.5 * ((NIR - RED) / (NIR + 6 * RED - 7.5 * BLUE + 1))',
    {
      NIR: scaled.select('B8'),
      RED: scaled.select('B4'),
      BLUE: scaled.select('B2')
    }
  ).rename('EVI')
   .copyProperties(image, ['system:time_start']);
}

// Apply to a collection
var eviCollection = ee.ImageCollection('COPERNICUS/S2_SR_HARMONIZED')
  .filterDate('2023-01-01', '2023-12-31')
  .filterBounds(ee.Geometry.Point([-82.3, 29.6]))
  .filter(ee.Filter.lt('CLOUDY_PIXEL_PERCENTAGE', 20))
  .map(computeEVI);

print('EVI collection size:', eviCollection.size());

Which Index Should I Use?

Use this decision table to choose the right index for your analysis:

Next steps

• Compute NDVI and NDWI on the same scene; compare histograms and maps.
• Test MNDWI on a coastal area to isolate water bodies.
• Write a reusable function that computes EVI for any Sentinel-2 image and .map() it over a collection.
• Read the Earth Engine docs for normalizedDifference and expression.