NDVI and Spectral Indices: The Arithmetic of Earth Observation

A spectral index compresses multiband imagery into one diagnostic number per pixel. NDVI — the Normalized Difference Vegetation Index, computed from red and near-infrared reflectance — is the most famous: healthy vegetation scores high, bare soil low, water negative. Dozens of siblings target water (NDWI), built-up areas (NDBI), burn severity (NBR) and soil-adjusted vegetation (SAVI, EVI). Indices are simple, fast, physically motivated — and easy to over-interpret.

Why NDVI works

Chlorophyll absorbs red light; healthy leaf structure scatters near-infrared strongly. NDVI = (NIR − Red)/(NIR + Red) captures that contrast on a −1 to +1 scale: dense vigorous canopy typically 0.6–0.9, sparse or stressed vegetation lower, soil near 0.1–0.2, water below zero. The normalised ratio suppresses illumination differences, making values comparable across scenes.

Its power multiplies in time: an NDVI curve traces a crop’s season — green-up, peak, senescence — enabling crop identification, sowing-date estimation and stress alerts from trajectory anomalies.

The wider index family

EVI and SAVI correct NDVI’s saturation over dense canopy and soil-background sensitivity in sparse cover. NDWI variants highlight open water or canopy moisture; NBR differencing before/after fire grades burn severity; NDBI leans toward built surfaces; red-edge indices from Sentinel-2 sharpen crop nitrogen and stress signals.

Choosing an index is choosing which physical contrast to exploit — and acknowledging what else moves it.

Using indices honestly

Indices computed on uncorrected imagery inherit atmosphere and calibration noise — surface reflectance first. Thresholds are not universal: “NDVI > 0.4 is vegetation” varies by biome, season and sensor; calibrate locally. Mixed pixels blur meaning at coarse resolution, and NDVI measures greenness, not species, yield or health directly — those are inferences requiring ground truth.

Treated as calibrated evidence within a designed analysis, indices are superb; treated as universal truth-meters, they mislead with great confidence.

Frequently asked questions

What do NDVI values mean?
As a rough guide: below 0 water; 0–0.2 bare soil and built surfaces; 0.2–0.5 sparse or stressed vegetation and crops mid-cycle; 0.6–0.9 dense healthy canopy. Exact interpretation depends on biome, season and sensor — local calibration beats universal tables.
Which is better, NDVI or EVI?
Neither universally. EVI resists saturation over dense canopy and reduces soil/atmosphere effects, at the cost of an extra band and coefficients; NDVI is simpler and comparable across decades of archives. Many programmes compute both.
Related Topics
Need This Done Professionally?
More in GeoAI & Machine Learning

Keep learning

GeoAI: Where Machine Learning Meets GIS

What GeoAI covers — from feature extraction to prediction — and what it takes to make machine learning work on spatial problems.

Read the article →

Deep Learning on Satellite Imagery

How convolutional networks and segmentation models extract buildings, roads and land cover from imagery at scale.

Read the article →

Training Data and Annotation for Geospatial AI

Why labelled data decides model quality, how geospatial annotation is done well, and what quality control looks like.

Read the article →

Cloud Geospatial Platforms: Google Earth Engine and Beyond

How planetary-scale cloud platforms changed Earth-observation analysis — data next to compute, and what that enables.

Read the article →

Put this knowledge to work

Every encyclopedia topic maps to a production service we deliver worldwide. Tell us what you're building.

Request a Free Quote
Have a project in mind? Proposal within 48 hours.Get a Quote