What NDVI is:

NDVI (Normalized Difference Vegetation Index) is a number between about -1 and 1 that expresses how “vegetation-like” and how vigorous a pixel is, based on how it reflects red and near‑infrared (NIR) light. Healthy leaves strongly absorb red (for photosynthesis) and strongly reflect NIR, so NDVI is high there.

Step 1: Capture multispectral bands

A multispectral drone camera captures several separate images at the same time, each in a narrow spectral band (for example: green, red, red‑edge, NIR), not just RGB.

At acquisition time:

• Each exposure creates one image per band; a Mavic 3M or Sequoia, for example, will output separate Red and NIR frames for each position along the flight path.
• Flight planning ensures sufficient overlap (front lap/side lap) so those frames can later be stitched into orthomosaics.

Step 2: Radiometric calibration (turn DN into reflectance)

Raw band images come out as digital numbers (DNs), which depend on exposure, sensor gain, and illumination conditions. To compare Red and NIR meaningfully, they must be converted to reflectance (fraction of incoming light reflected).

Typical workflow:

• Before or after the flight, you shoot a calibration panel with known reflectance in each band.
• The processing software (Pix4Dfields, DJI Terra, Agisoft, WebODM, etc.) uses those values plus camera metadata (exposure time, ISO, etc.) and sometimes a sunshine sensor to compute a transform from DN → reflectance for each band.
• The result is that each pixel in the Red/NIR images now stores an approximate surface reflectance value (often 0–1 or scaled).

Without this step, NDVI can still be computed, but values can be out of the expected -1 to 1 range or inconsistent between flights and lighting conditions.

Step 3: Build georeferenced band orthomosaics

Next, all those overlapping frames in each band are aligned and stitched into orthomosaics:

• Structure‑from‑motion (SfM) and bundle adjustment align camera positions using tie points, GNSS, and possibly RTK/PPK data.
• A digital surface model (DSM) is estimated and used to project each image onto the ground, correcting perspective and terrain distortions.
• For each band (Red, NIR, etc.), the software produces a single raster with one value per pixel, geometrically aligned and georeferenced.

After this step, you typically have a multiband orthomosaic (e.g., band stack) or separate single‑band GeoTIFFs for Red and NIR, all in the same projection and resolution.

Step 4: Perform band math to compute NDVI

Now the core NDVI step is just band math:

NDVI=(NIR−Red) / (NIR+Red)

This is applied per pixel over the aligned Red and NIR rasters.

Typical implementations:

• In a GIS (QGIS, ArcGIS Pro), you use a raster calculator, referencing bands by index (e.g., “Band 5” as NIR, “Band 4” as Red in Landsat) and entering the NDVI formula.
• In drone mapping software (Pix4D, DroneDeploy, Pix4Dfields, WebODM, DJI Terra), NDVI is often a built‑in “vegetation index” option; you select the Red and NIR bands, and it creates the NDVI layer automatically.

The normalization (dividing by NIR + Red) keeps NDVI in a bounded range, theoretically between -1 and 1 for physically realistic reflectance values.

Values are interpreted as:

• Near +1 → dense, healthy vegetation
• Around 0 → bare soil, stressed vegetation, or mixed pixels
• Negative → water, snow, clouds, or non‑vegetated surfaces

Step 5: Visualize NDVI as a map

The NDVI raster is a single‑band image, but to interpret it, software applies a color ramp:

• Often, reds/yellows for low NDVI, greens for high NDVI.

• Some tools auto‑stretch the min/max values, others use fixed thresholds so that NDVI 0.2–0.4, 0.4–0.6, etc., map to specific colors.

This visual NDVI map is what you usually deliver for agronomy or vegetation assessments, sometimes combined with contour overlays, zones, or polygons.

Step 6: Quality checks and practical considerations

In practice, several things affect how “trustworthy” the NDVI is:

• Band assignment: Ensuring you are using the correct band for Red and NIR; mis‑assigned bands yield nonsense NDVI.
• Sun‑angle and illumination: Flights at different times of day or under clouds can change apparent reflectance if calibration is poor.
• Calibration method: Camera‑only vs. camera + sun sensor calibration; some users report NDVI ranges only behaving as expected when using camera‑based calibration alone in certain pipelines.

Ground truthing: NDVI variations must be linked to actual field conditions (visual inspection, leaf samples, soil tests) to interpret thresholds and stress patterns correctly