Field NotesPlant techNo. 08

What is NDVI? What plant-health cameras actually measure

Satellites, drones and plant cameras all report NDVI. It compares two kinds of reflected light, and it says more about how much green leaf is in view than about what is wrong with a plant.

Aerial view of young trees in straight rows on bare tan soil

The short answer NDVI (normalized difference vegetation index) compares the near-infrared light a surface reflects with the red light it reflects. Green leaves absorb red light and reflect near-infrared strongly, so dense green vegetation scores high (about 0.6 to 0.9 by USGS figures), barren rock and sand sit near zero, and water goes negative. It tracks how much healthy green leaf a sensor sees, not what is wrong with a plant.

You have probably met NDVI as a satellite map shaded green, a drone image of a striped field, or a feature on a “plant health” camera. It is simpler than it looks: two reflectance readings, one subtraction, one addition and one division.

A plant is always reporting on its condition, and NDVI listens to two channels of that report. That works well for a square kilometer of forest and less well for one pothos on a shelf.

Why healthy leaves reflect near-infrared and absorb red

Chlorophyll absorbs visible light. NASA puts this at about 0.4 to 0.7 µm (400 to 700 nm), absorbed for use in photosynthesis. That is the light that matters when you measure light for houseplants. USGS singles out red as the visible light leaf pigments absorb strongly.

The leaf’s cell structure reflects near-infrared. NASA gives about 0.7 to 1.1 µm (700 to 1,100 nm), light your eyes can’t see.

The more leaves a plant has, NASA notes, the stronger both effects get. Dense, healthy vegetation looks dark in red and bright in near-infrared. Unhealthy or sparse vegetation reflects more visible light and less near-infrared.

Real sensors read narrower slices: Landsat 8’s red and near-infrared bands, for example, are centered at 655 nm and 865 nm (Huang et al., 2021).

The NDVI formula

NDVI = (NIR − Red) / (NIR + Red)

  • NIR is near-infrared reflectance: the fraction of incoming near-infrared light the surface reflects, from 0 to 1.
  • Red is red reflectance, the same fraction for the red band.

Because the top is a difference and the bottom is the sum of the same two numbers, NDVI always falls between −1 and +1 (NASA, USGS). Much more near-infrared than red pushes it toward +1, equal amounts give 0, and more red than near-infrared gives a negative value. Only the ratio of the two matters: Huang et al. note that a near-infrared to red ratio of 2 corresponds to an NDVI of 0.33.

What the numbers mean

The sources group surfaces differently.

Surface NASA Earth Observatory USGS Huang et al. (2021)
Water Not given Not given Negative
Barren rock, sand or snow 0.1 and below 0.1 or less Close to zero (rock, sand, concrete; snow not listed)
Shrubs, grassland, sparse vegetation 0.2 to 0.3 About 0.2 to 0.5 (also senescing crops) Positive
Dense vegetation 0.6 to 0.8 (temperate and tropical rainforest) About 0.6 to 0.9 (forests, crops at peak growth) Positive

NASA adds that values close to +1 (0.8 to 0.9) mean the highest possible density of green leaves. Sensor maker MicaSense, writing about crops, treats 0.5 as typically low vigor and 0.9 as very high, so “good” depends on what you measure.

None of these sources gives a figure for garden or potting soil, and USGS says NDVI is sensitive to soil color, so don’t assume soil reads as exactly zero.

A worked example

Example 1, a healthy leaf: NIR reflectance 0.50, red 0.08.

  • NIR − Red = 0.50 − 0.08 = 0.42
  • NIR + Red = 0.50 + 0.08 = 0.58
  • NDVI = 0.42 ÷ 0.58 = 0.724, about 0.72

Example 2, a sparse or struggling patch: NIR 0.40, red 0.30.

  • 0.40 − 0.30 = 0.10, and 0.40 + 0.30 = 0.70
  • NDVI = 0.10 ÷ 0.70 = 0.143, about 0.14

These are illustrative numbers, not measurements, following the pattern NASA describes.

Now the top end. Keep NIR at 0.50 and halve red to 0.04: NDVI = 0.46 ÷ 0.54 = 0.85. Halve it again to 0.02: NDVI = 0.48 ÷ 0.52 = 0.92. Each step doubles the near-infrared to red ratio (6.25, then 12.5, then 25), yet NDVI rises by 0.13 and then by only 0.07; Huang et al. note that an NDVI of 0.80 already takes a ratio of 9. The formula squeezes high values together, which matters when two lush plants read almost the same.

Limits: why the same plant can give different numbers

Saturation. USGS says NDVI tends to saturate over dense vegetation, and MicaSense says it often loses sensitivity once plants reach a critical level of leaf cover or chlorophyll content. Huang et al. cite saturation in broadleaf forests above about 100 tons of biomass per hectare.

Soil background. Where soil shows through, it shifts the value, and USGS notes the effect depends on soil color.

Light and shadow. USGS says NDVI can compensate for changing illumination, surface slope and viewing angle, which is why it is especially useful for continental- to global-scale monitoring. Closer in, Public Lab notes that NDVI can differ between shady and sunny areas: shade is lit by different-colored light than sun, and in underexposed areas the channel values get so small that NDVI becomes more sensitive to random variation.

Mixed pixels. Each pixel’s NDVI blends whatever falls inside it, and a pixel from the AVHRR satellite instrument covers about 1 square kilometer (NASA). Huang et al. list “pixel components” with atmosphere and soil as things NDVI is sensitive to.

The atmosphere and the sensor. Clouds and aerosols can block a satellite’s view entirely (NASA). A drone sensor is often so close to the ground that the atmospheric effect is minimal, but Huang et al. warn that many commercial drone NDVI sensors may carry substantial geometric and radiometric errors that need calibration, and that different sensors can deliver notably different NDVI behavior. A 0.7 from one camera is not automatically a 0.7 from another.

Calibration targets. Hence calibrated reflectance panels, targets whose reflectance has been tested across the sensor’s bands. MicaSense says panel images compensate for the lighting at the time of capture and “anchor” the data to that date, time and location; without a calibration process, it says, data captured on different days or at different times of day cannot be accurately compared for changes.

It doesn’t name the problem. NASA groups unhealthy and sparse vegetation together: both reflect more visible and less near-infrared light. Huang et al. state that a specific NDVI value does not have a sole interpretation. A drop can mean fewer leaves, sicker leaves or different light; you still have to go and look.

  • NDRE (normalized difference red edge) can only be calculated when the sensor has a red edge band, MicaSense notes. It explains that red-edge light is more translucent to leaves than red light and less likely to be completely absorbed by a canopy, calls NDRE a better indicator of vigor than NDVI for mid to late season crops that have accumulated high levels of chlorophyll in their leaves, and lists uses including leaf chlorophyll content and nitrogen uptake.
  • EVI (enhanced vegetation index) corrects for some distortion from particles in the air and from the ground cover below the vegetation. NASA says it doesn’t saturate as easily as NDVI over rainforests and other areas with large amounts of chlorophyll.
  • OSAVI (optimized soil-adjusted vegetation index) is meant for relatively sparse vegetation where soil is visible through the canopy, per MicaSense.

Why a normal phone camera can’t measure NDVI

Public Lab, the project behind Infragram cameras, says a CCD or CMOS camera can capture near-infrared only after its infrared-blocking filter is removed. For a smartphone, its FAQ says, you would have to permanently modify the phone, and Public Lab can’t confirm the filter is removable on any particular model.

Public Lab’s method is to remove the IR-block filter and add its own filter, which passes near-infrared along with blue and green light, so near-infrared shows up in the camera’s red channel. That leaves no red band, so the visible reading comes from blue or green light instead, and the result is not the red-based NDVI defined above. The FAQ adds that the sensitivity of most inexpensive consumer cameras seems to drop off sharply beyond about 800 nm, and that results depend partly on exposure.

Purpose-built multispectral cameras, like those flown on agricultural drones, record the needed bands and are used with reflectance panels (MicaSense). If an app on an unmodified phone shows you a plant-health map, it is working from visible color.

NDVI for one houseplant versus a field

Over a field or forest, NDVI is a summary tool. NASA describes comparing each place with its average for the same month over the previous 20 years (the difference is the NDVI anomaly). Odd patches still get checked on the ground. Up close, things change:

  • The frame is full of things that aren’t leaf. Pot, soil, wall and shelf usually read lower than leaf and can skew a whole-image average. Average only the leaf pixels.
  • Shade works leaf by leaf. Inner leaves sit in shadow, where Public Lab’s warning about noisy, underexposed values applies.
  • Indoor light varies. Readings depend on the light at capture time, and indoors that changes with the lamp, the window and the hour.
  • A lush plant may already be near the top. Dense, overlapping foliage is where USGS and MicaSense say NDVI loses sensitivity, and the arithmetic above squeezes high values together.
  • Numbers don’t transfer. Don’t expect a modified-camera reading to match a satellite map or another brand of sensor (Huang et al.).
  • The obvious changes are visible anyway. On a plant a foot away, the leaf loss and yellowing that pull NDVI down are often things you can see, and NDVI won’t tell you the cause.

If you want to try it anyway, use NDVI to compare a plant with itself:

  1. Keep the camera, distance, angle and light the same, at the same time of day.
  2. Put a reference target of known reflectance in every frame.
  3. Mask out everything but leaf before averaging.
  4. Watch the trend over weeks, not single readings, and inspect the plant when it moves.

If you suspect the plant is short on water, NDVI can’t confirm it on its own. How to measure plant water stress covers leaf temperature, stem diameter and sap flow.

Sources

  1. Measuring Vegetation (NDVI & EVI)NASA Earth Observatory
  2. NDVI, the Foundation for Remote Sensing PhenologyU.S. Geological Survey
  3. A commentary review on the use of normalized difference vegetation index (NDVI) in the era of popular remote sensingJournal of Forestry Research 32:1-6 (Huang et al., 2021)
  4. Infragram FAQPublic Lab
  5. How do calibrated reflectance panels improve my data?MicaSense (sensor maker support documentation)
  6. Overview of Agricultural IndicesMicaSense (sensor maker support documentation)

General information, not professional advice for a specific plant or crop. Spot something wrong? Email info@terrathink.io.

Keep reading

All field notes

Join the waitlist

New field notes by email, and first word when the hardware goes live.

By joining, you agree to get email updates from Terrathink. Unsubscribe anytime. Privacy policy