Visible-Light vs Multispectral Drone Imagery: What Each Tells a Grower

Two very different kinds of imagery get sold as "drone crop maps." One comes from an ordinary color camera. The other comes from a sensor that records light the eye cannot see. They answer different questions and fail in different ways, and it is worth knowing which is which before the map arrives in a red-to-green color scale.

Up front: we currently fly visible-light (RGB) imagery only. Our multispectral service is coming soon and is not offered today. This article explains both because the comparison is the useful part. The shorter version, focused on vineyards, is in what visible-light scanning shows and where NDVI fits.

What a visible-light camera records

A color camera records three broad bands, red, green and blue, roughly matching what your eye sees. Flown as a mapping mission and stitched together, the photos become an orthomosaic: one flat, scaled, georeferenced image of the whole block, at a few centimeters per pixel. Gaps, dead trees and a washed-out turn row are obvious from above in a way they never are from the end of the row.

The pixel values in that image are brightness numbers from the camera, not calibrated measurements. That matters later.

What color indices do with it

A visible-light index is arithmetic on those three bands that makes green vegetation stand out from everything else. The three you will see most often:

• ExG (Excess Green), 2G − R − B. Woebbecke and colleagues tested it in 1995 for telling weeds apart from soil and crop residue, and found that the color of living plants differed enough from background soil and residue to make the separation feasible.
• GLI (Green Leaf Index), (2G − R − B) / (2G + R + B). The same idea, normalized so values fall between −1 and 1.
• VARI (Visible Atmospherically Resistant Index), (G − R) / (G + R − B). Gitelson and colleagues developed it in 2002 to estimate vegetation fraction, the share of ground covered by green canopy, from visible bands alone, and designed it to be minimally sensitive to atmospheric effects.

Rendered as a map, any of these shows spatial variation in canopy color, density and greenness: which rows are thinner, smaller or yellower than their neighbors, and whether the difference forms a pattern. In practice that covers:

• Missing vines and trees. A hole in a regular grid. Countable.
• Canopy cover. How much of the ground the canopy shades, block by block.
• Irrigation lines not running. A strip along one drip lateral, a wedge at the end of a line, a ring under one pivot span. Once the plants show it, the pattern is plain.
• Visible damage. Broken limbs, flattened or scorched patches, washouts.

Where visible-light indices run out

They are sensitive to light. An uncalibrated camera reports brightness, and brightness changes with sun angle, cloud and exposure. A map flown on a hazy morning and one flown at noon in clear sky are not comparable value for value, even over an unchanged block. Patterns within one flight are reliable; small differences between dates are not.

They are sensitive to soil background. In a vineyard, most of the ground between the rows is soil or cover crop, and pixels along the canopy edge mix leaf and ground. Pale caliche and dark loam pull the index in different directions. A row on lighter soil can read differently from an identical row on darker soil.

They have no near-infrared. A color camera sees what you see. It does not see stress before stress is visible to the eye, and a visible-light index does not diagnose water stress, nutrient deficiency, pests or disease. A low patch might be any of those, or a soil change, a different rootstock, or a younger planting. The map says where to look, not why.

What multispectral adds

A multispectral sensor records several narrow bands instead of three broad ones, typically including red-edge (roughly 700 to 730 nm, where reflectance climbs steeply from red toward near-infrared) and near-infrared. Healthy leaf tissue absorbs red light and reflects near-infrared strongly, and that contrast is what the standard indices exploit.

• NDVI, (NIR − Red) / (NIR + Red), runs from −1 to 1. The USGS describes it as a measure of vegetation greenness that helps show vegetation density and changes in plant health.
• NDRE uses the red-edge band in place of red, (NIR − RedEdge) / (NIR + RedEdge). It is widely used where NDVI flattens out over dense canopy.

The other thing multispectral work adds, done properly, is radiometric calibration. A reflectance panel of known value is photographed before and after the flight, and a sunlight sensor on the aircraft records how incoming light changes during it. Together they convert raw camera numbers into surface reflectance, which is what makes one date comparable with the next. That comparability across a season is the strongest practical argument for multispectral over RGB.

Where multispectral runs out

Calibration has to be done well, and even then there is error. Assmann and colleagues, writing up best practice for drone multispectral work, found that sun angle alone shifted NDVI by just over 0.02 within their flight windows, that calibration panels degraded 4 to 10 percent over one field season, and that uncontrolled errors combined could reach about 0.094 NDVI at the one-hectare scale, roughly 10 to 13 percent of peak-season values. They recommend flying within two to three hours of solar noon and recording sky conditions.

NDVI saturates in dense canopy. Viña, Henebry and Gitelson describe NDVI as suffering "a rapid decrease of sensitivity at moderate-to-high densities of photosynthetic green biomass." Gitelson's 2002 paper makes the same point from the other side: near-infrared reflectance levels off or falls once vegetation fraction passes about 60 percent. Over a mature pecan canopy or a pivot crop at full cover, NDVI can read uniformly high while real differences sit underneath. NDRE helps; it does not remove the problem.

An index is not a diagnosis. A low NDVI zone can be water, nitrogen, disease, salinity, compaction, a cover crop or a replant. The index measures reflectance. Deciding what caused it still takes a soil probe, a pressure chamber, a tissue test or someone who knows the block. For pecans, University of Arizona Cooperative Extension's leaf-sampling guide describes leaf analysis as the tool for evaluating nutrient status, with interpretation standards based on samples taken in late July and early August. No aerial index replaces that.

It still needs ground truth. Multispectral can show differences a color photo does not, because it measures wavelengths the eye cannot see. Whether it shows a given problem earlier depends on the problem; it is not a guarantee.

Which one fits which question

Locally, that sorts out roughly like this. Willcox AVA vineyards are row crops with open middles, where much of the useful information is structural: gaps, row-to-row vigor patterns, drip problems, weeds. RGB covers a lot of that, and our pre-harvest scouting article describes what one September flight is for. Pecan orchards around Bowie and San Simon are the opposite: dense mature canopy, where tree count and canopy cover are RGB jobs and NDVI saturation is a real limit; see pecan canopy mapping. Pivot crops in the Sulphur Springs Valley show span and nozzle problems as rings and arcs, visible in RGB once the crop responds; pivot inspection from the air covers that.

The grower stays in charge of the answer

Both kinds of imagery do the same job at bottom: they cover the whole block at once and point at the places worth walking. The value is in the walk that follows. A map with a red patch on it is a question. The answer comes from the person who digs a hole there, pulls leaves, checks the emitters or reads the pressure, and from the grower or agronomist who knows the block's history. We do not give agronomic advice, and nothing on this page is a recommendation for a particular block.

What we offer today

Visible-light orthomosaics and RGB index maps, labeled with the index used, plus GeoTIFF exports and a short written note of what looked different and where, written as observations rather than diagnoses. Multispectral is coming soon; until it launches we do not sell NDVI, and we will not label an RGB map as NDVI.

On the rate card, a vineyard block up to 50 acres is a flat $425 per visit, and a season subscription of six scans from bud break to post-harvest is 25 percent off. If your question genuinely needs near-infrared data now, tell us and we will be straight about whether to wait or use another provider. The agriculture service page has the deliverables, and the contact page is the place to send a block location and the question you want answered.