Multispectral survey

The multispectral survey is a precision-agriculture cornerstone. DroneField computes NDVI, GNDVI and NDRE from imagery captured with DJI Phantom 4 Multispectral (P4M) and Mavic 3 Multispectral (M3M).

What does “multispectral” mean?

A normal RGB camera records 3 spectral channels (Red, Green, Blue). A multispectral camera records 4–5 channels:

Band	Wavelength	What it’s good for
Blue	450 nm	P4M only; reconstructed for M3M
Green (G)	560 nm	Chlorophyll content
Red (R)	650 nm	Chlorophyll absorption
Red Edge (RE)	730 nm	Chlorophyll saturation, canopy health
NIR	840 nm	Leaf structure, canopy density

Vegetation indices (NDVI, GNDVI, NDRE) are difference ratios of these bands (see Vegetation indices).

Why are multispectral images black & white? A 3-minute video explainer:

P4M vs M3M

Property	P4M (2018)	M3M (2022)
RGB camera	2 MP (small)	20 MP
MS bands	5 (B, G, R, RE, NIR)	4 (G, R, RE, NIR)
Auto-exposure	Manual (sensor)	Automatic
RTK	Yes	Yes
Flight time	~25 min	~45 min
Take-off mass	1.5 kg	0.95 kg

The M3M is significantly more modern — faster, longer flight, much better RGB. For a fresh purchase the M3M is the right call.

P4M’s Blue band absent from M3M: DroneField reconstructs the Blue band from the JPEG — with limited accuracy, but enough for the classical indices.

Planning

Altitude

• 60–80 m AGL for typical agricultural NDVI
• Higher (100 m+): the alignment noise between bands can grow

Overlap

For multispectral, high overlap is even more important — each band is a separate image, and band-to-band registration has to be tight:

• Along-track: 80% (or more)
• Cross-track: 75% (or more)

Weather

Overcast is mandatory (officially recommended). Why:

• Direct sun casts shadows — uneven reflectance
• Thin clouds change moment-to-moment, introducing differences between neighbouring images

Current DJI drivers can’t fully compensate, nor can DroneField.

Time of day

10:00–14:00 (Europe). The sun’s perpendicular position reduces shadow length.

Calibration panel

MANDATORY on both drones (P4M and M3M):

1. Before take-off — photograph the panel on the ground, horizontally, ~1–2 m above ground, with the drone stationary (DJI Pilot 2 prompts automatically).
2. After take-off, immediately — still over the field, same settings.

Calibration shots calibrate the band-to-band reflectance scale. Without them, NDVI absolute values are not credible.

DroneField detects calibration panels and excludes them from the render pool (but uses them for reflectance computation).

Image grouping

DJI M3M captures 5 images per exposure:

• DJI_0001.JPG — RGB
• DJI_0001_MS_G.TIF — Green band
• DJI_0001_MS_R.TIF — Red band
• DJI_0001_MS_RE.TIF — Red Edge band
• DJI_0001_MS_NIR.TIF — NIR band

DroneField groups them automatically by prefix. Don’t rename the files!

Processing time

For a 100-image Mavic 3 M RTK RGB survey:

Configuration	GSD	Time
Intel i9 / 16 GB RAM / Windows	6 cm	< 40 seconds
Mac M4 / 24 GB RAM / macOS 26	5 cm	< 30 seconds

Multispectral processing adds roughly 40-60% to the runtime (separate rendering of 4-5 bands).

Reading the result

What matters on NDVI / GNDVI / NDRE maps are the relative differences, not the absolute values. An NDRE of 0.55 alone says little — but if the field average is 0.62 and there’s a patch around 0.40, that’s stress or weakness there.

Typical interpretation mistakes

• “NDVI is above 0.85 everywhere — depleted” — no, that’s saturation, no differentiation across the dense canopy. NDRE differentiates.
• “Banded NDVI” — weather change, not the plant.
• “Low NDVI in shadowed area” — tree, wire and cloud shadow all lower the index. Not stress, just illumination.

When to re-survey?

Index changes should be tracked at 2–3 week intervals during the season. Within a “survey season”:

• Spring, growth phase — first survey
• Before flowering — second survey
• Before harvest — third survey (if relevant)

Save these as separate projects in DroneField, not the same one.