LATTICE example · Studio · Grape leaves

NDVI says these leaves are healthy. The red edge disagrees.

We photographed 34 grape leaves, healthy to dead, with one 9-camera LATTICE L9 array: twelve narrow bands and a color reference from a single trigger. Pick a layout below, then step through every camera, band and index.

  • 9cameras, one trigger
  • 12narrow bands, 430–958 nm
  • RGBtrue color, same trigger
  • 34leaves, healthy to dead

The studio setup

One leaf at a time, on a backing that stays black in the infrared

Each leaf lay flat on carbon black foam. Foam that looks black can still glow in the near-infrared; this sheet reads 2–4% from 450 to 940 nm, so it adds almost nothing to the leaf's red edge.

All nine cameras used the wider-angle L87 lens (3.5 mm, 90° across). From 1.0 m, each pixel covers about 1 mm of leaf. Settings stayed fixed for the whole shoot.

A T4P-R125 reflectance target sat beside every leaf, and every band was referenced to its panels in the same frame. A DAQ-E-S light sensor recorded the light.

The layouts further down are simulated: each leaf was captured on its own, with the same height, light and exposures, then placed in a sheet of eight at true relative size.

The set as the array's RGB camera sees it: leaf 31 on the carbon black foam beside the T4P target and the DAQ-E-S
The set as the array's own RGB camera saw it, cropped just below the foam: the T4P panel in its case, the DAQ-E-S on the case hinge in the middle of the frame, leaf 31 on the foam.
Halogen domes and LED floods hanging around the LATTICE array
Halogen domes and LED floods around the array, in the middle beside the fan.
  • CamerasLATTICE 9-camera L9 array, every camera with the L87 lens. The modules we chose for this set: six M3M mono (F450, F520, F685, F725, F750, F940) and three M3C color (RGB, RGN, OCN), from the filter options
  • ReferenceT4P reflectance target, in every frame
  • Light sensorDAQ-E-S spectroradiometer
  • LightsFour halogen lamps and four RGBW LED lamps, all at full power: about 3,700 K together at the set, measured by the DAQ-E-S
  • BackingCarbon black foam, 2–4% reflectance from 450 to 940 nm
  • CaptureChloros, 12-bit, gain 0, all nine cameras on one trigger
  • AlignmentCalibrated alignment (depth-aware), every band within a fraction of a pixel at the leaf
The LATTICE L9 array: nine M3 cameras in a three by three block, cooling fins on the side
LATTICE L9The array behind every image on this page: nine M3 cameras in one block, all on one trigger, each with its own filter.See the L9 array

Leaf 19, every band

One trigger, twelve bands plus RGB

Every band the array recorded for leaf 19, in wavelength order. The six mono cameras carry one narrow filter each and the RGN and OCN cameras three each: twelve narrow bands. The RGB camera adds broad blue, green and red, and true color. Values are the leaf's median reflectance (ρ).

Leaf 19, True color
True color FRGBwhite-balanced
Leaf 19, Blue 450
Blue F450430–458 nmρ 0.06
Leaf 19, Blue FRGB
Blue FRGB400–506 nmρ 0.08
Leaf 19, Cyan 490
Cyan FOCN471–509 nmρ 0.08
Leaf 19, Green 520
Green F520504–536 nmρ 0.12
Leaf 19, Green FRGB
Green FRGB480–602 nmρ 0.13
Leaf 19, Green 550
Green FRGN535–565 nmρ 0.16
Leaf 19, Orange 615
Orange FOCN604–626 nmρ 0.10
Leaf 19, Red FRGB
Red FRGB582–662 nmρ 0.11
Leaf 19, Red 660
Red FRGN650–670 nmρ 0.07
Leaf 19, Red 685
Red F685675–698 nmρ 0.08
Leaf 19, Red edge 725
Red edge F725712–738 nmρ 0.31
Leaf 19, Red edge 750
Red edge F750729–763 nmρ 0.35
Leaf 19, NIR 808
NIR FOCN801–815 nmρ 0.36
Leaf 19, NIR 850
NIR FRGN835–865 nmρ 0.36
Leaf 19, NIR 940
NIR F940923–958 nmρ 0.35

Every filter, and the light

mono filters, measuredcolor-camera channels, approximate
02550751004005006007008009001000wavelength (nm)transmission (%)450FRGB BFOCN C520FRGB GFRGN GFRGB RFOCN OFRGN R685725750FOCN NFRGN N940DAQ-E-Slight sensor3,720 K (3,670 – 3,800 K)

Halogen lamps give a smooth spectrum that reaches past 940 nm but little blue; the RGBW LEDs add blue and green and no near-infrared. Together they light every band the array records.

One leaf, three answers

Leaf 31 looks healthy. NDVI agrees. The red edge doesn't.

NDVI compares red with near-infrared. Once a leaf has a fair amount of chlorophyll, red is almost fully absorbed and NDVI flattens out near the top of its scale.

The red edge, the steep climb in reflectance between 700 and 750 nm, keeps moving as chlorophyll falls. For this set we put a 725 nm red-edge camera in the array, one of the filtered modules you can choose for any LATTICE position, so the red-edge chlorophyll index (CIre) can see what NDVI can't.

Leaf 31, Your eye
Your eyeLooks healthy in RGBGreen,
no spots or browning
Leaf 31, NDVI (RGN)
NDVI (RGN)0.64Healthy,
says NDVI
Leaf 31, Red edge (CIre)
Red edge (CIre)0.12Stressed,
says red edge

Three leaves, eight bands

Leaf 20, healthyLeaf 31, NDVI healthy, red-edge stressedLeaf 27, dead
red edge0.00.10.20.30.40.5450550650750850950wavelength (nm)reflectance

At 725 nm leaf 31 reflects 0.38 against leaf 20's 0.26: its red edge has moved toward shorter wavelengths, the classic sign of falling chlorophyll. In red (660 nm) the two read 0.09 and 0.06, and in near-infrared (850 nm) 0.42 and 0.35, so NDVI barely moves.

The six mono cameras plus the RGN's red and near-infrared. F940 sits on a water band, so dry tissue reads brighter there.

Explore the set

Eight leaves, every camera

Pick a layout, then a view. Sweep steps through the bands in wavelength order.

Color
Single bands
Indices

Outlined: NDVI says healthy, the red edge says stressed. Dead tissue (no chlorophyll) is drawn red in every index map.

Health ladder: eight leaves
True colorFRGB camera · white-balanced

The RGB camera, white-balanced on the T4P panels with the camera's own color matrix, then given more exposure (×1.5), local contrast and deconvolution, and enlarged. What your eye sees.

RGN false colorFRGN · 660 / 550 / 850 nm

The RGN camera's own three channels. Near-infrared rides in the blue channel, so leaves with healthy cell structure glow blue.

OCN false colorFOCN · 615 / 490 / 808 nm

Orange, cyan and near-infrared, with near-infrared in blue. A relative map: the OCN's cyan correction was tuned for this set, and its cyan is visibly noisy.

Blue 450M3M F450 · 430–458 nm

Chlorophyll and carotenoids both absorb blue, so every living leaf is dark.

Blue FRGBM3C RGB · 400–506 nm

The color camera's blue channel. Broad and overlapping with its neighbors: a color reference, not a calibrated band.

Cyan 490M3C OCN · 471–509 nm

Cyan, the edge of carotenoid absorption. A relative map on this set; the grain is real sensor noise.

Green 520M3M F520 · 504–536 nm

The rising side of the green peak. Yellowing leaves brighten here first.

Green FRGBM3C RGB · 480–602 nm

The color camera's green channel: a color reference, not a calibrated band.

Green 550M3C RGN · 535–565 nm

The green reflectance peak that makes leaves look green. A relative map on this set.

Orange 615M3C OCN · 604–626 nm

Orange: chlorophyll absorbs strongly, so healthy leaves stay dark and browning ones light up. A relative map on this set.

Red FRGBM3C RGB · 582–662 nm

The color camera's red channel: a color reference, not a calibrated band.

Red 660M3C RGN · 650–670 nm

Red, near chlorophyll's strongest absorption. The red half of NDVI.

Red 685M3M F685 · 675–698 nm

The bottom of the chlorophyll well, just before the red edge begins.

Red edge 725M3M F725 · 712–738 nm

On the red edge itself. Reflectance climbs steeply here, and where it climbs moves with chlorophyll: the band that catches leaf 31.

Red edge 750M3M F750 · 729–763 nm

The top of the red edge, where the leaf turns bright.

NIR 808M3C OCN · 801–815 nm

Near-infrared, the NIR half of OCN NDVI. A relative map on this set.

NIR 850M3C RGN · 835–865 nm

The near-infrared plateau: the leaf's cell structure scatters it back. The NIR half of NDVI and CIre.

NIR 940M3M F940 · 923–958 nm

Far near-infrared, on a water absorption band: dry, brown tissue reads brighter here than green tissue.

NDVI · RGN(NIR 850 − Red 660) / (NIR 850 + Red 660)

The standard vegetation index, from the RGN camera alone. It separates living from dead well, but flattens out on any leaf with a fair amount of chlorophyll.

NDVI · OCN(NIR 808 − Orange 615) / (NIR 808 + Orange 615)

The same idea from the OCN camera's orange and near-infrared. A relative map on this set.

GNDVI(NIR 850 − Green 520) / (NIR 850 + Green 520)

Green in place of red, from the F520 camera beside the RGN's near-infrared. Green is absorbed less than red, so GNDVI keeps responding a little further up the chlorophyll range.

NDRE(NIR 850 − Red edge 725) / (NIR 850 + Red edge 725)

The normalized red-edge difference. It needs the F725 camera beside the RGN.

CIre, red-edge chlorophyll indexNIR 850 / Red edge 725 − 1

Tracks chlorophyll across its whole range, where NDVI flattens. Brown pigment makes a red-edge slope too, which CIre would read as healthy; that is one reason dead tissue is drawn red.

The numbers

Every leaf in the layouts

Median of each leaf's pixels, edge trimmed. Values compare the leaves on this set; they are not field values. CIre values in red are the leaves it rates very stressed while NDVI rates them 0.5 or more.

Leaf NDVI (RGN) NDVI (OCN) GNDVI NDRE CIre
Leaf 15deep green
0.74 0.71 0.62 0.14 0.32
Leaf 20deep green
0.72 0.70 0.60 0.14 0.33
Leaf 13green, one browning lobe
0.70 0.62 0.54 0.09 0.20
Leaf 19green
0.67 0.54 0.51 0.07 0.15
Leaf 16green, purpling at the edge
0.64 0.56 0.60 0.11 0.26
Leaf 31pale green
0.64 0.51 0.57 0.06 0.12
Leaf 12green with yellow patches
0.58 0.44 0.51 0.07 0.15
Leaf 25pale olive
0.56 0.41 0.46 0.12 0.27
Leaf 10pale yellow-green
0.51 0.34 0.44 0.04 0.07
Leaf 6yellow and brown blotches
0.47 0.40 0.50 0.08 0.1658% dead tissue
Leaf 22brown
0.45 0.39 0.57 0.08 0.1892% dead tissue
Leaf 24half green, half brown
0.43 0.29 0.43 0.04 0.08
Leaf 21brown
0.34 0.31 0.53 0.08 0.1891% dead tissue
Leaf 29half brown, half dry
0.29 0.35 0.54 0.07 0.1573% dead tissue
Leaf 28dry and brown
0.16 0.23 0.44 0.07 0.15100% dead tissue
Leaf 27dead
0.10 0.11 0.30 0.02 0.04100% dead tissue

All 34 leaves

The whole set, healthy to dead

Every leaf from the shoot on one sheet, at true size relative to each other, in leaf-number order. Switch between true color, NDVI and the red edge to see which leaves each one calls stressed.

All 34 grape leaves from the set, RGB

How it was made

Capture to color map

  1. CaptureAll nine cameras fire on one trigger. 12-bit, gain 0, each camera's exposure fixed so nothing clips. Each tile sits where its camera sits in the array, seen from behind, as if looking down on the set over the cameras.
  2. ReflectanceEach band is divided by the T4P target's 94% and 50% panels in its own frame, capture by capture, before anything is moved: the leaf still sits a little differently in every camera.
  3. RegistrationThen the bands are registered to each other. Standard alignment, one warp per camera, is exact at a single depth; on this shoot it left them a median of 13.7 px apart at the leaf. Calibrated alignment works at the leaf's own depth: 0.15 px, so edges do not fringe.
  4. Color mapsOne red-to-green ramp for every index, green at the healthy end, stretched from the 2nd to the 98th percentile of all 34 leaves. Dead tissue (no chlorophyll, PSRI above 0.10) is drawn red in every index map.