The View From Up Here is Great!

Ian MacRae, Tim Baker, & Robert Koch

Dept of Entomology

Univ. of Minnesota

The View From Up

Here is Great!Remote Sensing and Drones in

Pest Management

Manitoba Agronomists Conference

Winnipeg, Manitoba, Dec 14-15, 2016.

Plant Light

InteractionDifferent wavelengths

reflected by pigment,

internal leaf structure &

H2O content.

Plant

geometry

- Orientation,

density &

distribution

UAVs fill a platform

gap, provide more

options/tradeoffs for

resolution, speed, &

immediacyHigher resolution than a

satellite or plane, more

immediate/convenient,

but covers smaller

footprint.

Less detailed info than a

person/ATV, but covers a

bigger footprint.

Economics depend on

what’s being sensed &

how

Hyperspectral Sensors – measure many bands at

narrow intervals (e.g. 1-3 nm / interval)

Can get hyperspectral sensors / imagers, BUT $$$$

Multispectral

Sensors –

measure multiple

broader blocks,

several to many

wavelengths in width

Most UAS based sensors / imager are multi-spectral

GoPro Hero3

ADC TetraCam

Sentera

NIR imagers

Imaging sensors

Tend to be multi-spectral but

hyperspectral available (costly)

3-D Robotics Iris

Flight planning

200’

5 run

lengths

= 40%

image

overlap

Stitching the mosaic

Individual images of

plots ‘stitched’ into

mosaic representing

entire field using

AGISoft PhotoScan

(AGISoft LLC St

Petersburg RU)

Stitching imagery

Basically 3 tools• Hyperspectral

– Usually reflectance = spectral curves

– Often non-image (spectroradiometers =

actually numerically measure the amount

of incident energy being reflected from the

target)

• Multispectral

– Most imaging sensors in the cost window

for scouting will tend to be multi-spectral

• Visible

Visible data• Currently collaborating

on stand counts in

experiment plots

• Think hail damage,

herbicide drift, planter

skips, flooding, heavy

defoliation….

• Any problem you can easily see from the

ground, you can probably see from the air.

http://www.prairieagri.com/cvaphoto.html

Estimating defoliation• Takes time

• Doesn’t cover entire field

• Sometimes are guestimates

Canopy segmentation• Process - Identify

spectral reflections of

representative non-

plant areas to create

selection sieve values

for selection decisions

• Delineate the area of

interest and perform

analysis.

• % canopy coverage

then calculated,

geocoded images, so

actual area can be

determined

• Most image analysis software have “supervised

classification” – you ‘train’ the software to

differentiate between vegetative and non-

vegetative pixels (e.g. soil)

• based on min & max red & NIR values (e.g.

red<6, NIR <20), non-vegetative pixels rendered

to a user-specified solid color while leaving

pixels representing vegetative material

unchanged.

P<0.001, R2=0.587

With VIS dataR2 = 0.893, p < 0.001

Non-visible pest damage

• Disease, aphids, whitefly (sap feeders)

• Require spectral remote sensing

– Looking at other factors indicative of the

problem

– We can’t see/count aphids but can detect

changes they cause in plant physiology /

structure

Spectral reflectance

Cell

structure Water content

Dominant factors controlling leaf reflectance

Lowest AIC is better

Sugarbeet Root

Maggot• Sugarbeet Root Maggot

(SBRM) feeds on root sof

sugarbeet

– Difficult to scout

(underground)

– Stresses plant within

season

• Researching ability to

remotely scout for SBRM

Low SBRM Populations

High SBRM Populations

Tease out the varieties in

the spectral data and you

see the difference

Disease reflectance• Also in that 700-900 range

• BUT – disease affects stomatal opening

and diseases, like Downy Mildew, have

been demonstrated as lowering leaf

surface temperature (already used to

remotely sense DM)

Thermal IR

cameras small

enough to be

mounted on small

UAS

Chaerle, Laury, and Dominique Van Der

Straeten. 2000. ‘Imaging Techniques And

The Early Detection Of Plant

Stress’. Trends In Plant Science 5 (11):

495-501. doi:10.1016/s1360-

1385(00)01781-7.

Symptom models• Symptoms leading to probability

diagnoses

– Climate (temp & rH)

– Time of year

– condition / variety of crop

– Reflectance & other remotely

sensed data (temp by FLIR)

GoPro Hero3

Sony NEX-T5

ADC TetraCam

Sentera

NIR imagers

Imaging sensors

Imagers tend to be multi-

spectral but hyperspectral

available (costly)

Spectral

sensitivity of

digital cameras

RGB – 3 bands

Some bleed of energy from one

wavelength into neighboring

wavelengths

Calibration vital

Imaging sensor issues

• Lens distortion

– Higher the altitude, wider the footprint BUT >>

barrel distortion

– Fixed focal length lenses

• Quantum efficiency of sensor

(CCD/CMOS) at various wavelengths

– Take caution at the edges of sensitivity

• Format of saved RAW data

Data acquisition issues• Canopy structure

– Wind, plant architecture, etc. all influence reflectance

of plant

• Shadow– position of sun (data acquisition best @ solar zenith)

• Incident light– Cloud cover and reflection can influence reflectance

(70%+ clear skies, incident light sensor)

Some e.g... Many factors can introduce variation into reflectance

Gravity sucks!

What goes up, will come down

– sometimes in a spectacular

fiery manner…

Acknowledgements -

Field, Lab & UAS Staff – Christian Halos, Guthrie Dingman,

Abbie Anderson, Alex McGregor, Camila Inacia Costa,

Nicole Dudycha, Joe Wodarek

The staff at the UMN Sand Plains Research Farm (Becker)

and the Northwest Research & Outreach Center (Crookston)

This research was supported by funding from the Minnesota

Department of Agriculture Crop Research Program, MDA

Specialty Crop Research Block Grant Program, the Northern

Plains Potato Growers Assoc. and MN Area II Potato

Growers Assoc., Univ. of Minnesota MNDrive, MDA Rapid

Agricultural Response Fund,