Affordable field high-throughput phenotyping - some tips

J.L. Araus, A. Elazab, J. Bort, M.D. Serret, J.E. Cairns

Affordable field high-throughput phenotyping - some tips

Outline

Why field phenotyping?

Some examples of traits and tools

Affordable high-throughput phenotyping

Outline




Crop breeding pillars

Araus and Cairns 2014

INIA 5

ClimaticData

EnvironmentData

Pedigree

PhenotypicData

Molecular Markers

Model

Estimation of highest value

crosses

Genotyping in next

generations

Estimation of Genomic Breeding Values

Genomic Selection

Crop Database

Sequence Server

Marker assisted selection

Molecular Breeding

After Passioura (2006) Funct. Plant Biol. 33,

Outline




Different categories of traits


Proximal sensingLaboratory analyses Near infrared reflectance spectroscopy



How to implement proximal sensing in practice?

Phenomobiles

Rebetzke et al. 2013 FPB 40: 1-13

Aerial platforms

Zimbabwe February 2013

Different categories of imaging systems for remote-sensing evaluation of vegetation and examples of prototypes capable of being carried by UAPs of limited payload are shown: A) RGB/CIR cameras; B) Multispectral cameras; C) Hyperspectral VIS-VNIR imager; D) Longwave infrared cameras or thermal imaging cameras; E) Conventional digital (RGB) cameras.

Outline




Proximal sensing: Low cost approaches

http://www.farmworks.com/products/greenseekerhandheld

1 m

green biomass

Digital photography

http://www.google.co.uk/imgres?imgurl=http://www.product-reviews.net/wp-content/userimages/2008/04/the-nikon-coolpix-l15-digital-camera.jpg&imgrefurl=http://www.product-reviews.net/2008/04/04/the-nikon-coolpix-l15-digital-camera-easy-auto-mode-for-great-pictures/&usg=__qwTWYSS12Hhl4CEJLI80y0bOeJM=&h=300&w=400&sz=17&hl=en&start=19&itbs=1&tbnid=79vyJgn3HJ4JlM:&tbnh=93&tbnw=124&prev=/images?q=digital+camera&hl=en&gbv=2&tbs=isch:1

•IHSIntensity, Hue, SaturationPractical for image analysis

0º

120º240º

Hue wheel:

Numerical representation of color

•RGB: Red, Green and Bluerelated with color reproduction by computer screens, etc.

•CIE-lab~ sensitivity of human visual system Consistent distance practical for arithmetics

CIE-Lab

There are a number of different systems for representing a given color.

Overview of the process

Picture-derived Vegetation Indices calculated by BreedPix

•Components of the average color of the image•H (from HIS color-space)•a* (from CIE-Lab color-space)

•Counting green pixels

•Green Area (% pixels with 60<Hue<120)•Greener Area (% pixels with 80<Hue<120)

Casadesus and Villegas J. Integ. Plant Biol. 2013

Casadesus and Villegas 2013 J. Integ. Plant Biol.

Casadesus et al. Ann. Appl. Biol. 2007

Validation: Pic-VIs correlate with leaf area (however, the relationship may change with phenology)

The relationships between LAI and Hue, a* and u* were similar to these

Casadesus and Villegas 2013 J. Integ. Plant Biol.

http://www.google.es/url?sa=i&rct=j&q=greenseeker&source=images&cd=&cad=rja&docid=zQB8gwlBOvuDOM&tbnid=yUh2Z4t3WNVtjM:&ved=0CAUQjRw&url=http://www.farmworks.com/products/greenseekerhandheld&ei=XA5fUYf5FIKN0AWk1YCYBg&bvm=bv.44770516,d.ZGU&psig=AFQjCNEdlfKKoLsFJaflfTyWidk6XiNT4A&ust=1365270488684783

Cereal leaf beetle Oulema melanopus L. (Coleoptera, Chysomelidae). Started in May.

Yellow rust Puccinia striformis f. sp. tritici. A very virulent new strain in Europe named Warrior/Ambition, first cited in England in 2011. Started mid-April.

Pests and diseases monitoring

Correlation coefficients of Grain Yield (GY) with leaf chlorophyll content and color parameters calculated from the digital images at jointing (no infested), heading (mildly infested) and two weeks post-anthesis (severely infested) across 12 wheat genotypes.

Jointing Heading Post-anthesis

GY Chl GY Chl GY Chl

Chl -0.39* ― -0.29 ―

0.54*** ―

Intensity 0.1 0.27

0.23 -0.15

-0.04 0.01

Hue -0.17 0.2 -0.04 0.37* 0.87*** 0.66***

Saturation 0.13 -0.22 -0.09 -0.42* -0.68*** -0.50**

Lightness 0.19 0.11 0.23 -0.25 0.14 0.16

a* -0.08 0.09

-0.12 0.52**

-0.88*** -0.72***

b* 0.14 -0.18 0.09 -0.55*** -0.45** -0.30

u* 0.01 -0.03

-0.14 0.38*

-0.87*** -0.72***

v* 0.15 -0.15 0.15 -0.54*** -0.08 0.01

GA -0.2 0.13

0.33 -0.32

0.87*** 0.72***

GGA -0.22 0.21 0.36* -0.14 0.89*** 0.57***

Chl, flag leaf chlorophyll content (SPAD value); Intensity hue saturation (IHS) color space and each of its components; lightness, a* and b*, color component from Lab; u* and v*, color component from Luv; GA, green area; GGA, greener area. (*, P< 0.05; **, P < 0.01 and ***, P < 0.001, n = 36).

0.0 .2 .4 .6 .8 1.0

Gra

in y

ield

(t ha

-1)

0

2

4

6

8

10

GA

GGA

r2 = 0.74***

r2 = 0.79***

GA and GGA

Relationships between G and GAA against grain yield across a set bread wheats

MLN in hybrid maize field in Tanzania – Dr. B.M. Prasanna

Potential applications

Conclusions: Advantages of Pic-VIs

•Very low sampling cost and high resolution•Sampling [almost] not conditioned by weather•Calculation of Pic-VIs can be automated

(a trial with hundreds of plots can be sampled and processed in the same day)

•Good repeatability and representativity (taking several pictures per plot allows accounting for its spatial variability)

•Validated as Vegetation Indices (before anthesis, GA, a* and u* show R2>0.8 with LAI, GAI and CDW)

Conclusions: Comparison between Pic-VIs

•GA, GGA, a* and u* are more robust than Hue to environmental conditions•GA and GGA are almost unaffected by soil color•GA is the easiest to interpret

(% soil covered by green canopy)

•GGA may be useful at late grain-filling stages to exclude pixels representing senescent leaves

•As other VI, they get saturated at high LAI (e.g. at stages with much green biomass, under irrigated

conditions)

•As other VI, they get disturbed after anthesis by the structure of the canopy•Effect of spikes•Vertical distribution of green biomass

Conclusions: Limitations of Pic-VIs

Gitelson et al. 2002 Remote Sensing of Environment 80

Normalized Green Red Difference Index (NGRDI)NGRDI = [(Green – Red)] / (Green + Red)]

Tucker, C.J., 1979. Remote Sensing of Environment 8

NGRDI = [(Green – Red)] / (Green + Red)]

• Image analysis was performed with ImageJ 1.46r (http://imagej.nih.gov/ij/).

• ImageJ is a public domain Java image processing and analysis program created by NIH Image.

• The original images stored by the camera were converted to its main 3 channels (red-green-blue)

http://imagej.nih.gov/ij/

Anthesis Grain filling

Durum wheat (Sula)

SI

RF

Anthesis Grain filling

Genotype Sula

SI

RF

Relationship between Normalized Difference Vegetation Index (NDVI.2, left A, B) and the Normalized Green Red Difference Index (NGRDI.2, right C, D) at anthesis versus grain yield (GY) and aerial biomass (AB) at maturity.

Aerial picture about three weeks after anthesis of a maize trial with 6 different N fertilization treatments (Fontagro Project. Algerri, Lleida, Spain)

Experimental design

Tetracam mini MCACanon Eos 5D

Beyond vegetation indicesOther parameters could be estimated from digital images.

•Total soil cover (green+dry vegetation)

•Physiological status(N-content, Chl,...)from the color of the green area only.

•Agronomical yield components (e.g. spikes m-2)



Technique IRMS EA AACC Method NIRS-prediction

Parameter 13C 18O N content Ash content 13C* 18O Ash N

Cost per sample 10€ 20€ 3€ 1.5€ 0.5€

Time <10 min <10 min <10 min ≈24 h ≈3 min

Equipment EA-IRMS EA Muffle furnace NIR spectrometer

*previously reported by Clark et al. 1995; Ferrio et al. 2001; Kleinebecker et al. 2009

NIRS a surrogate analysis of 13C

Calibration Samples

Measured 13C Discrimination (o/oo)

12 13 14 15 16 17 18

NIR

S P

redi

cted

13C

Dis

crim

inat

ion

(o /

oo)

12

13

14

15

16

17

18 N = 135Y = 2.10 + 0.86xr2 = 0.86***RMSEP = 0.46

Validation Samples

Measured 13C Discrimination (o/oo)

12 13 14 15 16 17 18

N = 179Y = 1.48 + 0.90xr2 = 0.82***RMSEP = 0.55

Breda RainfedTel Hadya RainfedTel Hadya Irrigated

Breda RainfedTel Hadya RainfedTel Hadya Irrigated

NIRS prediction of δ13C and δ15N

Kleinebecker et al. 2009 New Phytologist 184: 732-739

Trait N Mean SD Range CV SEC R2c SECV R2cv RPD SlopeNkernels 126 1.81 0.24 1.15-2.38 13.4 0.09 0.87 0.09 0.87 2.76 0.90Nleaves 152 1.57 0.22 1.04-2.05 14.1 0.10 0.80 0.12 0.72 1.86 0.80ASHkernels 129 1.47 0.24 0.91-1.90 16.2 0.11 0.79 0.13 0.72 1.89 0.79ASHleaves 150 14.31 2.89 8.78-21.46 20.2 0.54 0.97 0.65 0.95 4.42 0.9818Okernels 128 31.69 1.43 28.05-34.99 4.5 0.82 0.66 1.04 0.49 1.38 0.6618Oleaves 151 32.97 1.25 29.37-36.46 3.8 0.79 0.54 1.00 0.38 1.26 0.57

NIRS prediction of ash content and δ18O

Trait N Mean SD Range CV SEC R2c SECV R2cv RPD SlopeNkernels 73 1.73 0.24 1.15-2.24 13.71 0.07 0.87 0.08 0.87 2.79 0.87Nleaves 86 1.49 0.22 0.92-1.95 14.71 0.08 0.86 0.09 0.83 2.46 0.86ASHkernels 75 1.37 0.27 0.91-1.80 19.71 0.10 0.82 0.14 0.70 1.92 0.82ASHleaves 84 14.89 2.92 10.02-20.82 19.64 0.49 0.97 0.78 0.93 3.76 0.9818Okernels 70 31.03 1.05 29.06-33.53 3.37 0.50 0.77 0.76 0.51 1.38 0.77N, number of samples; SD, standard deviation; CV, coefficient of variation; R2c, determination coefficient of calibration; R2cv,

determination coefficient of cross-validation; RPD, ratio of performance deviation; SEC, standard error of calibration; SECV, standard error of cross calibration. All correlations were significant at P<0.001 level.

Calibration statistics for hybrid sample set for leaf and kernel N and ash content and kernel 18O

Calibration statistics for global sample sets (including inbred lines and hybrids) for N, ash content and 18O in kernels and leaves

Conclusions

There are different low-cost methodological approaches that makes high-throughput field phenotyping affordable for NARS

Ackowledgements

• Affordable field-based high Throughput Phenotyping Platforms (HTPPs). Maize Competitive Grants Initiative. CIMMYT

• Adaptation to Climate Change of the Mediterranean Agricultural Systems – ACLIMAS.. EuropeAid/131046/C/ACT/Multi. European Commission

• Durum wheat improvement for the current and future Mediterranean conditionsMejora del trigo duro para las condiciones mediterráneas presentes y futuras. AGL2010-20180 Spain.

• Breeding to Optimise Chinese Agriculture (OPTICHINA). FP7 Cooperation, European Commission - DG Research. Grant Agreement 26604 .

http://www.optichinagriculture.com/



Organizers: Chinese Academy of Agricultural Sciences and the OPTICHINA Project




Many thanks….

Education

Affordable field high-throughput phenotyping - some tips