Near infrared spectroscopy: A rapid nondestructive method for - - PowerPoint PPT Presentation

Near infrared spectroscopy: A rapid nondestructive method for measuring wood properties and its application to tree breeding

• L. R. Schimleck

Warnell School of Forestry and Natural Resources The University of Georgia

— 2 —

Overview

• Introduction to NIR spectroscopy
• Rapid estimation of pulp yield by NIR spectroscopy
• Multiple-species, multiple-site calibrations
• Application of calibrations to new sites
• Estimating whole-tree properties using core spectra
• Estimation of the wood properties of radial strips
• Field based NIR spectroscopy
• Estimation of genetic parameters using NIR data
• Conclusions

— 3 —

Introduction to NIR spectroscopy

Increasing Frequency

200nm 380nm 800nm 2500nm 25,000nm

X-Ray UV Visible

NIR

IR FIR, Microwave

Increasing Wavelength

50,000 cm-1 12,500 cm-1 4,000 cm-1 400 cm-1

• Frequency = 1 / wavelength
• NIR spectrum just above visible region of the electromagnetic

spectrum (EMS) 800 to 2500 nm

• Overtones, combinations of fundamental IR vibrations (stretch, bend)

— 4 —

Introduction to NIR spectroscopy

1100 1300 1500 1700 1900 2100 2300 2500 Log (1/Reflectance) 0.1 0.2 0.3 0.4 0.5 0.6 Wavelength (nm)

O-H N-H S-H N-H O-H N-H C-H vibr ation thr

• ughout spectr

um

— 5 —

Introduction to NIR spectroscopy

Estimation of a parameter involves the following steps:

• Collect spectra of calibration samples
• Develop a calibration (regression)

(y = B0 + X1*B1 + X2*B2 + ………..+ XN*BN)

• Collect NIR spectra of test (or unknown) samples
• Estimate parameter of interest for test set samples

using the calibration

— 6 —

Rapid estimation of pulp yield by NIR

• Earliest work reported in the late 1980’s
• Wood chemistry (pulp yield, cellulose and lignin)
• Pulp yield particularly important and a rapid

method for its estimation had long been sought

• Some good results obtained, but calibrations tested
• n samples drawn from the same population
• Calibration / prediction statistics variable –

accuracy of data and its range both important

— 7 —

Rapid estimation of pulp yield by NIR

Many practical questions raised:

• Are multi-site, multi-species calibrations possible?
• Can a calibration from one site be used to

accurately estimate the wood properties of samples from a different site?

• Can whole-tree wood properties be estimated using

NIR spectra from cores?

— 8 —

Multiple-species, multiple-site calibrations

Common in agriculture, 20,000+ samples, multiple countries represented

• Yet to be achieved for wood – NIR research

Potential reasons:

• NIR applied to wood for only a short time
• Difficulty and cost of measuring pulp yield
• Pulping methods differ between laboratories

— 9 —

Multiple-species, multiple-site calibrations

• Garbutt et al. (1992) - 13 eucalypt species, 1 hybrid
• Michell (1995) - Tasmanian E. globulus, 10 locations
• Michell & Schimleck (1998) – Tasmanian E.

globulus and E. nitens, multiple sites

• Schimleck et al. (2006) - 7 eucalypt species, 5

hybrids from 3 locations in Brazil

• Hodge & Woodbridge (2004) – 5 pine species
• Ensis – 700+ samples, multiple eucalypt species

— 10 —

Multiple-site E. nitens calibration

8 factors R2 = 0.91 SEC = 0.64 SECV = 0.78 RPDc = 2.79

44 46 48 50 52 54 56 44 46 48 50 52 54 56 NIR-estimated pulp yield (%) Measured pulp yield (%)

Source: SCHIMLECK, L.R.; KUBE, P.D.; RAYMOND, C.A.; MICHELL, A.J.; FRENCH, J. 2006: Extending near infrared reflectance (NIR) pulp yield calibrations to new sites and species. J. Wood Chem. Technol. 26: 299-311.

— 11 —

Multi eucalypt species, multiple-site calibration

R2 = 0.90 35 40 45 50 55 60 65 35 40 45 50 55 60 65 NIR-predicted pulp yield (%) Measured pulp yield (%)

Source: G. Downes, Ensis

— 12 —

GLOBAL versus LOCAL calibrations

LOCAL approach – build a specific equation to predict a given property for a new sample

• The new calibration is obtained using samples selected from a large

database on the basis of their similarity to the unknown

• Several studies have demonstrated that LOCAL calibrations provide

smaller predictive errors than GLOBAL calibrations

• Provides the benefits of using a GLOBAL strategy (large database

encompassing the variation), with the accuracy of specific calibrations

• Establishment of a large database (1000’s of samples) is critical

GLOBAL approach – build large data set, hope to include all sources of variation

— 13 —

Application of calibrations to new sites

A true test of a calibration……..

• Performance of the calibration may suffer owing to

multiple differences between sites

• Presently no way of testing if a calibration will

accurately predict wood properties or not

• Findings indicate that relationship between lab

measured data and NIR predicted data will be good

• Addition of a small number of samples from the

new location will reduce predictive errors

— 14 —

PPY for Southern Tas using Northern Tas calibration

46 48 50 52 54 56 45 47 49 51 53 55 Laboratory determined pulp yield (%) NIR predicted pulp yield (%) R2 = 0.87 9 of the top 12

Source: SCHIMLECK, L.R; RAYMOND, C.A.; BEADLE, C.L.; DOWNES, G.M. KUBE, P.D.; FRENCH, J. 2000: Applications of NIR spectroscopy to forest research. Appita J. 53: 458-464.

— 15 —

PPY for Gog (Tasmania) using Northern Tas calibration

Rp

2 = 0.70

SEP= 4.60 RPDp = 0.37

46 48 50 52 54 56 58 46 48 50 52 54 56 58 NIR-predicted pulp yield (%) Measured pulp yield (%)

Source: SCHIMLECK, L.R.; KUBE, P.D.; RAYMOND, C.A.; MICHELL, A.J.; FRENCH, J. 2005: Estimation of whole-tree kraft pulp yield of Eucalyptus nitens using near infrared spectra collected from increment cores. Can. J. For. Res. 35: 2797-2805.

— 16 —

PPY for Gog (Tasmania) using Northern Tas calibration

Rp

2 = 0.77

SEP = 1.03 RPDp = 1.65

48 50 52 54 56 58 48 50 52 54 56 58 NIR-predicted pulp yield (%) Measured pulp yield (%)

Source: SCHIMLECK, L.R.; KUBE, P.D.; RAYMOND, C.A.; MICHELL, A.J.; FRENCH, J. 2005: Estimation of whole-tree kraft pulp yield of Eucalyptus nitens using near infrared spectra collected from increment cores. Can. J. For. Res. 35: 2797-2805.

— 17 —

Sample selection

How to select which samples to add to the calibration?

• Select trees that represent different diam classes,

provenances, families, clones, site conditions etc.

• Use an existing calibration to predict the property of

interest and select samples that encompass the range

• Analyse the spectra to identify unique samples
• Selected samples are analysed and used to update the

existing calibration (small number required)

— 18 —

Whole-tree calibrations based on core spectra

Whole-tree data and NIR spectra from cores used to develop wood property calibrations = non destructive estimation of wood properties

• Schimleck et al. (2006) – calibrations based on NIR

spectra from milled whole-tree chips and cores from 0.65 and 1.30 m provided similar results

• Similar findings for hybrid poplars (western USA)

and E. nitens (Tasmania)

— 19 —

Comparison of Aracruz calibrations

0.2 0.4 0.6 0.8 1 D e n s i t y N a O H C h a r g e P e n t

• s

a n s P u l p y i e l d S p e c i f i c C

• n

s T

• t

a l l i g n i n

R2

whole-tree 0.65 m 1.30 m

Source: SCHIMLECK, L.R.; REZENDE, G.D.S.P.; DEMUNER, B.J.; DOWNES, G.M. 2006: Estimation of whole-tree wood quality traits using near infrared spectra collected from increment cores. Appita J. 59: 231-236.

— 20 —

Hybrid poplar – core versus whole-tree

5 factors R2 = 0.96 SEC = 0.35 52 53 54 55 56 57 58 52 53 54 55 56 57 58 NIR fitted pulp yield (%) Lab determined pulp yield (%) 6 factors R2 = 0.90 SEC = 0.55 52 53 54 55 56 57 58 59 52 53 54 55 56 57 58 59 NIR fitted pulp yield (%) Lab determined pulp yield (%)

Whole-tree chips Increment cores

Source: SCHIMLECK, L. R.; PAYNE, P.; WEARNE, R. H. 2005: Determination of important pulp properties of hybrid poplar by near infrared spectroscopy. Wood and Fiber Science 37: 462-471.

— 21 —

Summary of the calibration options

must be pulp tested * Most accurate rankings * Maximal cost as all trees * Most accurate estimate of yield Site specific calibration * Improved ranking and identification of top trees * Some additional pulping required (minimal) * Estimated yield closer to true yield Enhanced calibration * Able to identify majority of top trees * Rankings generally OK * Yield may be under or over estimated * Minimal cost as no additional pulping is required Existing calibration Disadvantages Advantages Option

Source: SCHIMLECK, L.R.; KUBE, P.D.; RAYMOND, C.A.; MICHELL, A.J.; FRENCH, J. 2005: Estimation of whole-tree kraft pulp yield of Eucalyptus nitens using near infrared spectra collected from increment cores. Can. J. For. Res.35: 2797-2805

— 22 —

Estimation of wood properties of radial strips

• NIR spectra collected from

radial strips (10 mm)

• Air-dry density
• Microfibril angle
• Estimated stiffness
• Tracheid morphology
• Tracheid length
• Cellulose
• Lignin

— 23 —

Estimation of wood properties of radial strips

The same questions discussed for pulp yield apply to the work conducted on radial strips:

• Multi-site, multi-species calibrations; and
• Application of existing calibrations to new sites

Other important questions for calibration purposes: – how many spectra are required to represent a core? – at what spatial resolution do we have to collect spectra?

— 24 —

10mm MFA calibration (729 spectra)

5 15 25 35 45 5 15 25 35 45 Fitted MFA (degrees) Measured MFA (degrees) Factors = 8 R

2 = 0.90

SEC = 2.33 SECV = 2.38 RPD = 3.11

Source: JONES, P.D.; SCHIMLECK, L.R.; PETER, G.F.; DANIELS, R.F.; CLARK, A. 2005a: Nondestructive estimation

• f Pinus taeda L. wood properties for samples from a wide range of sites in Georgia. Can. J. For. Res. 35: 85-92.

— 25 —

10mm stiffness calibration (729 spectra)

5 10 15 20 25 5 10 15 20 25 Fitted Stiffness (GPa) Measured Stiffness (GPa) Factors = 5 R2 = 0.91 SEC = 1.48 SECV = 1.53 RPD = 3.19

Source: JONES, P.D.; SCHIMLECK, L.R.; PETER, G.F.; DANIELS, R.F.; CLARK, A. 2005a: Nondestructive estimation

• f Pinus taeda L. wood properties for samples from a wide range of sites in Georgia. Can. J. For. Res. 35: 85-92.

— 26 —

10mm MFA – prediction (225 spectra)

5 10 15 20 25 30 35 40 45 50 10 20 30 40 50 Predicted MFA (degrees) Measured MFA (degrees) Factors=8 R2= 0.84 SEP= 3.12 RPD= 2.34

Source: JONES, P.D.; SCHIMLECK, L.R.; PETER, G.F.; DANIELS, R.F.; CLARK, A. 2005a: Nondestructive estimation

• f Pinus taeda L. wood properties for samples from a wide range of sites in Georgia. Can. J. For. Res. 35: 85-92.

— 27 —

10mm MFA prediction – single core

5 10 15 20 25 30 35 10 20 30 40 50 60 70 80 90 100

Distance from pith (mm) MFA (degrees)

Source: JONES, P.D.; SCHIMLECK, L.R.; PETER, G.F.; DANIELS, R.F.; CLARK, A. 2005a: Nondestructive estimation

• f Pinus taeda L. wood properties for samples from a wide range of sites in Georgia. Can. J. For. Res. 35: 85-92.

— 28 —

FOSS spectrometer – 5mm MFA calibration

10 20 30 40 50 10 20 30 40 50 NIR-measured MFA Lab-measured MFA 7 Factors R2 = 0.94 SEC = 1.72 deg. SECV = 2.40 deg. RPD = 2.98

JONES, P.D.; SCHIMLECK, L.R.; SO, C.-L.; CLARK, A.; DANIELS, R.F. 2007: High resolution scanning

• f radial strips cut from increment cores by near infrared spectroscopy. IAWA Journal (accepted).

— 29 —

MFA prediction – 2mm

5 10 15 20 25 30 35 10 20 30 40 50 60 70 80 90 Distance from pith (mm) MFA (deg.)

— 30 —

The problem of increasing resolution

• As resolution increases variation in wood property

data increases

• Accurate positioning of sample becomes

increasingly important

• It becomes increasingly difficult to manage spectra

90 strips at 2 mm = 4156 spectra 9 strips = 80 spectra at 10 mm resolution 160 spectra at 5 mm, 400 spectra at 2 mm

• Existing spectrometers not suited to scanning radial

strips at high spatial resolution

— 31 —

Options for increasing spatial resolution

Collect NIR spectra from the tangential surface

• Transmission NIR (thin sections) eg. Yeh et al. (2004 and 2005)
• Reflectance NIR (thick sections) eg. Schimleck et al. (2007)

10 20 30 40 50 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 Sample number MFA (degrees)

Source: SCHIMLECK, L.R.; SUSSENBACH, E.; LEAF, G.; JONES, P.D.; HUANG, C.L. 2007: Microfibril angle prediction of Pinus taeda L. wood samples based on tangential face NIR spectra. IAWA Journal 28: 1-12.

— 32 —

How many spectra are required per core?

• For calibration a single spectrum per core is adequate

provided that sections from other cores representing juvenile, mature and transition wood are included

• 3-5 spectra per core will improve calibration statistics slightly
• For calibration 7 cores per plantation are recommended

0.0 0.2 0.4 0.6 0.8 1.0 1 2 3 4 5 6 7 8 FULL Number of NIR spectra per core Rp2 0.0 0.2 0.4 0.6 0.8 1.0 1 2 3 4 5 6 7 8 9 FULL Number of cores per plantation Rp2

Source: SCHIMLECK, L.R.; TYSON, J.A.; JONES, P.D.; PETER, G.F.; DANIELS, R.F.; CLARK, A. 2007b: Pinus taeda L. wood property calibrations based on variable numbers of NIR spectra per core and cores per plantation. Journal of Near Infrared Spectroscopy (submitted).

— 33 —

Field based NIR spectroscopy

NIR spectroscopy is increasingly being used in the field providing data immediately, eg. fitted to harvesters to estimate forage quality

• For the estimation of standing tree quality of P. taeda

Jones et al. (2007) reported several problems, including: – Design of a suitable probe – Performance of fiber-optic probe systems – Resin bleeding into core hole (OK for eucalypts)

• Acuna and Murphy (2006) report promising results for

estimating Douglas fir density using chain saw chips

— 34 —

Estimation of genetic parameters based on NIR data

• Raymond et al. (2001) estimated genetic parameters

and genotype-by-environment interactions for pulp yield and pulpwood productivity in E. globulus

• Raymond & Schimleck (2002) estimated genetic

parameters for cellulose content in E. globulus

• Schimleck et al. (2004) estimated genetic parameters

and genetic gains for cellulose content in E. nitens

— 35 —

Estimation of genetic parameters based on NIR data

3 sites in Northern Tasmania 538 E. nitens (core samples) Elevation range 100–300 m, rainfall range 1060–1200 mm Planted 1984, sampled at age 13 Dial (168) Gog (182) Kamona (188)

— 36 —

Estimation of genetic parameters based on NIR data

• Selections made without knowledge of cellulose content
• Two methods:
• 1. Apply an existing pulp yield calibration and ID samples

that cover range (20, 40, 60 samples)

• 2. Select samples based on spectral features (WinISI II)
• Calibrations developed for each site and used to predict

cellulose content of all samples at a site

• Genetic gains determined and compared (genetic gain =

increase in cellulose after selecting the top 5 % of trees)

— 37 —

Estimation of genetic parameters based on NIR data

91% 91% 90% 80% Off-site model 3 (Kamona calibration) 83% 70% 73% 64% Off-site model 2 (Gog calibration) 82% 87% 86% 61% Off-site model 1 (Dial calibration) 89% 88% 83% 73% Local site model WinISI 1 60/site 40/site 20/site Selection of samples for calibration model Type of calibration model 99% 89% 91% 90% Off-site model 3 (Kamona calibration) 100% 79% 97% 83% Off-site model 2 (Gog calibration) 100% 81% 95% 84% Off-site model 1 (Dial calibration) 100% 87% 94% 93% Local site model WinISI 1 60/site 40/site 20/site Selection of samples for calibration model Type of calibration model 1 The number of samples used by WinISI was 37 at Dial, 59 at Gog and 45 at Kamona.

Genetic gains – forward selection Genetic gains – backward selection

Source: SCHIMLECK, L.R.; KUBE, P.D.; RAYMOND, C.A. 2004: Genetic improvement of kraft pulp yield in Eucalyptus nitens using cellulose content determined by near infrared spectroscopy. Canadian Journal of Forest Research 34: 2363-2370.

— 38 —

Conclusions

• NIR spectroscopy provides a rapid method for the

assessment of multiple wood properties and it could play a very important role in tree breeding programs

• Several questions remain, 2 of the most important are:
• 1. The applicability of calibrations to new sites; and
• 2. NIR instruments that can be used in the field
• Using NIR for appropriate applications is important
• What is the best way to provide tree breeders NIR data?

— 39 —

Acknowledgements

• Conference organisers for inviting me to the conference
• The scientists who worked with me on many of the

projects described in this presentation