Passive field reflectance measurements

Post on 27-Jan-2023

0 views 0 download

transcript

This content has been downloaded from IOPscience. Please scroll down to see the full text.

Download details:

IP Address: 193.140.28.22

This content was downloaded on 05/11/2014 at 02:25

Please note that terms and conditions apply.

Passive field reflectance measurements

View the table of contents for this issue, or go to the journal homepage for more

2008 J. Opt. A: Pure Appl. Opt. 10 104020

(http://iopscience.iop.org/1464-4258/10/10/104020)

Home Search Collections Journals About Contact us My IOPscience

IOP PUBLISHING JOURNAL OF OPTICS A: PURE AND APPLIED OPTICS

J. Opt. A: Pure Appl. Opt. 10 (2008) 104020 (8pp) doi:10.1088/1464-4258/10/10/104020

Passive field reflectance measurementsChristian Weber1, Daniel C Schinca1,2, Jorge O Tocho1,3 andFabian Videla1,2

1 CIOp (CONICET-CIC), CC 124, 1900 La Plata, Argentina2 Facultad de Ingenierıa, Universidad Nacional de La Plata, Argentina3 Facultad de Ciencias Exactas, Universidad Nacional de La Plata, Argentina

E-mail: cweber@ciop.unlp.edu.ar

Received 14 March 2008, accepted for publication 12 August 2008Published 2 September 2008Online at stacks.iop.org/JOptA/10/104020

AbstractThe results of reflectance measurements performed with a three-band passive radiometer withindependent channels for solar irradiance reference are presented. Comparative operationbetween the traditional method that uses downward-looking field and reference white panelmeasurements and the new approach involving duplicated downward- and upward-lookingspectral channels (each latter one with its own diffuser) is analyzed. The results indicate that thelatter method performs in very good agreement with the standard method and is more suitablefor passive sensors under rapidly changing atmospheric conditions (such as clouds, dust, mist,smog and other scatterers), since a more reliable synchronous recording of reference andincident light is achieved. Besides, having separate channels for the reference and the signalallows a better balancing of gains in the amplifiers for each spectral channel. We show theresults obtained in the determination of the normalized difference vegetation index (NDVI)corresponding to the period 2004–2007 field experiments concerning weed detection in soybeanstubbles and fertilizer level assessment in wheat. The method may be used to refinesensor-based nitrogen fertilizer rate recommendations and to determine suitable zones forherbicide applications.

Keywords: remote sensing, synchronous measurement, nitrogen fertilization, weeds detection

(Some figures in this article are in colour only in the electronic version)

1. Introduction

Reflectance-based measurement techniques are valuable toolsto study culture stress [1]. Most of these effects rely onfrequency-dependent energy interactions, which enable theiranalysis through characteristic wavelength-based methods [2].Spectral reflectance of cultures in specific wavelength regionshas been correlated with plant parameters such as leaf area,soil cover percentage, dry-matter accumulation, nitrogen andhydric content [3].

The direct spectral reflectance signal may comprise factors(i.e. soil reflectance) that can obscure the relation betweenthe measurement and the effect of the issue under study. Soit is necessary to derive some parameter that may take thesame value for the same culture cover degree, independentof the optical properties of the different kinds of soil. Thisproblem has been traditionally addressed by defining the so-called vegetation index (VI) which, ideally, should be sensitive

to cover degree only. One of the most used VIs is the so-callednormalized difference vegetation index, NDVI [4].

Direct passive spectral reflectance sensor signals mustbe normalized to follow incident light intensity variations,which may otherwise be misinterpreted as changes in thestudied parameters (i.e. cover degree, nitrogen defect). This iscommonly accounted for by taking several spectral reflectancesignals from a flat white panel placed over the soil. Thiswhite panel is a Spectralon-like Lambertian diffuse reflectivepanel that is traditionally used as a primary reflectancestandard [5, 6] in passive reflectance measurements. However,when atmospheric light conditions are not uniform or arerapidly changing (i.e. uneven clouding, rapidly movingclouds), the measurements become cumbersome and timeconsuming, which makes the method almost impractical inthese situations. For such cases, a more user-friendly methodshould be devised.

In this work we describe a comparative study betweenpassive spectral reflectance sensors normalized through a

1464-4258/08/104020+08$30.00 © 2008 IOP Publishing Ltd Printed in the UK1

J. Opt. A: Pure Appl. Opt. 10 (2008) 104020 C Weber et al

Table 1. Optical characteristics of the measurement channels.

Channel WavelengthBandwidth(FWHM) Transmittance

Red 656 nm ± 2 nm 10 nm ± 2 nm >50%Infrared 1 830 nm ± 2 nm 10 nm ± 2 nm >50%Infrared 2 980 nm ± 2 nm 10 nm ± 2 nm >50%Red (withdiffuser)

656 nm ± 2 nm 10 nm ± 2 nm >50%

Infrared(with diffuser)

830 nm ± 2 nm 10 nm ± 2 nm >50%

Infrared(with diffuser)

980 nm ± 2 nm 10 nm ± 2 nm >50%

reference panel and through synchronous detection of zenith-looking twin channels with Teflon diffusers, for changingcloud conditions. This methodology is suitable for similarapproaches in other fields of research where open columnsof samples have to be measured (i.e. oceanography, andatmospheric science [5–7]). To the best of our knowledge,this synchronous method for in-field culture reflectancemeasurements is described for the first time. Fieldmeasurements for weed detection and wheat nitrogenic fertilityare also presented. This point has a relevant importance inmodern agricultural methods aiming for higher profitabilityand lower environmental impact relying on an optimalefficiency in nitrogen and herbicide use, since insufficient orexcess quantities have a direct impact on crop production [8, 9].In Argentina, nitrogen-based fertilizers are usually applied inthe Tillering growth stage (GZ 20–30), [10] to obtain a largerstand of plants, which is related to a larger number of headsand kernels per unit area [11]. Late applications usually have apositive impact in commercially important parameters such asquality [12, 13]. Herbicides are applied in full area covering,but weeds present an inhomogeneous distribution in the field.

2. Materials and methods

Radiometric measurements were done using a multibandsensor developed at CIOp (Centro de Investigaciones Opticasde La Plata, CONICET-CIC) as part of the research carriedout in spectroscopy applications. This sensor results from aevolution of previous developments which had no referencechannels for simultaneous measurements. In these cases,intensity normalization was carried out using ground-basedreference panels. The characteristics of the new sensor areshown in table 1.

The radiometer is built in modules, each one comprisingan interferential optical filter and a Hamamatsu S1226-44BQsilicon photodiode (figure 1). The soil-looking channels havean aperture limiting their field of view (FOV) to 30◦, whichdetermines a sensing area of 1 m2 at a sensor altitude of2 m. All photodiodes were used in current mode, renderinglinearity and high signal-to-noise ratio. Output voltage wasrecorded in a Hewlett–Packard 34970 A data logger at a rateof 1 signal/channel/30 s. All instruments were powered by aDC–AC converter and a small 12 V battery. A laser pointerwas mounted in the central part of the radiometer to mark thecenter of the sensing area.

FOV

Filter

Photodiode

O’ring

Teflon

Filter

Photodiode

O’ring

Figure 1. Sensor’s optical modules.

2.1. Comparative measurements between reference panel anddiffuser modules

The measurements were taken under variable cloud conditionsbetween 10 am and 2 pm at the city of La Plata, BuenosAires, Argentina (Latitude 34◦ 53′ S, Longitude 58◦ 00′ W),during winter. For this work, three days with different skycloud coverage were selected (clear sky, high cloud cover,and mild cloud cover). Zenith-looking and reference panelmeasurements were taken simultaneously and recorded aftersubtracting the dark current.

2.2. Field measurements for determining nitrogenic fertilityvariations in wheat

For the field measurements, the sensor was normalized usingthe diffuser channels only. The reflectance values for eachchannel were calculated by subtracting the dark currentand dividing by the corresponding upward-looking diffuser-channel signal, as follows:

Rλ = Sλ − Zλ

Riλ − Zλ

× kλ,

where Sλ is the vegetation signal centered in band λ, Riλ is

the reference signal for the same λ, Zλ is the dark signal forboth vegetation and diffuser channels, and kλ is the absolutereflectance for λ.

As mentioned above, a way to minimize unwantedeffects such as different soil optical properties, illuminationgeometries and meteorological factors is to define the NDVIinvolving red (650–670 nm) spectral bands related withchlorophyll absorption and near infrared (NIR, 800–1000 nm)for green biomass detection:

NDVI = NIR − RED

NIR + RED.

One of the characteristics of the NDVI is that it allowsremoving the dependence from absolute calibrations in passiveoptical systems, although this is mainly related to satelliteimage applications, since a determined scene holds the samelevel of illumination. However, when different field scenes areto be compared, such as we are concerned with, it is necessaryto normalize each channel to avoid fluctuations in illuminationlevels being misinterpreted as changes in the parameters understudy (nitrogen or vegetation coverage). This is because the

2

J. Opt. A: Pure Appl. Opt. 10 (2008) 104020 C Weber et al

Z

Ω

X(E)

ϕ

Y(N)

θZ

Sol

H

Sensor

Figure 2. Coordinate system. The azimuthal angle is measured fromNorth to East.

influence of the atmospheric factors on the received signalis not proportional in all channels, so it would modify thecalculated NDVI value.

As most natural surfaces, vegetation reflects light in ananisotropic way. When the observation geometry changes,this behavior depends on the canopy structure and the solarzenithal angle; for the sun at nadir, isotropic scattering isfavored, while for the sun near the horizon anisotropic behavioris enhanced. During radiometric measurements, the sensor washeld at approximately one meter over the canopies, with a smallswinging movement about the vertical direction, as can be seenin figure 2. This figure shows the system of measurementand coordinates used. This system is capable of measuringreflectance from the culture placed in the XY plane.

Field measurements were carried out in experimentalparcels of wheat cultures (Triticum aestivum L.), onArgiudol Typic soil [14] belonging to Experimental StationJ J Hirschhorn (Facultad de Ciencias Agrarias y Forestales dela Universidad Nacional de La Plata, city of La Plata, province

Table 2. Soil analysis results at 0–20 cm.

pH (potentiometricmeasurement in saturated soil)

5.4

Organic matter (% organicC multiplied by f 1.724)

2.98

N total (Kjeldahl method) 0.18Rel C/N 9.6P assimilable (ppm)(Bray–Kurtz I method)

9.5

of Buenos Aires, Latitude 34◦ 55′ S, Longitude 57◦ 50′ W).Different nitrogen fertilization levels (urea, 0, 80 and 160 kgN) within each parcel were prepared. Chemical soil analysisdetermined that phosphor was not a limiting factor for culturedevelopment (table 2).

To find the ability of this method for determining changesin nitrogen fertility, a multifactor ANOVA was done. Thesensor performance was also compared with a commercialchlorophyllometer [15, 16]. This study was made for aperiod of three years with emphasis in growth stage Z 60(pre-anthesis) since the dynamics of nutrients shows a patternin which the accumulation rate (particularly of nitrogen) bywheat increases as the foliar area is augmented (the leaf areais maximum at this stage), with the total requirements ofnutrients concentrated in this period (ca 80% N absorbed) [17].The variations in nitrogen fertilization levels were efficientlydetected. This was most noticeable in the basic fertilizationlevel (0–80 kg N), which is a typical value in Argentina.

2.3. Identification of weeds in soybean stubble

Zones with different weeding degree were selected and thesensor was placed so as to measure a 1 m2 area. Photographs ofeach measurement zone were taken so as to visually comparethe reflectance values with the vegetal cover degree. Theimportance of crop damage associated with weed harrowinghas often been demonstrated [18, 19].

Local time (h)

10 12 14 16 18

Re

flect

an

ce

0.00

0.01

0.02

0.03

0.04

0.35

0.40

0.45

0.50

Blue

Green

NIR

Red

NIR

Local time (h)

10 12 14 16 18

Sig

nal (

V)

0.00

0.05

0.10

0.40

0.60

0.80

Red

Green

Blue

CloudsClouds

Figure 3. Disturbances observed in the signal caused by the passage of clouds. In the right graph the vertical axis corresponds to the rawsignal without normalization of the photodetectors. The vertical axis in the left graph shows the ratio between the raw signal and the averagereference signal obtained from a white panel taken at the beginning, middle and end of the measurement time.

3

J. Opt. A: Pure Appl. Opt. 10 (2008) 104020 C Weber et al

A2

B2B

Local time (h)

Sig

nal [

V]

-3,00E+00

-2,50E+00

-2,00E+00

-1,50E+00

-1,00E+00

-5,00E-01

0,00E+0010

:00:

00

10:3

0:00

11:0

0:00

11:3

0:00

12:0

0:00

12:3

0:00

13:0

0:00

13:3

0:00

14:0

0:00

14:3

0:00

980diffuser

980 reference panel980 nm reference

panel

980 nm diffuser

Sig

nal [

V]

Local time (h)

-2,50E+00

-2,00E+00

-1,50E+00

-1,00E+00

-5,00E-01

0,00E+00

5,00E-01

1,00E+0010

:00:0

0

10:36

:16

11:12

:32

11:48

:48

12:25

:04

13:01

:20

13:37

:36

980nm reference panel

980nm difuser

980 nm reference

panel

980 nm diffuser

C2

C1

-2,50E+00

-2,00E+00

-1,50E+00

-1,00E+00

-5,00E-01

0,00E+0010

:00:0

0

10:25:4

0

10:51:2

0

11:1

7:00

11:42:4

0

12:08:2

0

12:34:0

0

12:59:4

0

13:25:2

0

13:51:0

0

Local time (h)

Sig

nal [

V]

A1

1

Figure 4. Reference panel and diffuser reflectance measurements under different sky conditions: A1 clear sky, B1 high cloud cover andC1 mild cloud cover.

3. Results

3.1. Comparison between reference panel and diffusermeasurements

Previous work carried out by our group has shown theusefulness of the information obtained with passive opticalsensors developed at CIOp for agronomic applications [20].However, field measurements may typically take many hours,

during which the sky conditions may change and alter thevalues of the recorded signals. Moreover, these variationsmay be a function of wavelength and can be different forthe different measurement bands. Figure 3 shows the signalchange in a four-channel sensor due to the passage of cloudsduring the measurement time. In the right graph the verticalaxis corresponds to the raw signal without normalization of thephotodetectors. Notice the spikes produced by passing clouds

4

J. Opt. A: Pure Appl. Opt. 10 (2008) 104020 C Weber et al

Infrared 1 in cloudy condition

y = 2,985x + 0,055R2 = 0,995

-3,00

-2,50

-2,00

-1,50

-1,00

-0,50

0,00-1,00-0,80-0,60-0,40-0,200,00

Reference panel signal [V]

Dif

fuse

rs s

ign

al [V

]

Red channel in cloudy condition

y = 1,470x + 0,444R2 = 0,995

-7,00

-6,00

-5,00

-4,00

-3,00

-2,00

-1,00

0,00-5,00-4,00-3,00-2,00 -1,000,00

Reference panel signal [V]

dif

fusers

sig

nal [V

]

InfraRed2 channel in cloudy condition

y = 0,494x + 0,091R2 = 0,994

-1,20E+00

-1,00E+00

-8,00E-01

-6,00E-01

-4,00E-01

-2,00E-01

0,00E+00-2,50E+00-2,00E+00-1,50E+00-1,00E+00-5,00E-010,00E+00

Reference panel signal [V]

Dif

fus

ers

sig

nal

[V

]

Figure 5. Linear regressions for diffuser and reference panel signals.

during the measurement period. The vertical axis in the leftgraph shows the ratio between the raw signal and the averagereference signal obtained from a white panel (reflectance)taken at the beginning, middle and end of the measurementtime. It can be seen that this normalization procedure does notremove the effect of clouds.

On the other hand, figures 4(A1)–(C1) show a comparisonbetween a set of four-hour field measurements of reflectancefor the reference panel (upper trace) and diffuser (lower trace)for the 980 nm channel. Both sets closely follow eachother during different cloudy conditions, as can be seen infigures 4(A2)–(C2). It was found that the other channel presentsa similar pattern. When data for each channel are plottedone versus the other, the relation can be fitted with a linearregression with excellent match, as can be assessed by itsR2 coefficient (figures 5(a)–(c)). These results show thatthe diffuser-based normalization measurements closely followthose obtained with the standard reference panel technique,as can be seen by the very small dispersion of the points infigure 5 with respect to the straight line of the linear regression.This feature is regularly used as a validation procedure whenintercomparison with a standard technique is performed.

3.2. Detection of nitrogen in wheat using the NDVI

Figure 6 shows the ability of the sensor for detecting variationsin the degree of fertilization in a wheat culture. In figure 6(a)

there can be seen the increase in the NDVI as the nitrogenfertilization increases. The variations in nitrogen fertilizationlevels were efficiently detected. This was most noticeable inthe basic fertilization level (0–80 kg N), which is a typicalvalue in Argentina. The points in the figure correspond tothe mean value for the three years for each treatment in thisphenologic stage. This behavior is less evident in irrigatedparcels (lower slope curve), since the NDVI tends to saturatefor high biomass values. Figure 6(b) shows a plot of NDVIvalues obtained with our instrument versus those predictedusing a commercial chlorophyllometer. It can be seen thatthey can be reasonably fit with a linear regression. Figure 6(c)shows the evolution of the NDVI (average values) during thedevelopment period (winter–spring) for different fertilizingamounts. Use of the NDVI is based on chlorophyll absorptionin the red band and is related straightforwardly to the nitrogencontent. As can be seen in figure 6(a), the NDVI shows apositive response to fertilization. However, this relationshipmay not be unique since it is influenced by the amount ofbiomass (irrigated culture). Thus, the interpretation of NDVIvalues must take into account both irrigation and biomass.

When point-to-point NDVI measurements were carriedout over relatively large parcels, NDVI values may be plottedversus X–Y parcel positions. The upper panel in figure 7 showsthis kind of graph, where the X–Y dimensions are in metersand the NDVI values are represented on the Z axis. The lower

5

J. Opt. A: Pure Appl. Opt. 10 (2008) 104020 C Weber et al

Plot of NDVI

predicted

Obs

erve

d

0,4 0,45 0,5 0,55 0,6 0,65 0,70,4

0,45

0,5

0,55

0,6

0,65

0,7

b

Correlation Coefficient = 0,94 R-squared = 87,57 percentage

-0,2-0,1

00,10,20,30,40,50,60,70,8

160 N

80

0

ND

VI

c

Period of measurements

Kg N

0 080 160

Noirrigated

Irrigated

a

0,52

0,56

0,6

0,64

ND

VI

0,48

0,68

14_0

9

20_09

6_10_0

6

23_1

0_06

1_11_06

9_11_0

6

22_11

18_08

Figure 6. NDVI behavior and nitrogen fertilization levels. (a) ANOVA analysis of different nitrogen treatments. (b) Plot of observed versuspredicted NDVI values. (c) NDVI behavior during the whole growing period (average valued).

ND

VI

Val

ue

Y and X Distance in meters

Y

X

Figure 7. Correspondence between NDVI value and crop placement in soil.

panel shows an actual photograph of the culture. It can be seenthat the NDVI plot closely follows the real crop distributionwith acceptable spatial resolution. The apparent disagreementin the narrow valley at the front of the plot with respect to theforefront of the photograph is due to the fact that no pointswere measured there during the measurement protocol.

3.3. Detection of weeds in soybean stubble

Weed detection is important in countries like Argentinawhere direct seeding (no soil preparation) is carried out and

large amounts of herbicide are used with full area covering.However, the weeds show an inhomogeneous distribution inthe parcel. In this case, the NDVI was used since it is avegetation index with a good response to soil cover when thevegetal coverage is poor. Figure 8 shows the NDVI valuesobtained in soybean stubble for zones with increasing degreeof weed coverage, taken as area covered by weed divided bytotal sensed area. The large discontinuities in the NDVI signalcorrespond to increasing degree of weed coverage (increase inquantity of weeds in the sensed area).

6

J. Opt. A: Pure Appl. Opt. 10 (2008) 104020 C Weber et al

N D V I

-1 ,00E -01

0 ,00E + 00

1,00E -01

2 ,00E -01

3 ,00E -01

4 ,00E -01

5 ,00E -01

6 ,00E -01

7 ,00E -01

N D V I

1 2 3 4 5

Figure 8. NDVI values for different degrees of weed coverage. Signals 1, 2, 3 and 4 correspond to a monotonic increase in weed coverage.Signal 5 corresponds to zero weed coverage.

4. Conclusions

The excellent agreement between the two methods formeasuring incident reflectance suggests that it is possible touse the technique of synchronous measurement of vegetaland diffused sky light. This method shows comparativeadvantages over traditional reference-panel-based methodsused in field measurements. It has been shown that it is capableof responding to rapid changes in illuminating conditions,thus making the measurements independent of cloudy skyconditions. This was verified through field measurementsunder different sky conditions (very cloudy, partly cloudy andclear).

The device was also shown to be an efficient detectorof variations in nitrogen fertilization in wheat cultures. Itcan be used to determine spatial variations of this nutrientwhen applying variable fertilization protocols. Although thiswas noticeable in the non-irrigated parcels, it did not show alarge span for irrigated treatment (high biomass) due to NDVIsaturation.

Additionally, the results show high sensitivity fordetecting small coverage percentage for weeds in stubble, dueto the NDVI’s ability in detecting variations in the soil coveragedegree at the beginning of the vegetable soil covering. Thismethodology could lead to spatial analysis of weed coverageand help to decide on herbicide application sites.

Acknowledgments

This work is supported by: CONICET (PIP 5997), ANPCyT(PICT 26090) and UNLP (11X456) research projects.

References

[1] Carter G A 1993 Responses of leaf spectral reflectance to plantstress Am. J. Bot. 80 239–43

[2] Strachan I B, Pattey E and Bosivert J B 2002 Impact of nitrogenand environmental conditions on corn as detected byhyperspectral reflectance Remote Sens. Environ. 80 213–24

[3] Gitelson A and Merzlyak M 1996 J. Plant Physiol. 148494–500

[4] Rouse J W, Haas R H, Schell J A, Deering D W and Harlan J C1974 NASA/GSFC, type III Final Report (Greenbelt, MD,USA) pp 1–371

[5] Bruegge C, Chrien N and Haner D 1998 LABSPHEREReflections Newsletters 40 231–9

[6] Bazalgette Courre‘Ges-Lacoste G, Groote Schaarsberg J,Sprik R and Delwart S 2003 Modeling of Spectralondiffusers for radiometric calibration in remote sensingOpt. Eng. 42 3600–7

[7] Duntley S Q, Uhl R J, Austin R W, Boileau A R and Tyler J E1955 An underwater photometer J. Opt. Soc. Am. 45 904

[8] Ma B L, Morrison M J and Dwyer L M 1996 Canopy lightreflectance and field greenness to assess nitrogenfertilization and yield of maize Agron. J. 88 915–20

[9] Hatfield J L, Gitelson A A, Schepers J S and Walthall C L 2008Application of spectral remote sensing for agronomicdecisions Agron. J. 100 S-117–31

[10] Zadoks J C, Chang T T and Konzak Y C F 1974 A decimalcode for the growth stages of cereals Weed Res. 14 415–21

[11] Klepper B, Rickman R W and Peterson C M 1982 Quantitativedevelopment in small cereal grains Agron. J. 74 789–92

[12] Wang Z J, Wang J H, Liu L Y, Huang W J, Zhao C J andWang C Z 2004 Prediction of grain protein content in winterwheat (triticum aestivum L) using plant pigment ratio (PPR)Field Crop. Res. 90 311–21

[13] Le bail M, Jeuffroy M H, Bouchard C and Barbottim A 2005 Isit possible to forecast the grain quality and yield of differentvarieties of winter wheat from Minolta SPAD metermeasurements? Eur. J. Agron. 23 379–91

7

J. Opt. A: Pure Appl. Opt. 10 (2008) 104020 C Weber et al

[14] NRCS 1999 Soil Taxonomy US Department of Agriculture(Washington, DC: US Government Printing Office)

[15] Arregui L M, Lasa B, Lafarga A, Iraneta I, Baroja E andQuemada M 2006 Evaluation of chlorophyll meters as toolsfor N fertilization in winter wheat Ander humidMediterranean conditions Eur. J. Agron. 28 140–8

[16] Peltonen J, Virtanen A and Haggren E 1995 Using achlorophyll meter to optimize nitrogen fertilizer applicationfor intensively-managed small grain cereals J. Agron. CropSci. 147 309–18

[17] Fischer R A 1993 Irrigated spring wheat and timing andamount of nitrogen fertilizer: II. Physiology of grain yieldresponse Field Crop. Res. 33 57–80

[18] Rasmussen J, Nørremark M and Bibby B M 2007 Assessmentof leaf cover and crop soil cover in weed harrowing researchusing digital images Weed Res. 47 299–310

[19] Jensen R K, Rasmussen J and Melander B 2004 Selectivity ofweed harrowing in lupin Weed Res. 44 245–53

[20] Weber C, Videla F, Schinca D C and Tocho J O 2004 Passivesensor for wheat reflectance measurements Proc. SPIE Int.Soc. Opt. Eng. 5622 244–9

8