IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 5, NO.

6, DECEMBER 2012 1793

Radiometry of Proximal Active Optical Sensors

(AOS) for Agricultural Sensing
Kyle H. Holland, Member, IEEE, David W. Lamb, Senior Member, IEEE, and James S. Schepers

Abstract—Over the last decade, portable, active optical sensors The advantages of proximal, active, optical sensing for rapid
(AOS) have become tools in production agriculture both for map- assessment of pasture/crop canopies have been outlined in
ping crops and soils and also for applying agrochemicals. These [1] and [2]. AOS have increased in popularity for quantifying
sensors are often referred to as “proximal” active optical sensors,
in recognition of their deployment at sensor-target distances in the nutritional requirements of crops [3]–[5], for measuring
the order of meters. Unfortunately most users have little under- biomass in pastures [6], [7], [2] and as a tool for applying
standing of the underlying physics (optics) or construction of these agrochemicals in real-time [8], [9].
sensors. This paper sets out to document the fundamental electro- The development of spectral indices that reduce multiple-
optical principles by which these devices operate and to document
the mathematical rules governing the use of data produced by these
wavelength measurements to a single numerical measure has
devices. In particular, emphasis is placed on the inverse-square law underpinned the development of AOS technology. For example,
of optics and how it affects AOS measurements. The basis for uti- spectral indices that use combinations of red (Red) and near in-
lizing sensor measurements in more complex mathematical func- frared (NIR) reflectance, are potential candidates for estimating
tions is also presented, including manipulation of the individual the quantity and quality of green vegetation. Indices such as the
wavelength response data to derive information regarding the dis-
tance variation between the sensor and the target. Normalized Difference Vegetation Index (NDVI) and Simple
Ratio Index (SRI) have well-known relationships with pigment
Index Terms—Agriculture, optics, optoelectronic devices, ra-
diometry, remote sensing. content, leaf water stress and green biomass [10], [11]. Other
spectral indices have been proposed to enhance sensitivity to
the characteristic of interest and minimize the confounding ef-
I. INTRODUCTION fects. The Soil Adjusted Vegetation Index (SAVI) was designed

A CTIVE optical sensors (AOS) are specialized instru- to minimize soil induced variations [12]. The Non-Linear Veg-
ments that irradiate a target with radiation and measure etation Index (NLI), the modified Non-Linear Vegetation Index
that which is scattered back to the sensor’s integral photo-de- (MNLI) and Modified Simple Ratio (MSR) were designed to
tector. The key advantage of AOS is that measurements can be account for non-linear relationships with leaf area index (LAI)
performed under any ambient illumination condition, including [13], [14]. Both the SAVI and MNLI include a “transformation
at night. factor” , designed to render them insensitive to soil reflectance
On the other hand, “passive” sensors require sufficient or surface factors unrelated to the actual plant canopy. The value
ambient illumination to detect the target. At-sensor irradiance of in both indices is selected to be 0.5 for intermediate vege-
values vary with solar zenith and azimuth angles, as well as tation densities [12], [13], [2].
sensor position relative to the target (altitude, view angle). In With respect to plant canopy assessment, AOS technology
terms of airborne or satellite-based passive sensors, consider- has a long history dating back over 70 years. One of the ear-
ation must also be given to ensuring the field-of-view to the liest active electro-optical sensors was designed to be spectrally
distant target is not obscured by clouds and that cloud shadows sensitive to a plant’s carotenoid peak located at 550 nm for the
are not present on the target. Moreover atmospheric effects intended purpose of plant detection and selective thinning [15].
must also be accounted for if at-canopy optical characteristics In 1948, a sensor was developed for the selective application of
are sought. herbicides and fertilizers using color filters to detect plant char-
acteristics [8]. Palmer and Owen [16] reported on a sugar beet
Manuscript received October 25, 2011; revised January 12, 2012; accepted “singling” system to automatically reduce plant populations.
April 10, 2012. Date of publication June 11, 2012; date of current version De- This instrument incorporated four optically modulated sensors
cember 28, 2012. This work was supported in part by the Cooperative Research
to create a 2-dimensional “kill map” of plants to be eliminated.
Centre for Spatial Information (CRCSI) under the Australian Governments Co-
operative Research Centres Program. When a plant to be eliminated was detected, the system would
K. H. Holland is with Holland Scientific, Inc., Lincoln, NE 68516 USA (cor- respond by spraying the plant with an herbicide. It should be
responding author, e-mail: [email protected]).
D. W. Lamb is with the Precision Agriculture Research Group, University of
noted that this was also the first sensing system to incorporate
New England, and with the Cooperative Research Centre for Spatial Informa- a digital computer to orchestrate both plant sensing and agro-
tion, NSW Australia (e-mail: [email protected]). chemical control.
J. S. Schepers is with the United States Department of Agriculture, Agri-
cultural Research Service (retired, collaborator) and Agronomy and Horticul- From 1975 to 2002, fully solid-state sensors were developed
ture Department, University of Nebraska (Emeriti), Lincoln, NE, USA. (e-mail: that utilized light emitting diodes (LEDs) to illuminate plant
[email protected]). canopies. The first AOS instrument to use LEDs as an illumi-
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. nation source was designed to measure leaf reflectance char-
Digital Object Identifier 10.1109/JSTARS.2012.2198049 acteristics [17]. Stafford et al. [18] developed a portable hand-

1939-1404/$31.00 © 2012 IEEE

1794 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 5, NO. 6, DECEMBER 2012

held sensor that contained two NIR monochromatic LEDs emit-

ting 940 and 1150 nm to measure turf moisture content. Sub-
sequently, Beck and Vyse [19] developed a weed sensor with
LED sources emitting at 670 and 750 nm (WeedSeeker, Patchen,
Ukiah, California USA). A major advance was source moni-
toring methodology to stabilize the variation in the instrument’s
LED radiance [20]. Künnemeyer et al. [7] tested a sensor that
used eight LEDs and determined that the combination of 598
and 896 nm (approximately equivalent to yellow and NIR wave-
lengths) resulted in the lowest error when used to estimate the
herbage mass of a ryegrass pasture (Lolium perenne). Beck and
Kinter [21] developed a sensor to discriminate weeds from bare
soil. The sensor utilized three or more LED light sources to
evaluate target slope changes due to the presence of weeds in
the sensor’s field-of-view. Similarly, Paap et al. [22] evaluated
three laser diodes (635, 670 and 785 nm) in a line-scan imaging
sensor to discriminate weeds against a bare soil background.
The multi-wavelength approach allowed them to calculate the Fig. 1. Example of the physical sensor enclosure of an AOS (Crop Circle ACS-
slopes between the three wavelengths, although ultimately only 210). The diagram depicts sensor optical ports, network connection and mechan-
the NDVI was used as the basis to discriminate between green ical features typical of all active proximal light sensors. (Adapted from [23]).

leaves and bare soil.

Unlike previous workers, Holland et al. [23] developed a
plant biomass sensor utilizing a polychromatic LED light source Alternatively, sensors with modulated light sources can be
that emitted both visible and NIR radiation. With polychromatic operated without concern for ambient background illumination.
LEDs, sensors could simultaneously measure visible and NIR With a modulated source, the modulated radiation reflected
reflectance using a single source to illuminate the target. This from the target is measured by the sensor’s synchronous detec-
approach also reduced the errors associated with the drift in irra- tion hardware. The lock-in amplification stage of the detection
diance of each wavelength as well as ensuring each wavelength electronics is capable of differentiating (and amplifying) only
component of the source illuminated precisely the same target. the modulated portion of the returning optical signal, thereby
Although much has been written about the application of excluding the influence of the background light. This unique
AOS in agriculture, the majority of the scientific literature is feature of active sensors is why they can operate equally well
“applications-oriented” and offers little detail of the underlying under all lighting conditions although some care should be
optics and radiometry of the devices. This lack of information exercised when operating them in conjunction with modulated
limits opportunities for users to develop novel but scientifically- ambient lighting or other sources.
sound applications of the sensors. To this end, a primer for An example of the mechanical structure of an AOS is shown
in Fig. 1 (adapted from [23]). The enclosure is designed to fa-
practitioners and scientists is necessary regarding the operation
cilitate the sensor’s optical functions as well as protect the elec-
and use these devices; particularly with respect to discussing
tronic circuitry (Fig. 2). The sensor is composed of: 1) optics to
the underlying physics, optics/radiometry and data manipula-
facilitate collimation and collection of radiation, 2) a modulated
tion opportunities. Our goal is to provide practitioners and sci-
light source with associated modulated driver and power con-
entists with the fundamental optical and radiometric principles
trol electronics, 3) multichannel photo-detectors, 4) high-speed
by which such devices operate and to document the basic math-
preamplifier array with ambient light cancellation, signal con-
ematical rules for using raw, unprocessed reflectance data pro-
ditioning and data acquisition circuitry, 5) a micro-control unit
duced by these sensors.
(MCU) or digital signal processor (DSP) and 6) an input/output
interface to communicate sensor data to an operator or con-
II. ANATOMY OF A PROXIMAL ACTIVE OPTICAL SENSOR troller.
Configurations of AOS electronics depend on the spectral
Active optical sensing techniques irradiate a plant canopy and characteristics of the light source. For monochromatic sources
measure a portion of the radiation scattered (reflected) from the [Fig. 2(a)], one photo-detector is utilized and each spectrally
canopy. Active sensors can have either steady-state or modu- different radiation source alternately pulsed on and off. Re-
lated light sources. With steady-state sources, care must be taken flectance measurements are typically separated, temporally,
to adequately shield the measurement scene from ambient light. by the sensor’s electronics [Fig. 2(a)]. Polychromatic sources
Spectrophotometric measurements of plant canopies typically emit more than one wavelength [Fig. 2(b)] and two or more
use an integrating sphere and a halogen lamp. Alternatively, an spectrally filtered detectors are used.
opaque chamber can be placed over the target (typically a single The light source for the sensor is composed of an array of
leaf or section of plant canopy). The target is illuminated with solid-state emitters, typically either LEDs or laser diodes (LDs).
a lamp and reflected radiation is measured with a multi-band Most LEDs or LDs have emission characteristics that are classi-
sensor or a spectrophotometer [6]. fied as being mostly monochromatic or quasi-monochromatic,
HOLLAND et al.: RADIOMETRY OF PROXIMAL ACTIVE OPTICAL SENSORS (AOS) FOR AGRICULTURAL SENSING 1795

Fig. 2. Simplified block diagram of active proximal sensor electronics when utilizing (a) monochromatic emitters and (b) polychromatic emitters.

that is, the wavelengths composing the light are strongly peaked an LED that has an InGaN-GaN structure and emit light of
about a central wavelength (Fig. 3). wavelengths between 350–470 nm. Part of this light is ab-
The spectral characteristic of an LED is defined by an sorbed by the phosphor coating and re-emits this light to a
emission band having a center wavelength (CWL) and a much broader spectrum. Phosphor coatings typically consist
spectral-line half-width [Fig. 3(a)]. The center wavelength of rare-earth element compounds that are bound in silicone
defines the peak emission wavelength of the LED and the or polymer carrier materials that are deposited directly to the
spectral-line half-width defines the spectral bandwidth (BW) surface of the LED. The elements in the phosphor material can
of the LED. However, some ultra-bright LEDs exhibit a be adjusted compositionally to produce LEDs that have various
polychromatic spectrum with a peak in the visible portion color temperatures ranging from 2700 to 6000 K. A problem
of the spectrum (CWL1) and a smaller peak in the near in- associated with all solid state light sources is that the output
frared (NIR) (CWL2) [Fig. 3(b)]. Typical wavebands for intensity of a LED is temperature dependent. Depending on the
these polychromatic LEDs are: green/NIR (570 nm/880 nm), composition of the LED, the output can typically drift from 0.3
yellow/NIR (590 nm/880 nm), orange/NIR (612 nm/880 nm) to over 1%/ C. To achieve good output stability with respect to
and red/NIR (650 nm/880 nm). These LEDs are well-suited thermal and aging effects, LEDs should be properly driven and
for plant biomass and plant pigment measurements since basic monitored. Output intensity drift can be compensated using a
vegetative calculations require a reflectance measurement in number of electronic and software techniques.
the visible portion of the spectrum and a reflectance measure- Optical sensing is performed using silicon photodiodes with
ment in the NIR portion of the spectrum. Other polychromatic filters. The spectral range of silicon photodiodes is typically
light sources include phosphor coated LEDs and emit various 400 to 1100 nm. Blue enhanced photodiodes improve sensitivity
color temperatures of white light. These light sources utilize below 450 nm. The output signals from each photo-detector
1796 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 5, NO. 6, DECEMBER 2012

Fig. 4. Simplified optical configuration of a proximal active optical sensor.

( ) with radiance , source objective with area having

transmittance and photo-detector ( ) having transmittance
. Light is projected away from the source to irradiate a target
( ) over a distance meters (target irradiance). Target ( ) re-
flects a portion of the incident radiation back toward the sensor
where it is detected by the sensor’s photo-detection electronics
(detector irradiance).
On the source side, source irradiance ( ; units W/m ) in the
near field is given by the equation

(1)

where is the transmission constant of the optical system,

is the radiance of the source (LEDs; radiance units are W/m )
and is the solid angle subtended by the illuminating radiation
(at infinity; assumed to be small).
However, even when used with sensor-target distances in the
order of meters, we are dealing with far-field optics, that is
Fig. 3. Typical emission spectra emitted of a (a) monochromatic and (b) poly- where the sensor-target distance exceeds the ratio of the lens di-
chromatic light source. Spectral peaks are defined by CWL1 and CWL2. (Ex-
tracted from [23]). ameter divided by the angle subtended by the light source from
the lens. The target irradiance is therefore governed by
are amplified by preamplifiers with ambient light compensation (2)
electronics.
The output signals from the preamplifiers are routed into the where is the area of the source at the objective lens and is
signal conditioning subsystem of the sensor. Here, the detected the distance to the target. Note the inverse-square law of optics
and amplified signals are further amplified and band narrowed is in effect with respect to beam intensity as it projects away
in accordance with the modulation phase of the light source, from the objective lens.
and conditioned for sampling by the analog-to-digital con- On the detection side, the irradiance at the detector ( ; units
verter (ADC). The ADC is utilized to sample the conditioned W/m ) is governed by radiating energy from the diffuse target
reflectance signals after which the converted signals are read of area bounded by the extent of the illuminated region, and
by the MCU where they are utilized by the processor to make following similar considerations for (2), it can be shown that
reflectance calculations. Reflectance values can be utilized
to calculate vegetation indices, quantify color measurements, (3)
classify spectral attributes or make relative distance measure-
where is the radiance of the diffuse reflecting target (irradi-
ments.
ated by the source), is the transmittance of the detection op-
tics and is the angle by which radiation from a diffuse reflector
III. RADIOMETRY OF PROXIMAL ACTIVE OPTICAL SENSORS (target) arrives at the detector. Given the small geometric offset
Fig. 4 depicts the optical geometry of a typical proximal ac- between the source and detector compared to the sensor-target
tive optical sensor. The primary elements include a light source distance, the target-detector distance is taken to be the same
HOLLAND et al.: RADIOMETRY OF PROXIMAL ACTIVE OPTICAL SENSORS (AOS) FOR AGRICULTURAL SENSING 1797

as the source-target distance, (and hitherto referred to as the

sensor-target distance). It should be noted that in the case of
active sensors, where the geometric field-of-view (FOV) of the
collection optics exceeds the illumination footprint of the target,
the extent of the illuminated target effectively defines the sensor
FOV.
The magnitude of the diffuse target radiance, , is related
to the target irradiance (2) by

(4)

where it is assumed that the reflective surface is radiating

isotropically into a hemisphere, in other words it is a Lamber-
Fig. 5. The single wavelength irradiance response ( ) of a Crop Circle
tian surface. Here is the reflectance of the diffuse target and ACS-210 sensor measurement directed at a uniform soil surface for sensor-
for plant canopies this value typically ranges from 2 to 50% target distances ranging from 0.5 to 2.75 m. Note the numerical values (s.r.u.)
depending on the wavelength of the light source. is the are “scaled radiance units” whereby the electronic output of the sensor is scaled
relative to that for a given radiance level during factory calibration. The response
irradiance of the source on the target. demonstrates a near-perfect response.
Equation (4) can be modified to account for small reflectors
having surface areas less than the area of the irradiating source
by including an area fraction ( ) to scale the magnitude of the law effects (here the additional subscripts and indicated
reflector’s radiance, namely the parameters at two different wavelengths). This, of course as-
sumes that , , and are not affected by wavelength. It is
(5) worth mentioning here that in the case of a sensor that utilizes
multiple sources or multiple detectors to achieve multi-wave-
where is the area fraction of light projected by the source and
length functionality, care must be taken to ensure geometric ef-
intercepted by the reflector.
fects (particularly at small sensor-target distances) do not vary
Substituting (2) into (5) yields the radiance at the target sur-
, , and for each wavelength used.
face:
The ratio described above is the widely used simple ratio veg-
etation index (SRI):
(6)

and substituting (6) into (3) yields the irradiance on the detector (9)
as
Equation (7) can also be applied to the popular normalized
(7) difference vegetation index (NDVI):

Equation (7) is the general equation that describes the irra- (10)
diance at the photo-detector of the AOS. The presence of the
squared distance term in the denominator will cause a signifi- where, again, and are the target reflectance values at
cant change in detected signal magnitude when the relative dis- wavelengths and . The effect of distance normalization
tance between sensor and target varies (Fig. 5). as related to the SRI calculation (9) is illustrated in Fig. 6.
Fig. 5 illustrates the inverse-square law of optics phenomena As the sensor-target distance increases, the individual sensor
for reflected signals measured by an AOS. This phenomenon irradiance values also decrease according to the inverse-square
influences both the irradiance at the target and the irradiance at law. The derived SRI calculation remains almost constant
the detector. Thus, at a given wavelength the so-called inferred although noisy until sensor-target distances exceed 2.5 m
target reflectance ( ), as determined by the sensor-only param- where the irradiance, in particular that in the 650 nm waveband,
eters, namely begins to fall below the detection limit of the photo-detector.
Closer examination of Fig. 6 reveals a very small negative
(8) slope. For a uniform target any such systematic deviation from
invariance is most likely the result of deviations from linearity
exhibits the same inverse-square law response to sensor-target in the optics/electronics for each wavelength. This also explains
distance. This is an important operational point. These single the significant deviation occurring in the calculated index at
reflectance values vary with sensor-target distance. sensor-target distances 0.5 m. Consideration should also be
Assuming the transmission characteristic of the source and given to the situation where multiple sources are used to gen-
detector optics ( and ) are wavelength independent, (7) erate multi-wavelength measurements. At small sensor-target
shows that the ratio of the sensor irradiance values at two dif- distances the source/detector geometry may result in different
ferent wavelengths allows a determination of the ratio of the wavelength- illuminated regions of the target moving outside
actual target reflectance values ( ) for those two wavelengths, the geometrical limit of the detector FOV, and thereby pro-
namely , without any inverse-square ducing different values of the area fraction, (5). Given the
1798 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 5, NO. 6, DECEMBER 2012

. For our two reflecting targets, the sensor irradi-

ance components from each of the two targets can be added to
a single value at the sensor ( ), and for our two
wavelengths and , (7) and (9) can be combined to give:

(11)

Extending this concept to the general case of reflecting tar-

gets of progressively increasing distances from the sensor (in-
cremented by the multiplier ) and where :
Fig. 6. Distance sensitivity normalization via ratio calculation using the SRI.
Data was collected from an active sensor (Crop Circle ASC-210) positioned at
nadir over a uniform soil surface. Both the 650 nm and 880 nm bands exhibit
an inverse-square law dependence while the ratio-based index, in this case SRI
(9), is noisy but invariant.
(12)

Now, consider multiple, identical, non-obscuring reflecting

surfaces (resulting in a net LAI of 1 from leaves spaced along
the entire depth of field), all oriented the same way with respect
to the sensor (hence ), and a BRDF that is approximately
Lambertian (which effectively means the target radiance is not
sensor/target-distance dependent, i.e., is ).
When the sensor-target distance is much larger than the inter-
target spacing, that is , (12) reduces to

(13)

Fig. 7. Illustration depicting two reflectors placed at distances and from

Equation (13) is that of a single reflecting surface filling the field
the sensor. Reflector 1 and Reflector 2 have fractional intercept coefficients of view of the sensor, namely (9).
and , respectively. The importance of using band ratios to avoid the effect of
the inverse-square law is further illustrated in Fig. 8(a). Here
the single-band, sensor-inferred irradiance, has been
polychromatic source used in generating Fig. 6, the illuminated recorded for an early-growth stage wheat crop (80 ha field of
target is common to all wavelengths but the geometric offset of wheat (Triticum aestevum) at growth stage Zadok 39; prior to
the two detector lenses may introduce a discernible change in “flag leaf”), on a single day and at numerous locations, to high-
, , and as discussed earlier. light the considerable spatial variability in the canopy biomass
In the previous discussion, the target was assumed to be a [24]. In this situation the sensor has been positioned such that
planar surface exhibiting Lambertian reflectance characteristics. the sensor-ground distance is (typical of most hand-held
An actual plant canopy is a three dimensional structure with ar- configurations) and the sensor response is plotted against the
bitrarily inclined reflective surfaces that do not necessarily ex- green biomass quantity (GDM; as measured by harvesting the
hibit Lambertian reflectance properties. Does the deviation from plants at soil level in a 50 cm 80 cm quadrat, oven drying,
the assumption of an ideal planar and Lambertian target surface weighing and converting the weights to kilograms of biomass
invalidate the observations? per hectare). At locations with higher biomass, the total canopy
Consider a simple case having two reflecting targets at dis- chlorophyll absorbs more of the Red radiation. A minimum net
tances and from the sensor with fractional intercept coef- scattering baseline value occurred for a biomass of 3200 kg/ha
ficients and , respectively, as shown in Fig. 7. [Fig. 8(a)]. For biomass levels 3200 kg/ha, the increased
The reflectance ( ) of each surface is an intrinsic property of crop height reduced the sensor-target distance and introduced
the surface itself. However real targets often exhibit non-Lam- an inverse-square law response whereby the measured canopy
bertian behavior and the non-isotropic nature of the scattered reflectance was observed to increase with increasing biomass.
radiation is described by the target’s bidirectional reflectance Fig. 8(b), which is the NIR response, was different.
distribution factor (BRDF). The target radiance may therefore The inverse-square law effect evident in the Red wave-
change with sensor-target distance as the sensor views a dif- length ( ), while still present, is not obvious in the NIR
ferent angular segment of the BRDF. In effect this would mean ( ) because both increasing total chlorophyll (function
the reflectance ( ) of the target would appear to change. As- of LAI) and diminishing sensor-crop distance are driving NIR
sume one leaf is located further away from the sensor than (but upwards. Ratio-based indices such as the SRI (9) and NDVI
not obscured by) the other leaf such that where (10) remove this inverse-square law effect (Fig. 9) although the
HOLLAND et al.: RADIOMETRY OF PROXIMAL ACTIVE OPTICAL SENSORS (AOS) FOR AGRICULTURAL SENSING 1799

IV. THE ALGEBRA OF RATIO-BASED INDICES

The previous mathematical analyses have been confined to
simple ratio-based indices, where the ratio of the at-sensor ir-
radiances for two wavelengths is equivalent to
the ratio of the target reflectance values . There are,
however, a multitude of reflectance indices designed to eluci-
date some aspect of a target canopy that are not simply band
ratios. To this end, it is worth noting some basic algebraic rules
for utilizing AOS irradiance measurements; namely:
1) An irradiance ratio can be multiplied by a constant term(s)
without affecting the sensor-target distance invariance of
the ratio:

(14)

where , , and are constants; .

2) A constant can be added to an irradiance ratio without af-
fecting the sensor-target distance invariance of the ratio:

(15)

3) For any two bands of a proximal sensor, the ratio of the

irradiances with non-unity exponents ( ) will render the
Fig. 8. Sensor response for an individual wavelength (a) 650 nm (Red), and resulting ratio invariant with respect to the sensor-target
(b) 880 nm (NIR) as a function of increasing green biomass (GDM (kg/ha)) for distance:
an early-stage wheat crop. Crop conditions described in [24]. Fitted curves are
indicative only.
(16)

4) For any calculation using proximal sensor irradiances of

two or more bands, the resulting calculation will be in-
variant with respect to distance to the target if the resultant
power of the products in the numerator product is equal to
power of the denominator product:

(17)

5) A series of irradiance ratios may be summed and the re-

sultant calculation will be invariant with respect to the
sensor-target distance so long as the exponent of the nu-
merator and denominator for each term is identical:

(18)

A consequence (16) pertains to the use of non-equal expo-

nents in the denominator and numerator terms. When these ex-
ponents are not equal, the at-sensor irradiance values ( )
Fig. 9. Ratio-based index response as a function of increasing green biomass can be utilized to measure, or incorporate, relative sensor-target
(GDM (kg/ha)) for an early-stage wheat crop. Crop conditions described in [24].
(a) SRI (9), and (b) NDVI (10). Fitted curves are indicative only.
distance into vegetation index calculations. Here the inverse-
square law effect on individual wavelength AOS measurements
can have some potentially useful implications.
calculated values at high biomass levels ( 3200kg/ha) exhibit The above discussions have shown that calculating a ratio-
more variability. based index such as (9) and (10) will remove inverse-square law
1800 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 5, NO. 6, DECEMBER 2012

effects. However, as evidenced in Fig. 8(a), under high biomass

situations (for example 2000–3200 kg/ha) the single-band re-
flectance of a plant canopy may “saturate”, that is, no change
in reflectance occurs as biomass changes. This happens fre-
quently when vegetation indexes incorporating Red wavebands
are utilized for biomass measurements. Consider the NDVI re-
sponse in Fig. 9(b). Even when correcting for apparent inverse-
square law effects, the NDVI remains insensitive to total canopy
chlorophyll changes 3200 kg/ha. However, analyzing each
band independently shows that the NIR band often tends to in-
crease linearly with increasing biomass [Fig. 8(b)] while the
Red reflectance decreases hyperbolically [in Fig. 8(a) for the
range 0 to 3200 kg/ha]. This is due to that fact that the NIR re-
flectance is a function of scattering within the mesophyll layer Fig. 10. Linear distance measurement using square root of inverse irradiance.
of plant leaves and higher leaf transmittance, thereby facili- The NIR channel of a Crop Circle ACS-210 ( was utilized for the cal-
culation using (20). Reflectance measurement was made with sensor positioned
tating deeper canopy inter-leaf scattering effects (for example, at nadir over a uniform soil surface. The coefficients , and were set to 0.5,
as observed in [25]). However, as the leaf chlorophyll content 0.627, and 0.140, respectively.
increases, the Red reflectance decreases. The result is that for
high LAI and high chlorophyll-content canopies, vegetation in-
dices such as SRI and NDVI may not discriminate either LAI or
chlorophyll content variations. This “saturation” effect has been
widely cited in the literature.
The sensitivity of the single waveband response to sensor-
target distance variations is not necessarily an undesirable arti-
fact that must be corrected out by ratio-based indices. Indeed the
scientific literature suggests that including other measures, such
as plant canopy height along with ratio-based indices such as
NDVI increases the level of variance in the sensor response ex-
plained by variations plant canopy morphology. One important
example is attributed to canopy nitrogen (N) levels [26]. Active Fig. 11. Distance-sensitive simple ratio measurement using square root of nu-
merator irradiance. Reflectance measurement was made over a uniform soil sur-
sensors, like the devices examined thus far, offer the opportu- face. The coefficient was set to 0.5.
nity to measure both ratio-based indices as well as proximity
to the reflecting target. Two examples are presented below that
exploit the inverse-square law for measuring distance to a scat- known distance values allows determination of the slope and in-
tering target. tercept constants, yielding a distance measurement for a target
with approximately constant reflectance ( ) (Fig. 10).
A. Distance Measurement Using Square Root of Inverse
Irradiance B. Distance Measurement Using Dissimilar Powers
Assuming there is no appreciable change in target reflectance A form of the ratio-based index that is sensitive to the distance
with distance ( ), the sensor response to the inverse-square law between the sensor and target is
(sensor-target distance, ) follows:
(21)
(19)

where is the single wavelength detector irradiance, and where and are the sensor-inferred irradiances at two
and are constants related to the ADC conversion of detector wavelengths and , like in (20), is an irradiance expo-
irradiance to output values and the ADC offset, respectively. In- nent. The special case of when is equal to 0.5 (corresponding
verting (19) allows the sensor-target distance, , to be measured to the inverse-square law) will again result in a linear response
via a two-point calibration process using to sensor-target distance (Fig. 11).

(20) V. CONCLUDING REMARKS AND THE FUTURE OF AOS

Active optical sensors are likely to assume a greater role in
Here is an irradiance exponent which, owing to the in- agricultural remote sensing, primarily owing to the convenience
verse-square law is equal to 0.5 (but this can be optimized if of being able to operate them under any illumination condi-
not) and and are the slope and intercept constants, respec- tions and for the potential to incorporate them into systems that
tively. When the exponent is set to 0.5, the inverse-square law make real-time variable-rate applications (e.g., fertilizer). Fur-
response of the irradiance measurement to sensor-target dis- ther AOS development will most likely pertain to improving
tance is transformed into a linear response (Fig. 10). Using two sensor spectral selectivity and/or spatial resolution. Distance
HOLLAND et al.: RADIOMETRY OF PROXIMAL ACTIVE OPTICAL SENSORS (AOS) FOR AGRICULTURAL SENSING 1801

measurement and spatial resolution improvements in conjunc- [10] C. F. Jordon, “Derivation of leaf-area index from the quality of light
tion with multispectral reflectance measurements will provide on the forest floor,” Ecology, vol. 50, pp. 633–666, 1969.
[11] J. W. Rouse, J. R. Haas, J. A. Schell, and D. W. Deering, S. Freden,
more information concerning a plant’s biophysical attributes. E. Mercanti, and M. Becker, Eds., “Monitoring vegetation systems in
The combination of these elements is vital to the development the great plains with ERTS,” in Proc. 3rd Earth Resources Technology
of robust biophysical models needed for the accurate application Satellite-1 Symp., Washington, DC, 1974, pp. 309–317.
of agrochemicals or for the selection of particular plant genetic [12] A. R. Huete, “A soil-adjusted vegetation index (SAVI),” Remote Sens.
Environ., vol. 25, pp. 295–309, 1988.
characteristics. Additionally, technological improvements per- [13] P. Gong, R. Pu, G. S. Biging, and M. R. Larrieu, “Estimation of forest
taining to longwave and shortwave solid-state emitters and asso- leaf area index using vegetation indices derived from Hyperion hyper-
ciated detector technologies will allow AOS technology to per- spectral data,” IEEE Trans. Geosci. Remote Sens., vol. 41, no. 6, pp.
form measurements into the mid infrared and ultraviolet regions 1355–1362, 2003.
[14] D. Haboudane, J. R. Miller, E. Pattey, P. J. Zarco-Telada, and I. B.
of the optical spectrum, respectively. Currently, solid-state poly- Strachan, “Hyperspectral vegetation indices and novel algorithms for
chromatic light source technology has a rather limited spec- predicting green LAI of crop canopies: Modeling and validation in the
tral range extending only a few hundred nanometers in the vis- context of precision agriculture,” Remote Sens. Environ., vol. 90, pp.
337–352, 2004.
ible and NIR bands. Applications utilizing extended wavelength
[15] A. M. A. Ferté and A. Balp, “Agricultural Machine,” U.S. Patent No.
ranges include assessing the plant water and/or disease status, or 2,177,803, 1938.
in the case of soil, for determining soil organic matter compo- [16] J. Palmer and G. M. Owen, “Automatic control of sugar beet singling
sition. and thinning by means of an on-line digital computer,” J. Agricultural
Engineering Research, vol. 16, no. 2, pp. 107–125, 1971.
Notwithstanding the above comments, ground-based and
[17] G. R. Henderson and A. G. Grafton, “Method and Apparatus for Mea-
proximal active optical sensing remains a viable alternative suring Light Reflectance Absorption and/or Transmission,” U.S. Patent
to field scale surveying of photosynthetically-active biomass. No. 3,910,701, 1975.
Thus a solid understanding of AOS radiometry is essential [18] J. V. Stafford, G. S. Weaving, and J. C. Lowe, “A portable infra-red
moisture meter for agricultural and food materials: Part 1, Instrument
to correctly use and interpret data from these active proximal development,” J. Agricultural Engineering Research, vol. 43, pp.
sensors. The radiometry fundamentals presented in this paper 45–56, 1989.
provide the basis for utilizing measurements made by the [19] J. L. Beck and T. Vyse, “Structure and Method for Differentiating One
sensors in more complex mathematical functions. Finally, Object From Another Object,” U.S. Patent No. 5,296,702, 1992.
[20] J. L. Beck and M. L. Kinter, “Detecting Plants in a Field by Detecting
manipulation of the individual wavelength response data can a Change in Slope in a Reflectance Characteristic,” U.S. Patent No.
provide useful information regarding the distance variation 5,789,741, 1996.
between the sensor and the target. Careful consideration of [21] J. L. Beck and M. L. Kinter, “Appartus and Method for Determining a
the discussions in this document should avoid many potential Distance to an Object in a Field for the Controlled Release of Chemicals
on Plants, Weeds, Trees or Soil and/or Guidance of a Vehicle,” U.S.
pitfalls. Patent No. 5,585,626, 1994.
[22] A. Paap, S. Askraba, K. Alameh, and J. Rowe, D. Sampson, S. Collins,
K. Oh, and R. Yamauchi, Eds., “Photonic-based spectral reflectance
sensor for ground-based plant detection and weed discrimination,” in
REFERENCES Proc. SPIE, 19th Int. Conf. Optical Fibre Sensors, Perth, W. Aust.,
2008, vol. 7004, pp. 5K.1–5K.4.
[1] D. W. Lamb, M. G. Trotter, and D. A. Schneider, “Ultra low-level air- [23] K. H. Holland, J. S. Schepers, J. F. Shanahan, and G. L. Horst, D. J.
borne (ULLA) sensing of crop canopy reflectance: A case study using a Mulla, Ed., “Plant canopy sensor with modulated polychromatic light
Crop Circle™ sensor,” Computers and Electronics in Agriculture, vol. source,” in Proc. 7th Int. Conf. Precision Agriculture and Other Preci-
69, pp. 86–91, 2009. sion Resources Management, Minneapolis, MN, 2004.
[2] M. G. Trotter, D. W. Lamb, G. E. Donald, and D. A. Schneider, “Ac- [24] D. W. Lamb, D. A. Schneider, M. G. Trotter, M. T. Schaefer, and I. J.
tive optical sensors for quantifying and mapping pasture biomass: A Yule, “Extended-altitude, aerial mapping of crop NDVI using an ac-
case study using red and near infrared waveband combinations from a tive optical sensor: A case study using a Raptor™ sensor over wheat,”
Crop Circle™ in Tall Fescue (Festuca arundinacea) pastures,” Crop Computers and Electronics in Agriculture, vol. 77, pp. 69–73, 2011.
and Pasture Science, vol. 61, pp. 389–398, 2010. [25] D. W. Lamb, M. Steyn-Ross, P. Schaare, M. Hanna, W. Silvester,
[3] D. Inman, R. Khosla, and T. Mayfield, “On the go active remote sensing and A. Steyn-Ross, “Estimating leaf nitrogen concentration in ryegrass
for efficient crop nitrogen management,” Sensor Review, vol. 25, pp. (Lolium spp.) pasture using the chlorophyll red-edge: Theoretical mod-
209–214, 2005. elling and experimental observations,” Int. J. Remote Sens., vol. 23, pp.
[4] F. Solari, J. Shanahan, R. Ferguson, J. Schepers, and A. Gitelson, “Ac- 3619–3648, 2002.
tive sensor reflectance measurements of corn nitrogen status and yield [26] A. Thomsen and K. Schelde, “Nitrogen fertilisation based on double
potential,” Agronomy J., vol. 100, pp. 571–579, 2008. sensor measurements,” NJF Seminar 390: Precision Technology in
[5] K. H. Holland and J. S. Schepers, “Derivation of a variable rate nitrogen Crop Production – Implementation and Benefits. NJF Report, vol. 2,
application model for in-season fertilization of corn,” Agronomy J., vol. no. 8, pp. 42–50, 2006.
102, pp. 1415–1424, 2010.
[6] M. M. Hanna, D. A. Steyn-Ross, and M. Steyn-Ross, “Estimating
biomass for New Zealand pasture using optical remote sensing tech-
niques,” Geocarto International, vol. 14, pp. 89–94, 1999.
[7] R. Künnemeyer, P. N. Schaare, and M. M. Hanna, “A simple reflec-
tometer for on-farm pasture assessment,” Computers and Electronics
in Agriculture, vol. 31, pp. 125–136, 2001. Kyle H. Holland (M’93) received B.Sc. and M.Sc. degrees in electrical en-
[8] L. A. Marihart, “Photoelectric Unit for Agricultural Operation,” U.S. gineering from Iowa State University in 1988 and 1990, respectively, and the
Patent No. 2,682,132, 1948. Ph.D. in biomedical engineering from Iowa State University in 1993.
[9] G. Falzon, D. W. Lamb, and D. Schneider, “The dynamic aerial survey He is President and Principal Engineer for Holland Scientific, Inc. in Lin-
algorithm architecture and its use in airborne fertilizer applications,” coln, NE. His current research activities/interests include active proximal sensor
IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens. (JSTARS) , 2012, R&D, biophysical plant nutrient modeling, real-time sensor-based control sys-
DOI 10.1109/JSTARS.2011.2179020. tems and high-speed flow control design for agricultural applicator systems.
1802 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 5, NO. 6, DECEMBER 2012

David W. Lamb (M’04–SM’07) received the Ph.D. James S. Schepers received the B.S. degree in 1968
in physics at the University of New England (UNE) in Agricultural Honors and the M.S. degree in soil
in gaseous electronics, and subsequently completed chemistry in 1970 from the University of Nebraska.
postdoctoral work in physical chemistry elucidating He received the Ph.D. in soil physical chemistry from
the anistropic nature of various electric and magnetic the University of Illinois in 1973.
properties of molecules. Prior to commencing a lec- From 1975 to 2008 he was employed by the
turing appointment at UNE in 2002, he taught physics USDA–Agricultural Research Service where he
at Charles Sturt University for eight years. At UNE worked to develop technologies to help producers
he has established major national research projects reduce nitrate losses to groundwater, including
in precision agriculture, applied photonics, and re- remote sensing and active crop canopy sensors.
mote and proximal sensing technologies for moni- He currently serves as co-editor for the Journal of
toring plants, animals and soil. He is presently head of Physics and Electronics Precision Agriculture and consults with various agricultural businesses.
in the School of Science & Technology and heads UNE’s Precision Agriculture
Research Group.