5. Spectral characteristics of Cannabis
The studies examining the spectral signature of Cannabis are limited and often
contradictory in nature regarding the actual separability of Cannabis from other
vegetation.
In general all healthy green vegetation has the same general spectral
properties (i.e. shape as defined by the location of absorption features or reflective peaks)
(Figure 3).
The general shape of the spectral response of healthy green vegetation.
Reflectance is reported as a unit-less ratio where 0.5 corresponds to 50% of incident
radiation being reflected.
The region centered around 550nm is known as the green peak in vegetation
spectra and is representative of photosynthetic pigments such as chlorophylls a and b and
photoprotection pigments associated with the xanthophyll cycle [13, 29-31]. Healthy,
green vegetation has strong absorption in the red wavelengths (620-700nm) and high
reflectance in the near infrared. The inflection point between the two is referred to as the
red-edge and can be related to foliar nitrogen and chlorophyll content [16]. The plateau
in the near infrared region is also a characteristic of vegetation reflectance spectra[32]; it
is predominantly controlled by internal structure at the leaf level [14, 32] and is also
influenced by leaf area index (total leaf area per ground area), water content, plant
physiology and stress at the canopy level [14].
One of the earliest publication examining the spectral properties of Cannabis is
[33]. In that study, the reflectance and transmittance of leaves from several varieties of
C. sativa was compared to the reflectance and transmittance of other vegetation native to
the East coast of the United States. Both sides of the leaves were measured along with the
effect of the application of nitrogen fertilizer. The wavelengths with the greatest
differences through time (i.e. plant growth) were reported to be around 700nm and 550-
600nm. Similar wavelengths were also reported to be influenced by the nitrogen
fertilization. Differences between strains of Cannabis were centered around 550nm and
near 720nm but were found to be minimal. The methodology used to assess differences
between spectra in this study constituted subtracting one spectrum from another. This
however, does not take into account the unusual properties of such hyperdimensional
data[34]. In comparison to the leaf spectra of seven tree species and eight herbaceous
species the greatest difference was observed in the 550 and 720nm regions of the
spectrum. The greatest similarity was found between Cannabis and herbaceous
vegetation in comparison to Cannabis and tree reflectance.
Other more recent studies examine the spectral signatures of Cannabis both at the
ground level and from airborne hyperspectral imagery [9, 35, 36]. Though findings were
inconclusive, [36] illustrate a potential unique feature in the second derivative of the
reflectance spectra at 695nm of Cannabis in comparison to other vegetation from
airborne imagery (19 band CASI data). Upon further analysis [36] determined however,
that the feature was not unique to Cannabis. Data analysis by [36] was conducted on
radiance data, not calibrated reflectance.
In comparison, [35] illustrate that the most likely reason for the unique hazy
emerald-green reflectance of Cannabis reported by trained spotters is due to specular
reflectance from the waxy cuticle layer of the leaves and from scattering due to
microscopic structures on the leaf surfaces. The scattering of blue sky-light was
hypothesized to add to the green reflectance (i.e. green peak) inherent in green vegetation
to create an emerald-green quality to the spectral signature of Cannabis. Thus, [35] state
that in clear conditions, the blue component of the reflectance will be stronger than under
cloudy conditions where green reflectance and internal leaf elements would be dominant.
The contrast of the emerald-green appearance of the Cannabis may be enhanced by the
introduction of fertilizer and elimination of water stress especially if other vegetation is
stressed due to a lack of water, nutrients or is subject to other conditions such as pests.
The canopy architecture of Cannabis was hypothesized to the cause of the hazy effect
when the plants are seen from a distance. In contrast to [33], no significant differences
were observed between the leaf reflectance of different strains/varieties, leading to the
conclusion by [35] that THC content has little effect on reflectance.
Interviews conducted with spotters indicated that Cannabis may have a unique directional
reflectance feature.
Photographs of maximum and minimum polarized reflectance
indicated that the waxy cuticle may in certain cases have a strong enough reflectance for
the leaves to appear white. Examination of cold-stage electron micrographs indicated that
the leaves are covered with patches of aligned rods, only a fraction of a visible
wavelength of light in diameter and multiple wavelength dimensions long. In addition,
the orientation of the rods in adjacent patches was found to be different. A waxy substrate
was also seen beneath the rods. The contribution of the rods and the waxy substrate were
found to contribute to the preferential scattering of blue light[35].
Using 16 band
airborne imagery, [35] concluded that the sixteen bands from the visible through near
infrared wavelengths were insufficient for discriminating Cannabis from other vegetation
types. The analysis methodology of the airborne imagery consisted of conventional
supervised and unsupervised classification across the entire airborne scene. The
Mahalanobis supervised classification provided the best results using various
combinations of the bands at 780, 800, 850, 880, 990nm. Many classification errors led to
the conclusion that the method was not successful at discriminating Cannabis from the
imagery.
In a recent study [9] examine the spectral reflectance of Cannabis at the leaf level
and from airborne and satellite imagery in British Columbia. Using a combination of
feature selection and pattern classification (machine learning)[37] perfect separability
was achieved with ten wavelengths (bands) at the leaf level. The areas of greatest
separability at the leaf level included the blue wavelengths and the region between 630-
680nm. The 550nm region previously reported was not found to be useful in separating
the reflectance of Cannabis from other herbaceous vegetation
(Figure 4).
Differences were also seen between the spectra between Cannabis growing in
two geographic regions investigated by [9].
The airborne imagery investigated in this study was the same as used by [36], converted
to reflectance. Results from the in-situ study were used to guide the examination of the
separability of the pixels indicated to represent known grow-ops. Using an n-dimensional
visualization of the spectra and a measure of the spectral angle locations were highlighted
as having the spectral signature the most similar to the known sites of Cannabis. In
addition, an assessment of high resolution satellite imagery illustrated a potential for
locating large grow operations but due to the limited spectral resolution [9] concluded
that hyperspectral imagery provided the greatest separability.
In ongoing surveys, the United Nations Office on Drugs and Crime (UNODC) has
been using satellite imagery to monitor the growth of illicit crops such as Cannabis,
opium and coca[38-44]. Currently, methodologies employed in the surveys conducted in
the different countries are not standardized, but follow similar principles of exploiting the
phonological differences between the crops and other vegetation. Multispectral imagery
(i.e. SPOT, IKONOS, Quickbird), containing only a few bands sensitive to radiation
spanning from the blue to shortwave infrared wavelengths, is the type of imagery most
often used in the UNODC surveys due to its relatively low cost and high repeat
acquisition. The illicit crop plantations in these countries are generally large in size and
limited efforts have been made to hide them in comparison to plantations in North
America which are considerably smaller and in many cases significant efforts have been
made by the growers to make their sites difficult to locate; posing a more difficult
detection problem.
As has been shown by [45], environmental factors greatly affect the spectral
signature of vegetation and as [8] found, specifically for Cannabis, the chemical
composition of the plant is affected. It is not currently known how these changes are
manifested in the spectral response of Cannabis but would need to be investigated in
order for region insensitive models to be developed.
