Professional Documents
Culture Documents
net/publication/239455366
CITATIONS READS
352 1,012
1 author:
Paul M Rich
Creekside Center for Earth Observation
81 PUBLICATIONS 7,854 CITATIONS
SEE PROFILE
All content following this page was uploaded by Paul M Rich on 27 May 2014.
Rich, P.M. 1990. Characterizing plant canopies with hemispherical photographs. Remote Sensing
Reviews 5:13-29.
Paul M. Rich
Paul M. Rich, Department of Biological Sci- of photographs. Digital image analysis tech-
ences, Stanford University, Stanford, CA, niques have recently been developed that al-
94305, USA. low for efficient analysis of large numbers of
____________________________________ photographs (Chazdon and Field, 1987b:
Rich, 1988, 1989; Becker et al., 1989).
I. INTRODUCTION
Hemispherical photography can theoretically
be used to study any plant canopy, including
A. The Nature of Hemispherical Canopy
forests, shrublands, and crops. The primary
Photography
limitations are the size of the camera setup
and the ability to position the setup within or
Hemispherical (fisheye) canopy photography
above the plant canopy. Photographs can be
is a technique for characterizing plant cano-
taken along transects or in horizontal or ver-
pies using photographs taken through an ex-
treme wide-angle lens (Anderson, 1964). tical grid patterns to sample spatial heteroge-
Typically, the viewing angle approaches or neity within canopies. Such sampling de-
signs can be used to map light penetration
equals 180°. Hemispherical photographs can
be taken either looking upward from within a and stand structure characteristics as a func-
tion of horizontal and vertical position (Fig-
canopy or looking downward from above a
canopy. The resulting photographs serve as ure 1), Dynamics and temporal variation can
be monitored by repeated sampling from the
permanent records of the geometry of canopy
same camera positions. Hemispherical pho-
openings. The geometric distribution of
tographs can be used to characterize forma-
openings can be measured precisely and used
tion and closure of canopy gaps through time
to estimate potential solar radiation penetra-
and to monitor seasonal changes in foliage
tion through openings and to determine as-
densities. Canopy structure and light envi-
pects of canopy architecture such as ground
ronment can be sampled above individual
cover, leaf area index (LAI), and leaf angle
plants for studies of demography and eco-
distribution. Hemispherical photographs can
physiology. Thus applications of the tech-
be analyzed by hand using sampling grids;
nique range from ecosystem level characteri-
however, hand analysis is extremely tedious
zation of canopy architecture to assessment
and generally impractical for large numbers
of local microenvironments.
PAUL M. RICH
14
CANOPY STRUCTURE: HEMISPHERICAL PHOTOGRAPHY
15
PAUL M. RICH
lected for vegetation or floristic studies, Direct sunlight penetration through canopy
hemispherical photographs can be collected openings is calculated by determining the
for studies of plant canopies. In general, intersection of the sun path with canopy
hemispherical photographs are best stored as openings. The apparent sky direction of the
negatives or transparencies. Photographic sun can be calculated with great accuracy, as
media have distinct advantages of low cost it varies with latitude, time of day, and time
and high resolution. As permanent records, of year (List, 1971; see also Pearcy, 1989;
hemispherical photographs can be studied Rich, 1989). Direct sunlight coming from a
using existing analytical methodology and given sky direction can be approximated by
saved for future study as methods are further correcting for attenuation as a function of the
developed and refined. pathlength that must be traversed through the
atmosphere (Gates, 1980). Variations due to
cloudiness and other site-specific atmos-
C. Estimation of Solar Radiation Penetra- pheric conditions are not easily predicted and
tion are best measured empirically. In the ab-
sence of empirical data, it is most common to
Hemispherical photographs taken looking calculate potential direct sunlight penetration
upward can be used to estimate solar radia- with only a correction for atmospheric at-
tion penetration to a given position in a plant tenuation.
canopy. The photographs must be taken such
that the film plane is level and the orientation Diffuse skylight penetration through canopy
relative to north is known. Radiation flux openings is calculated by assuming a distri-
density is calculated as the sum of direct bution of skylight as a function of sky direc-
sunlight and diffuse skylight that passes un- tion. Two diffuse skylight distributions are
impeded through canopy openings. Calcula- in common usage, the Uniform Overcast Sky
tion of solar radiation penetration from hemi- (UOC) and the Standard Overcast Sky (SOC)
spherical photographs is based on the as- (see Pearcy, 1989; Rich, 1989). The UOC
sumption that canopy openings allow unim- assumes that diffuse radiation flux is the
peded passage of solar radiation and foliage same from all sky directions. The SOC as-
absolutely blocks penetration. This simple sumes a distribution that depends on zenith
model neglects scattered radiation, radiation angle, with more diffuse radiation flux from
reflected from or transmitted through leaves. sky directions toward the zenith. It is widely
Using theoretical or empirical relations that recognized that these distributions do not
describe the way direct sunlight and diffuse adequately describe diffuse skylight distribu-
skylight fluxes vary as a function of sky di- tions, but no better alternatives are currently
rection and with knowledge of the distribu- available. Though somewhat arbitrary, the
tion of canopy openings, it is possible to UOC and SOC do allow standardization of
make reasonable estimates of solar radiation calculations between researchers. More re-
penetration to given positions under a plant search is required to provide diffuse skylight
canopy. Calculations of direct sunlight and distributions for both clear and overcast days,
diffuse skylight are generally expressed as and to determine site-specific diffuse sky-
direct and indirect site factors. Direct site light distributions that integrate contributions
factor is the proportion of direct sunlight and from clear and overcast days.
indirect site factor is the proportion of diffuse
skylight under the canopy relative to that Ideally, long-term monitoring of direct
outside the canopy. sunlight and diffuse skylight can provide
16
CANOPY STRUCTURE: HEMISPHERICAL PHOTOGRAPHY
empirical calibrations for a particular site. density (mole/m2/s) is useful for studying
However, reasonable results can be obtained radiation available for photosynthesis;
using theoretical sunlight and skylight distri- whereas, energy flux (W/m2) from total solar
butions. When practical, it is desirable to radiation is most useful for calculating en-
establish long-term, site-specific records of ergy balances. Empirical or theoretical mod-
solar flux. Monitoring is accomplished using els used to describe the sky distributions of
a pair of light sensors (pyranometers for total direct sunlight and diffuse skylight must be
radiation or quantum sensors for photosyn- tailored to the range of relevant radiation
thetically active radiation) attached to a data wavelengths; though for many purposes
logger, one to monitor global radiation flux these distributions are not expected to be
and the other to monitor radiation flux from very different.
skylight. The light sensors should be posi-
tioned in the open; one sensor should be un- A cosine correction can be included in calcu-
obstructed and used to monitor global radia- lations of light interception by a flat surface
tion flux, and the other sensor should be fit- within a canopy. Such a cosine correction
ted with a shadowband, to block radiation should be calculated relative to the normal of
along the sun path, and used to monitor dif- the plane of interception. For a horizontal
fuse radiation flux from skylight (Becker, surface, the normal is in the zenith direction.
1987). Radiation flux from direct sunlight Horizontal surface cosine corrections are use-
can be calculated as the difference between ful when making comparisons between
global and diffuse radiation flux. hemispherical radiation flux estimates and
measurements from cosine corrected light
The relative contribution of direct sunlight sensors. However, for purposes where it is
and diffuse skylight to global radiation flux desirable to measure solar radiation from all
can either be determined empirically or esti- directions, no cosine correction should be
mated from atmospheric clearness (Decker, included. An example of such a case would
1987). A typical atmospheric clearness of be calculation of the potential light intercep-
0.65 implies that 70-90% of global radiation tion by a plant that has leaves at many orien-
flux comes from direct sunlight. Radiation tations. In some cases it may be most mean-
flux estimates from photographs taken be- ingful to weight incoming radiation by the
neath plant canopies can be expressed in ab- leaf angle distribution for a particular plant or
solute units, such as mole/m2/s, or as a pro- level within a canopy.
portion of radiation flux in the open. When
possible both absolute and relative units Radiation flux can be predicted from hemi-
should be presented. Absolute units define spherical photographs for periods ranging
the biophysical limitations for a site; from minutes, to months, to an entire year.
whereas, relative units indicate the attenua- Determination of exact timing of direct
tion caused by the canopy itself. sunlight penetration (sunflecks) within a day
is theoretically possible, within the resolution
Analyses of radiation flux can be applied for of digitization. Such measurements require
any range of solar radiation wavelength, and very precise registration of photographic im-
to either photon or energy flux, because the ages for determination of sky directions
basis for analysis is the geometry of canopy (Chazdon and Field, 1987b). Seasonal varia-
openings, not direct measurements of radia- tion in light penetration for evergreen cano-
tion flux. For example, photosynthetically pies can readily be calculated from a single
active radiation (400-700 nm) photon flux photograph, based on the assumption that
17
PAUL M. RICH
canopy openings do not change significantly within a canopy for any wavelength by using
through the year. Integrated totals for an en- inverted values of LAI and mean leaf inclina-
tire year can similarly be calculated from a tion angle in a radiative transfer model, for
single photograph, based on the same as- example the model of Idso and de Wit
sumption. In cases where canopies undergo (1970). The radiative transfer model would
seasonal or directional changes during the be used to estimate the scattered radiation,
course of a year, a set of photographs can be while the estimate non-intercepted direct
taken at time intervals and analyzed indi- sunlight and diffuse skylight radiation would
vidually for given periods of the year. For come directly from the hemispherical photo-
calculations, sky distributions of direct graph. Although the inversion algorithm
sunlight and diffuse skylight should be for- may assume random foliage distribution, ra-
mulated with respect to the time period being diation models generally make the same as-
examined. sumption, so errors from this assumption are
minimized. The quantity most likely to be in
error from the inversion is the LAI and incli-
D. Measurement of Canopy Architecture nation angle, and these quantities usually
compare well with direct measurements.
Hemispherical photographs can be analyzed Furthermore, scattered radiation is not very
to infer various aspects of canopy architec- sensitive to randomness assumptions (Ross,
ture, including cover, foliage distribution, 1981).
LAI, and leaf angle distribution. Hemi-
spherical photography is an excellent alterna-
tive for obtaining gap fraction data that can III. METHODS
be used as input for inversion models to cal-
culate structural features of a canopy (Nor- A. Field Acquisition of Photographs
man and Campbell, 1989). A full discussion
of inversion techniques and alternative means Field acquisition involves taking hemispheri-
for measuring gap fractions is presented by cal photographs within or above plant cano-
Welles (1990) in this volume. Gap fraction pies. The goal is to obtain high contrast pho-
inversion techniques can produce excellent tographs from a known position and orienta-
estimates of LAI and leaf orientation for full- tion. Typically, set-up and acquisition of
cover canopies as well as heterogeneous each photograph requires less than five min-
canopies. Gap fraction data can be calcu- utes, so many samples can be rapidly ob-
lated from hemispherical photographs taken tained. The camera fitted with a hemispheri-
looking either upward from within a canopy cal lens should be supported so that the film
or downward from above a canopy (Norman plane is horizontal and the camera is oriented
and Campbell, 1989). Hemispherical pho- relative to north. A good practice is to align
tography offers distinct advantages over magnetic north with the top of the camera
other techniques, because it directly exam- and to correct for magnetic declination dur-
ines the distribution of gap fractions at high ing analysis. The camera set-up can be sup-
resolution, because photographs can be rap- ported on a tripod or monopod when taking
idly acquired in the field, and because the photographs pointed upward, or on a cantile-
photographs provide a permanent archival vered boom when taking photographs
record of gap fractions. pointed downward. Various researchers have
Results from hemispherical photographs can constructed self-leveling mounts that can be
also be used to estimate radiation levels supported on a tripod or monopod (for ex-
18
CANOPY STRUCTURE: HEMISPHERICAL PHOTOGRAPHY
ample, Rich, 1989). Meticulous care should ter wheels, filters can be mounted internally
be taken to level and orient the camera set- on the camera side of the lens. Hemispheri-
up, especially when it is desirable to register cal photographs taken looking upward are
sun paths with the photographs. Some re- best taken on overcast days or early or late in
searchers have found it useful to use small the day to achieve even sky illumination.
lights mounted at the edges of the field of Exposures can be determined using internal
view to mark the image boundaries (for ex- camera light meters or a series of spot read-
ample, Chazdon and Field, 1987b; Becker et ings with a hand-held light meter. Test pho-
al. 1989). The position of a number printed tographs should be taken to determine proper
by a camera databack can be useful both to exposure. A shutter speed of 1/125 second
uniquely label each photograph and to serve or greater will generally freeze foliage
as internal reference that can be used for de- movement caused by wind. Often it is advis-
termining image boundaries. Lights mounted able to take photographs at more than one
on poles in a known direction within the field exposure for each sample position, for exam-
of view can provide an internal cross-check ple, “bracketing” the exposure (taken at, one
of image orientation. F-stop above, and one F-stop below the me-
tered reading). If the lens has a focus ad-
Generally, high speed black-and-white film justment, the focus ring should be set at in-
should be used when looking upward from finity and taped in place to avoid slippage.
beneath canopies (see Rich, 1989) and black- Film should be developed promptly, the
and-white infrared film should be used when quality of photographs verified, and any poor
looking down on canopies. Satisfactory re- quality photographs retaken. Negatives or
sults can also be obtained with color trans- transparencies can be stored in archival-
parency film when looking upward. Some quality storage sleeves or plastic holders.
researchers have preferred high contrast, fine Table 1 provides a listing of equipment and
grain copy film (see Cohen and Fuchs, supplies for image acquisition in the field.
1987). To enhance contrast between foliage The listing includes information about vari-
and sky when working with visible light, a ous major manufacturers of hemispherical
red filter can be used on overcast days and a lenses, including Canon, Minolta, Nikon and
blue filter on sunny days. A filter that blocks Olympus. Inexpensive fisheye adapter lenses
visible wavelengths can be used for infrared that screw on in front of standard photo-
photographs. Many hemispherical lenses graphic lenses have also been used success-
have built in filter wheels. For lenses with fully (see Chazdon and Field, 1987b).
out filter wheels or filters not included in fil
19
PAUL M. RICH
20
CANOPY STRUCTURE: HEMISPHERICAL PHOTOGRAPHY
The functioning of CANOPY serves to illus- opening, and then, determining the contribu-
trate the steps involved in analysis of hemi- tions of direct sunlight and diffuse skylight
spherical photographs. The fundamental from the corresponding sky direction if it is
steps of analysis involve input of the nega- an opening. To enhance performance, many
tives, determination of a threshold that dis- of the calculations are made ahead of time
tinguishes openings from foliage, automated and stored in look-up tables. Direct site fac-
calculation, anti automated output of results tor, indirect site factor, and gap fraction data
(Rich, 1989). Precise registration of the (openings as a function of zenith and azimuth
negatives is critical to allow calculation of angles) are displayed on the screen or option-
sky directions. First negatives are placed in a ally output to data files. The system is fully
backlit film holder. Backlighting of nega- configurable for a given site. It can incorpo-
tives reduces problems of uneven lighting, rate either empirical or theoretical distribu-
which result when using reflected light to tions of direct sunlight and diffuse skylight.
digitize prints. Images are input with a solid- Sunlight and skylight distributions are calcu-
state black-and-white video camera fitted lated in a way that can accommodate differ-
with a macro lens. The images are digitized ent hemispherical projections and corrections
at 512(h) x 480(v) pixel resolution using a for lens distortion. A cosine correction can
framegrabber digitizer/display adapter that also be included, as can corrections for other
simultaneously digitizes and displays at the interception surfaces (for example, a correc-
rate of 30 images per second. With the use tion that parallels leaf angle distribution).
of hardware lookup tables, the digitized
negatives are displayed as positives in real Currently there are no commercially avail-
time on an image display screen. The dis- able systems for analysis of hemispherical
play screen includes a grid that facilitates ad- canopy photographs. Table 2 provides a list-
justment of the size, translation, and rotation ing of equipment for assembly of an image
of the image. Once the image is properly analysis system based on an IBM-compatible
positioned, a threshold is determined that dis- personal computer. This system is designed
tinguishes openings from foliage. This is for use with the program CANOPY (Rich,
determined by comparing a threshold image, 1988, 1989). As an add-on to an existing
in which all pixels are displayed as either personal computer, such a system would cost
black (foliage) or white (openings), with a less than $8,000 US. It should be noted that
continuous tone black-and-white image and the personal computer could be made avail-
interactively changing the threshold up or able for other purposes when not analyzing
down until the edges of openings in the photographs, and the image analysis system
threshold image match the edges observed in can also be used for many other applications
the continuous tone image. Automatic calcu- (see Rich et al., 1989). The program
lations and data output are then initiated. SYLVA runs on almost identical hardware,
The process can then be repeated for the next so the costs are comparable (Becker et al.,
photograph. From 10 to 15 photographs can 1988). The only difference is that SYLVA
be analyzed per hour, including setup, currently uses a Chorus Data Systems PC-
threshold selection, calculation and data out- EYE PC-1l00 6-bit digitizer (approximate
put. cost $700) and a Tecmar GRAPHICS MAS-
TER display adapter (approximate cost $700)
Calculations by CANOPY involve, first, ex- and two black-and-white display monitors in
amining each pixel within the image to de- place of the single color monitor (approxi-
termine whether it is counted as foliage or mate cost $600 per monitor). SOLARCALC
21
PAUL M. RICH
22
CANOPY STRUCTURE: HEMISPHERICAL PHOTOGRAPHY
space and therefore best integrated over time, Stringent protocols must be adopted to
for example, by long-term monitoring with a minimize compounding of errors. The
light sensor attached to a data logger. For means for distinguishing foliage from canopy
this reason, light sensor techniques are lim- openings presents a considerable challenge.
ited in the number of field locations that can A simple threshold technique involves binary
be sampled, both because of the expense and classification of pixel intensity values rela-
effort involved; whereas large sample sizes tive to a specified threshold value. For ex-
can be rapidly obtained with hemispherical ample, in positive digital images of hemi-
photographs. Both techniques can require spherical photographs, pixel values above the
considerable efforts for data analysis, though threshold are classified as openings and val-
hemispherical photography is more labor in- ues below the threshold are classified as foli-
tensive. Field equipment tends to be more age. Similarly, in negative images pixel val-
expensive for light sensor measurements; ues above the threshold are classified as foli-
whereas analysis equipment can be more ex- age and values below the threshold are classi-
pensive for hemispherical photography. The fied as openings. Problems arise because of
techniques should not be viewed as exclusive uneven exposure, uneven reflectance within
of each other, but rather as complementary. foliage and background, penumbral effects,
Light sensor measurements can be used to and edge sampling effects. Penumbral ef-
obtain basic site-specific radiation flux data fects cause intermediate reflectance at edges.
for calibration of hemispherical photographs. Edge sampling effects occur because edge
pixels include both foliage and opening, and
Table 3. Comparison of hemispherical photography therefore appear as an intermediate gray
and light sensor techniques. value. When working with photographs
HEMI- LIGHT taken looking upward, a simple threshold can
SPHERICAL SENSOR mistakenly classify reflections on leaves as
PHOTOGRAPHY openings and dark sky regions as foliage.
Measurements indirect direct
Field Data Acqui- low - moderate high
Also, large openings tend to be over-
sition Time represented and small openings under-
Sample Size large small represented. Even backlighting of the sky
Possible and reduced reflections can be achieved by
Sensitivity/ high? high taking photographs on overcast days; al-
Reproducibility
Cost of Field $300-1,500 $1,000-
though this severely limits field time. Alter-
Equipment $2,000 natively, photographs can generally be taken
Cost of Analysis $3,000-$15,000 $2,000- early or late in the day to prevent overex-
Equipment (includ- $5,000 posed regions around the sun; however,
ing microcom- problems still arise because the antisolar dark
puter)
Analysis Effort moderate - high moderate
spot of the clear sky is near the zenith. Simi-
- high larly, photographs taken looking downward
have problems of shadows and contrast be-
tween foliage and background. Infrared pho-
B. Limitations and Problems tography greatly enhances this contrast. Ju-
dicious editing of images can be helpful, for
Hemispherical photography involves many instance to remove reflections from leaves.
steps. Errors can occur during image acqui- Further work is needed to explore alternative
sition, digitization, and analysis. Table 4 means for classifying images, including use
lists steps at which errors can be introduced. of convolutions and Fourier transform tech-
23
PAUL M. RICH
24
CANOPY STRUCTURE: HEMISPHERICAL PHOTOGRAPHY
hancements could extend the utility of the crue significant errors at each step, including
technique for study of both within short errors in camera positioning and orientation,
canopies and at higher positions within tall exposure, registration during digitization, and
canopies. Eventually, active close-range re- selection of a threshold. Use of a threshold
mote sensing techniques using scanning la- to distinguish foliage from openings contrib-
sers could allow greater confidence in locat- utes subjectivity, over-represents large open-
ing canopy openings. ings, and under-represents small openings,
especially for poorly exposed images. For
evenness of lighting conditions, the tech-
V. SUMMARY nique is best used on overcast days, limiting
the time when field work may be practical.
1. Hemispherical (fisheye) canopy photog- Also, calculation assumptions, of radiation
raphy involves determination of the distribu- flux distributions or leaf distributions for in-
tion of canopy openings using photographs version, can lead to errors.
taken through an extreme wide-angle lens,
either looking upward from within a canopy 5. There is a need-to establish standard pro-
or looking downward from above a canopy. tocols for image acquisition and analysis, so
Hemispherical photographs can be used to results between sites can be compared. Fur-
estimate light penetration into canopies and ther work is needed to assess the sensitivity
to assess canopy architecture. Hemispherical and utility of the technique. To minimize
photography is an excellent means for ob- compounding of errors and increase flexibil-
taining gap fraction data to use as input for ity, it is desirable to develop a hemispherical
inversion models. field instrument that integrates image acqui-
sition, digitization, and analysis capabilities.
2. Recently developed digital image analysis It is also desirable to enhance the technique
techniques provide a means for rapid analysis for study both within short canopies and
of large numbers of photographs. A digital higher positions within tall canopies.
image analysis system can be assembled as a
moderate priced add-on to an existing per-
sonal computer system. In particular, video
image analysis systems allow input directly
from negatives, interactive image registra-
tion, interactive threshold determination to ____________________________________
distinguish foliage from canopy openings, Acknowledgements: This research was sup-
rapid calculation, and automated data output. ported by National Science Foundation
Grants #B5R85-16371, #BSR86-14769, and
3. The primary strength of hemispherical #BSR89-l 8185, and the Los Alamos Na-
photography is that it provides a means for tional Laboratory Environmental Research
direct measurement of canopy geometry. Park. NASA/NSTL provided support for
Large sample sizes can rapidly be collected constructing light maps at La Selva Biologi-
in a non-destructive manner. The photo- cal Station. I thank Christopher Field, Mar-
graphs provide a permanent archival record cel Fuchs, John Norman, and Robert Pearcy
of canopy architecture. for their editorial and scientific assistance.
___________________________________
4. The technique has several notable limita-
tions. It is a multi-step process that can ac-
25
PAUL M. RICH
26
CANOPY STRUCTURE: HEMISPHERICAL PHOTOGRAPHY
27