MENTS
coutimi,
yorku.ca
Jean-sur-
ada -
nthesis and
tigations in
[ prediction
2001, 2002
c, Canada).
ability for a
orne CASI
or greenbelt
saturation at
aluation of
je measured
on variables
d absorbed
lanc, 2000;
Gitelson et
farmers are
lity within
ivity and to
IS precision
oped during
y goal is to
Itural fields
ites that are
deficiencies
y particular
several crop
imbalances,
an indicator
onse to soil
for given
idy were (1)
elationships
tral indices,
predictions,
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
and (iii) to validate the indices’ prediction capability using
CASI (Compact Airborne Spectrographic Imager) hyperspectral
images and LAI measurements collected over fields different
from those used to collect ground spectra.
2. MATERIAL AND METHODS
2.1 Ground Spectra and Corresponding LAI Values for
Prediction Equations
The study area is located near Montreal, at the Horticultural
Research and Development Centre of Agriculture and Agri-
Food Canada, St-Jean-sur-Richelieu, Quebec, Canada. It is
known as the L’Acadie Experimental Research Sub-station
where various crops are grown on different experimental fields.
Intensive field campaigns (IFCs) were organized during the
growing seasons of 2000, 2001, 2002, and 2003 in order to
collect ground spectra and corresponding LAI values as well as
crop growth measures. Over these four years, different crops
(corn, beans, and peas) were monitored on experimental and
commercial fields. Acquisition dates were planned to coincide
with different phenological development stages, aiming to
monitor temporal changes in crop biophysical variables such as
LAI, ground cover, and other growth measures. A particular
emphasis was placed on acquiring ground data covering the
earliest, middle and latest periods of the growth season.
Spectral reflectance data over the region 0.4 — 1.1 mm
wavelength region were acquired with the ASD
spectroradiometer (Analytical Spectral Device, Boulder, CO).
A white Spectralon reference panel (Labsphere, North Sutton,
NH) was used to calibrate the spectroradiometer spectral
radiance measurements to reflectance, by measurements of the
reference panel under the same illumination conditions as the
ground targets. Reflectances were calculated and corrected for
the non-ideal properties of the reference panel as described by
Robinson and Biehl (1979). In all experiments, radiometric data
were collected close to solar noon; thus, changes in illumination
conditions (solar zenith angle) were minimized.
Crop LAI values corresponding to measured spectra were
determined using both non-destructive and destructive methods
using the Plant Canopy Analyzer (Li-Cor model LAI-2000) and
an area meter (LI-3100, Li-Cor, Lincoln, NE), respectively. The
latter was used to measure the LAI at the early growth stage, as
well as to determine separately LAI of green and dead leaves
during the senescence stage.
2.2 CASI Hyperspectral Images and LAI Values for
Validation
The study area is located in Ottawa (Canada), at the former
Greenbelt Farm of Agriculture and Agri-Food Canada. Over
three successive “years, different crops (corn, wheat, soybean)
were grown on a 30-ha field with a drained clay loam soil as
well as on adjacent fields owned by farmers. The experiments
consisted of dividing the main field into four regions receiving
various nitrogen treatments: 100% of the recommended
fertilization over a flat region, 100% of recommended nitrogen
over a region with a gentle topographic slope, 60% of the
recommended rate, and no nitrogen application (0%). They
were thus laid out to promote development of remote sensing
techniques for detection of plant stresses in precision
agriculture, particularly stresses due to nitrogen deficiency,
water deficit, and topographic influence. Within each region, a
grid of georeferenced points spaced every 25 m was established
109
on a representative section of 150 m x 150 m. These locations
were used to monitor crop biophysical parameters during the
growing season, particularly during intensive field campaigns
coinciding with image acquisition. Details on the experimental
site and design are presented in Pattey et al. (2001).
Hyperspectral images were acquired by the Compact
Airborne Spectrographic Imager (CASI), flown by Centre for
Research in Earth and Space Technology (CRESTech).
Simultaneously, a set of field and laboratory data were
collected for biochemical and geochemical analysis, along with
optical and biophysical measurements. Ground truth
measurements included: (i) collection of leaf tissue for
laboratory determination of leaf chlorophyll concentration, leaf
area index (LAI) measurements using the Plant Canopy
Analyzer (Li-Cor model LAI-2000), (ii) an area meter (LI-
3100, Li-Cor, Lincoln, NE), and (iii) crop growth measures.
During 2000 and 2001 growing seasons, CASI
hyperspectral images were collected in three different
deployments, using two modes of operation: the multispectral
mode, with 1 m spatial resolution and 7 spectral bands selected
for sensing vegetation properties (489.5, 555.0, 624.6, 681.4,
706.1, 742.3, and 776.7 nm); and the hyperspectral mode, with
2 m spatial resolution and 72 channels covering the visible and
near infrared portions of the solar spectrum from 408 to 947 nm
with a bandwidth of 7.5 nm. Acquisition dates were planned to
coincide with different phenological development stages,
providing image data covering the earliest, middle and latest
periods of the growth season.
The hyperspectral digital images collected by CASI were
processed to at-sensor radiance using calibration coefficients
determined in the laboratory by CRESTech (Centre for
Research in Earth and Space Technology). Subsequently the
CAMSS atmospheric correction model (O’Neill et al., 1997)
was used to transform the relative at-sensor radiance to absolute
ground-reflectance. To perform this operation, an estimate of
aerosol optical depth at 550 nm was derived from ground sun-
photometer measurements. Data regarding geographic position,
illumination and viewing geometry as well as ground and
sensor altitudes were derived both from aircraft navigation data
records and ground GPS measurements.
Reflectance curves derived from processed CASI images
showed the presence of spectral anomalies associated with
atmospheric absorption features at specific wavelengths.
Although we applied model-based atmospheric corrections, the
calculated reflectances are still affected by spectrally-specific
errors owing mostly to an under-correction of some
atmospheric components effects (oxygen and water vapour
absorption). The flat field calibration is a correction technique
used to remove the residual atmospheric effects from
hyperspectral reflectance image cubes. Its aim is to improve
overall quality of spectra and provide apparent reflectance data
that can be compared with laboratory spectra (Boardman and
Huntington, 1996). It requires the presence, and identification,
in images of spectrally-flat uniform areas where the spectral
anomalies can be unambiguously attributed, in narrow spectral
ranges, to atmospheric effects and the solar spectrum. In CASI
images, these features were observed over asphalt and concrete
areas within the same image where the reflectance spectra are
assumed to be flat or nearly flat over these features. Using
signatures of such scene elements, we calculated coefficients
that adequately compensate effects of atmospheric water and
oxygen absorption. After those coefficients were applied to the
entire image, but only in the specific spectral ranges affected,
we checked the signatures of different components of the image
and found that observed residual features have been
successfully removed.
