170
and Thomas, 1976; Jackson, 1983). The
reason Eq. [3] is location specific is
that efficiencies of conversion and
harvest indices are climate-dependent,
a fact not much acknowledged or
discussed yet in the crop science
literature.
The VI used can be calculated either
from reflectance factors (units of %)
or from digital counts (DC) which are
proportional to radiance
(W/m2/ym/sr). Because DC are
proportional to radiance, they will
change seasonally with sun zenith
angle, even if leaf area index or
amount of photosynthetically active
tissue is constant. Thus we speculate
that cumulations of VI based on DC will
take variations in incident PAR into
account, making comparisons of IVI
versus yield for a given sensor more
comparable among geographically
separated sites than if based on
reflectance factors. However, the
calibrations will differ among
sensors. On the other hand,
calibrations from observations
converted to exoatmospheric
reflectances should agree among
observation systems, ignoring
date-specific atmospheric effects on
the observations. For sites in
different parts of the world that are
climate analogs of each other, the
yield efficiencies in terms of VI as
well as the slopes of the right side
terms of Eqs. [3] and [5] should be
closely similar. The inverse should
also be true; for carefully taken data,
the difference in the slopes should be
a measure of the climatic effect if
yield potentials are genetically
alike. The hypotheses stated are
difficult to test because of
measurement errors in available data
sets, or incomplete data sets.
Because rainfed agricultural areas
dominate the production of crops
important in world trade and because VI
take time to respond to current
conditions, it would be highly
desirable to augment the VI with canopy
temperature observations which do
respond to current plant-available
water conditions. The thematic mapper
has a thermal band (10.4 to 12.5ym) but
the observations have to be corrected
for amount of water vapor in the
atmosphere and have not been used very
much in conjunction with the NIR and
Red bands; the ground resolution is
also larger than for the shorter
bands. Wiegand et al. (1983) have
reviewed drought detection and
quantification by reflectance and
thermal observations, but at that time
Eq. [5] and SCA did not exist.
Maas et al. (1989) have incorporated a
crop water stress index based on canopy
temperature into a crop simulation
model in which initial conditions and
parameter values are adjusted to make
simulated growth agree with remotely
sensed observations (Maas, 1988). In
the growth simulation model, daily dry
matter production is calculated from
daily PAR absorbed and an efficiency of
conversion, and during grain filling a
constant fraction is allocated to the
increase in grain dry mass. In
addition, the simulation model requires
weather data (ambient maximum and
minimum daily temperature, dewpoint
temperature, and solar radiation) that
is not always available. In contrast,
the pure spectral approach used here
assumes the same physiological
principles as the simulation model but
relates the spectral observations
directly to yield. Thus it is
simpler. The two approaches are best
viewed as complementary rather than
competing because when applied to
particular situations the yields
estimated should agree (in direction
and magnitude) with each other and
available ground truth and strengthen
confidence in both.
Another appealing feature of SCA is
that the functional relations of all
the left side terms of Eqs. [1], [2],
[3], and [5] are linear. This makes
them easy to use because slopes and
intercepts are easy to interpret and
have biophysical meaning. The
equations are linear because there is a
proportionality between photosynthetic
size of the canopies and each of the
dependent variables, APAR, DM, and
yield through the common process,
photosynthesis. Evapotranspiration is
also a function of the
photosynthetically active tissue and
insolation so that the relation
between ZET and ZVI is nearly linear.
In applying spectral components
analysis there are many pertinent
questions to ask in order to tailor the
required procedures to the specific
objectives of the application. These
include: Is information needed about
specific ecological communities (crops,
pastures, forests) or is synoptic
information sufficient? If information
is needed about specific ecological
communities, is enough known about the
annual calendars of the plant community
categories involved (planting and
green-up dates, relative rates that
ground cover develops or greening
occurs, growing season duration,
senescence rate or date differences) to
develop classifications algorithms that
distinguish the ecological categories
accurately enough to estimate the
hectarage of each? Are locations of
the perennial categories well enough
known that masks of the areas they
occupy can be prepared? Are indicator
fields or sites needed that can be
located repetitively, or will a random
sample from within the crop category or
the scene be adequate? What magnitude
of difference in VI or in yield is
