il 2004
————
canopy
lus the
ference
canopy
canopy
; of the
on from
gm ET
quently,
LO:
#q.13a)
opy [kg
s small,
es. TR,
Eq13b)
py was
on the
omplex;
canopy
| Konijn
veen 1.0
'M is set
defined
suggest
(Eq.14)
R o the
'rence in
(Eq.15)
tion and
urements
> that the
1e water-
n actual-
all yield-
scarcity,
diseases,
s a well-
pest and
> damage
ioning of
tion. The
1 ‘actual-
available
1pounded
ig of the
| (Eq.16)
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
Rugege and Driessen (2002) demonstrated that based on the same
production function indeed highly accurate estimates of maize
yield can be obtained. This has promise for regional applications
since it greatly reduces the computational data needs with few
other forcing variables needed then the difference between
ambient air and surface (or canopy) temperature. This is also the
weakness of the approach that the uncertainty of the estimates
inereases when AT is not available for all days in the crop cycle,
as often pixels are not entirely cloud-free at the moment of the
satellite pass.
Note that for computations of AT the following is needed: one
observer to record ambient air (7,) and another for surface (or
canopy) temperature (7,). Key to accurately quantifying the
thermodynamic process of transfer of energy from objects that are
warmer than their surroundings to the air, or from the air to cooler
objects, is that the absolute difference between the two are
observed at the same instantaneous moment, and using observers
that yield independent readings. Currently, ambient (air)
temperature is often taken from meteorological surface
observations. After these surface (point) observations have
undergone objective analysis, their spatial and temporal
resolutions have miraculously become commensurate with the
surface (or skin) temperature observation from satellites. Diurnal
variations of temperature (differences) can be highly dynamic,
and in such cases the resulting ¢fH20 or any other derivation of
the latent heat flux merely reflects the (lack of) quality of the
interpolation technique being deployed rather than approximating
the absolute transference of heat, inter alia crop stress. Even
though acknowledged, for this research it is assumed that errors
caused by this flaw can be neglected, and that positive differences
between 7, and 7, can be fully subscribed to canopy heating, and
thus crop stress.
2.2.1 Observing crop canopy heating from Space
A formidable challenge lies in combined use of data from
multiple satellites of complementary specifications to further
satisfy the spatial, temporal and radiometric requirements for
canopy temperature inference. Estimating ambient air temperature
from satellite data will possibly be dealt with in future
experiments; this study focuses on estimating canopy temperature
only.
22. Multi-sensor canopy temperature retrieval:
Accurate retrieval of surface temperature is complicated if
measurements are made by sensors aboard satellite platforms
far from the ground. Atmospheric attenuation processes
including absorption, upward atmospheric radiance and bi-
directional reflection of downward atmospheric radiance affect
transmission of the emitted radiation. Absorption of water
vapour is considered to be the most important factor influencing
radiance transfer in the thermal spectral range (Bastiaanssen,
1995; Qin and Karnieli, 1999).
Polar orbiting satellites have a relatively high signal-to-noise
ratio, and depending on the flight characteristics of the spacecraft
in question they cover the same spot of the earth surface once
every day. Coupled with cloudy conditions very few clear-sky
observations remain for repetitive skin temperature retrieval for a
fixed position on the earth’s sphere. For skin (or canopy)
temperature inference the so-called ‘split-window’ technique is
commonly applied for data from multi-thermal band sensors. The
technique eliminates effects of water vapour absorption and
emission by using split data in the 10 to 13 pm range, often
referred to as the 7,, and T,» bands. The concept exploits the
different absorption characteristics of the atmosphere within these
215
different but close wavelengths, assuming that surface emissivity
is constant over this spectral region. Detailed reviews of split-
window algorithms are provided by Caselles et al (1997), Qin and
Karnieli (1999) and by Parodi (2000).
Sensors aboard satellites with a geosynchronous orbit observe
diurnal changes of the atmosphere and earth surface for a fixed
region, but do so at a lower signal-to-noise ratio due to their
height (approx. 30.000 km). Another disadvantage is, despite their
attractive temporal resolution, that they observe the earth surface
at (nadir) resolutions of >5 km for TIR bands, actual resolutions
depending on the sensor in question and distance from at nadir.
Such spatial resolutions are insufficient for most regions, which
have scattered land use practices resulting in mixed-pixel
observations. Data from more *recent sensors with smaller at
nadir resolutions (<3.25 km) may further help to overcome this
drawback and permit to up scale the developed methodology to
regions with less homogenous land cover. Most sensors aboard
geo-stationary satellites also have two channels in the thermal
infrared range of the spectrum (10 to 13 jum). Researchers have
effectively exploited this to minimize the errors in estimating land
surface (or canopy) temperature, analogue to the technique
developed for polar orbiting satellites data as introduced above. In
an attempt to further improve the technique for geo-stationary
satellites. Sun and Pinker (2002) showed that by adding a second
term of the brightness temperature difference (T, rT, the
atmospheric effect can be further removed. They also noted that
when the satellite viewing angle increases, the optical path and
the atmospheric attenuation increase also. After McClain et al.
(1985) they added a zenith angle correction term (sec@/) to
further normalize the data for optical path variations. Yuichiroh
(2004) further elaborated on this idea, and signalled that the effect
of water vapor is not fully removed by the split-window
technique. After Coll and Caselles (1997) he attempted to further
improve the algorithm by calculating various coefficients for
different precipitable water levels using a radiative transfer model
over The Tibetan Plateau for the application of the method to
satellite data from GMS-3.
For this study satellite data from NOAA-/4 and GMS-5 were used
because of the similar specifications of the two instruments
onboard these satellites and their complementary viewing
frequency. The NOAA-14 spacecraft passes at approx. 14.00 h
local time (range: 13.00 — 15.00 h), whereas GMS-5 scans the
whole of South East Asia every hour.
2.2.3 Inter-calibration between GMS-5 and NOAA-14:
Instrument calibrated data from GMS-5 have been evaluated
against calibrated data from the polar orbiting satellites NOAA-14
by the Japanese Meteorological Satellite Center, JMA (Tokuno
M., 1997). The results of their evaluation revealed that the
brightness temperatures of /R/ (10.5-11.5 um) of GMS-5 are
about 1.2 (K) lower than those of Ch4 (10.3-11.3 um) of NOAA-
14/AVHRR on average, and the brightness temperatures of /R2
(11.5-12.5 um) of GMS-5/VISSR are about 0.6 (K) higher than
those of Ch5 (11.5-12.5 um) NOAA-14/AVHRR.
* SEVIRI (Spinning Enhanced Visible and Infra Red Imager) instrument
aboard the European geo-stationary MSG-/ and 2 and the spectral
channels of the Chinese Visible and Infrared Spin Scan Radiometer
(VISSR) aboard the geo-stationary FY-2C will enhance upon the
applicability of the procedures presented in this research by their
improved temporal and spatial resolution.
