You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Remote sensing for resources development and environmental management
Damen, M. C. J.

3-3- Cloud detection techniques
One of the major problems in satellite retrievals
of Earth-surface parameters is the correct
identification and removal of cloud contaminated
Currently available techniques (Saunders, 1985) are
only partially automatic in nature and include an
essential interactive stage, leaving the decision
on cloud clearance to an experienced operator.
However, to cope with the ever increasing stream of
remotely sensed environmental data, we decided to
embark on a fully automated cloud identification
Algorithms for temperature retrievals require
pixels to be flagged cloud contaminated even with
cloud covers as low as 1$, the error in the derived
brightness temperature having increased to the 0.2
K level and higher already.
During the course of our developments we found that
the usual cloud detection techniques can be used in
an automatic scheme if semi-transparent clouds can
be identified by a new procedure. In this context
we developped the "channel 4 - channel 5" -
The brightness temperature difference between
channels 4 and 5 can be used to detect semi
transparent cloud-layers, especially thin cirrus.
This type of cloud, though frequently occuring, is
hard to identify with other cloud detection
techniques, because the radiance of thin cirrus in
both visible and infra-red channels is very low.
The small difference in effective emissivity
between channels 4 and 5 can cause brightness
temperature difference (T^ - T b c) between the two
channels as high as 6 K for semi-transparent cloud
layers with an effective emissivity of 60$ (Inoue,
Thresholds, linearly related to T b ^ for T b c are
"empirically" determined. These have been obtained
by analyzing a large number of AVHRR-images of
different areas, seasons and times of the day. By
selecting obviously cloudfree and obviously cloudy
pixels and plotting the reswults in a 2-D histogram
(T b c versus T b ^ - T b 5) thresholds for land and
sea’nave been determined (see fig. 1):
-> T b , 5
Figure 1: 2-D histograph showing
principle of "ch4 - ch5" -
sea:cloudfree if T b ^ - T b 5 < 0.065 * T b 17.255
land:cloudfree if T b ^ - T b 5 < 0.094 * T b 5 -25.11
(brightness temperatures in’degrees kelvin)
Land/sea discrimination is based on a linear
combination of the channel 1 and channel 2 albedos.
3.4. Navigation/geometric correction
Navigation of satellite images involves assigning a
latitude and longtitude to any point in the
satellite image. For accurate navigation results,
the exact position of the satellite must be known
at any time, simple trigonometry then giving the
latitude and longitude for every pixel (in
The NOAA-information Service publishes every other
day orbital elements of the operational NOAA-
satellites. These elements can be used together
with the recording date of a scanline to determine
the coordinates of the sub-satellite-point in this
line and after that the positions of the pixels on
the scanline. In this way a navigational grid can
be generated with which an image can be resampled
in any desired map-projection.
The brighness temperature in the three (or two)
thermal-IR channels of the AVHRR are being used to
derive true Earth-surface temperatures. The
different characteristics of sea and land surface
require different approaches.
4.1. Sea surface temperature (SST)
4.1.1. Multi channel technique
Many uncertainties hamper the measurements of the
Earth-surface temperature from space. Nevertheless
for a sea-surface some of these can be eliminated.
In the thermal IR wavelengths the emissivity of sea
water is high and relatively constant, radiometric
efficiency is particularly high on account of the
nature of the Planck function at temperatures near
300 K and verification with in-situ measurements is
possible. In these circumstances by far the largest
source of measurement error remains in the
estimation of the atmospheric correction. The
wavelengths of the AVHRR thermal infra-red channels
have been chosen in such a way that atmospheric
effects are different for each channel. The
transmittance in these so called "window" is
dependent not only on the water vapour
concentration but also on its vertical
The SST can be obtained from a linear combination
of the brightness temperatures:
SST = a ♦ .? a. T,., where N is the number of
. .0 i=l 1 Ai
channels used.
The coefficients a Q , a i are determined, empirically
or theoretically, to give the optimum performance
in a given set of atmospheric conditions believed
to joinly represent those in a particular area, or
period, or for the whole globe. The correct
determination of the coefficients is critical for
the accurate measurements of SST from satellites.
A good set of coefficients for the North East
Atlantic Ocean and North Sea was determined by a
group of the Rutherford Appleton Laboratory in the
UK. (Llewellyn-Jones, 1984). The coefficents are
scan-angle dependent. During daytime the "split"-
window (channels 4 and 5) technique must be used,
because the reflection of solar energy in channel 3
is too high. During nighttime a "triple" window
technique gives more accurate results provided
channel 3 is not too noisy.
4.1.2. Verification
In October 1985 the Oceanographic Research

fig. 2).
derived ;
The in-s:
15.00 GM'
The SST':
in-situ t
the in-s:
by the s<
R ut her f 01
than the
between j
SST is m
RMS 0.241
is 0.022
These st£
4.2. Lane
4.2.1. Te
For the c
for the £
1 . the
2. Lack
Price (19
window te
can also
He estima
LST's , us
caused by
4.2.2. Va
In co-ope