which assist in the interpretation process. 'The cloud classification process is sup
ported by radio sonde reference data (RSA).
The following assumptions have been made for the cloud classification process:
* clouds are black body emitters
e the instantaneous field of view of the radiometer is completely covered
by clouds (no mixed pixel effects').
Classification of scene segments is implemented after division of the image frame
into sub-segments for which the RSA-values (approx. 35) can be assumed repre
sentative. For this process the minimum distance on earth surface (not in the
image matrix) is calculated. The cloud upper limit is assumed to be at a height
where radiation temperatur and the surrounding temperature coincide. This is.
however, only valid for a completely transparent atmosphere. Under real condi
tions an atmospheric correction would have to be performed. This paper uses a
one- channel parametric correction scheme. In so doing the radiation and tem
perature deficit above the clouds (between cloud top and the satellite) as a result
of the rest band water vapour absorption in the large atmospheric window (IR-
channel) is simulated. This is done by integration of the Schwarzschild-Emden
differential radiation transfer equation for stationary, horizontally homogeneous
conditions and neglection of IR scattering.
The spectral transmission function of wafer vapour is obtained in parametric form
using effective spectral mass absorption coefficient from literature. The reduced
water vapour mass above the clouds is obtained with the help of radio sonde
measurements (humidity, temperature). This procedure reduces the temperature
of the main pressure levels by an. amount induced through atmospheric influence
under the condition of a black body emitter. The in this manner corrected elima
to logical and regional temperature tables are used to perform a threshold classifi
cation of the METEOS AT infrared image segments.
3. image compression techniques
3.1. A parallel two-dimensional procedure for micro-adaptive image
coding
Following the classification process a further reduction of the data volume can be
performed by redundancy reduction. The relevant algorithms can also be applied
to achieve compression of non-classified meteorological images such as NOAA
scenes. To achieve that goal a two-dimensional adaptive compression procedure
is applied. The images are decomposed using a quadtree structure. Image coding
is performed by explicit transmission of resolution and gray level, while the loca
tion of the pixels is implicitly given by the algorithm. Resolution is determined by
the local contrast of images.
Higher compression rates can be achieved by tolerating differences between the
original and the decoded image. The resolution is controlled in such a way that
small details are transmit ted if the local contrast, is high. This procedure avoids to
a great extend blurring of edges.
This procedure has been implemented as a parallel as well a sequential algorithm.
