Among the many steps in every processing chain for optical satellite imagery, cloud detection is a vital step.  Clouds obstruct the viewing of the earth's surface, as well as part of the atmosphere, and it should be clear that for imagery targetted towards studying land cover or atmospheric composition, they need to be properly detected to be able to take them into account for further processing of the data.  Due to differences in the characteristics of the sensor, (the number and wavelength of the different spectral bands, the bandwidth,…), probably no two cloud detection schemes for different satellites can be the same.  In this paper the cloud detection scheme for the VEGETATION sensor aboard the SPOT4 satellite, launched March 1998 on the Ariane 4 carrier is discussed.
Detection of clouds is a classification problem.  Many ways exist to solve classification problems, but the boundary conditions in which the eventual algorithm has to operate are very restrictive.  To operate in near real time conditions, over 250 million pixels, measured in 4 spectral bands, have to be classified daily, which eliminates certain methods involving spatial pattern recognition such as self organising maps or segmentation algorithms.  This paper presents a thresholding approach, which uses a genetic algorithm to determine the optimal selection of variables derived from the information delivered by the sensor, as well as to determine the optimal thresholds for cloud and snow detection.

GIL LISSENS
TAP, Vito
Boeretang 200
Mol, B-2400, Belgium
lissensg@vito.be

Aircraft, portable tower and radiosonde measurements from August 4 1989 (Day 68) of the First International Satellite Land Surface Climatology Project (ISLCP) Field Experiment (FIFE), over fairly flat terrain in Kansas, are used for the reconstruction of the heat conservation equation. The calculation grid consists of three grouped flight levels and the surface, and three hour-blocks, within the midday convective boundary layer (CBL). The day chosen had a weak warm front disturbance in addition to the usual summertime southern flow over the Midwestern United States. Significant vertical and temporal structure which should not be bulk averaged was observed for all terms. The upper layer showed warming sustained at a constant rate around noon and appeared decoupled from the surface and the middle levels in the later hours. Excess warming was due to advection and possibly entrainment. Consequently, the heat flux divergence also had a non zero vertical gradient. Advection proves to be an important term and does not average out when driven by a synoptic feature. Its East-West component improved the balance of the heat equation at all levels. Partial flight inter-track and total CBL volume standard deviations for terms and residuals are carried through explicitly.

JOHN C. LUKAS
Complex Systems Center
Vermeer Goldmann Institute, Hellas
140 Tritonos Str., Paleo Faliro
Athens 17562, Greece
johnl@compulink.gr