of
the
sets
tral
and
and
find
and
n of
late
ired
nal
rite,
itor
J.S.
lon,
129,
| of
Petrosyan, Arakel
THE NEW PHOTOGRAMMETRIC METHOD FOR CLOUD REMOTE STUDIES
S. Pavlov, A. Petrosyan
Space Research Institute of the Russian Academy of Sciences, Russia
apetrosy@iki.rssi.ru
KEY WORDS: Clouds, Entrainment, Image Matching, Photogrammetry , Remote Sensing
ABSTRACT
The main objective of the suggested work is to improve predictions of cloud top entrainment phenomena with
photogrammetric methods based on satellite data analyses. The new analysis methodology for the cloud top 2D-flow
reconstruction from remote sensing data is developed and suggestions for further satellite missions and data sets is
discussed. Validation examples based on available data is provided. Method uses a set of images of the cloud layer with
known temporal and spatial resolution in order to obtain flow pattern. We consider visual movement and deformation of the
picture as an atmospheric flow perturbation and in the simplest situation imply it to be identical. The identity of two points
is determined by their neighboring vicinity. Thus the method, analyzing some vicinity of point at the first image, is to find
an identical vicinity at the second one. Algorithm analyses the couple of consecutive In that way we build an auxiliary
space in which small displacement of original image point will be in correspondence with that of auxiliary space point.
Also two points with equal in the above sense vicinities have the same image point in the space. The solution is developed
by use of iterative method.
1 INTRODUCTION
The main objective of the suggested work is to improve predictions of cloud top entrinment phenomena with
photogrammetric methods based on satellite data analyses. The new analysis methodology for the cloud top 2D-flow
reconstruction from remote sensing data is developed and suggestions for further satellite missions and data sets is
discussed. Method uses a set of images of the cloud layer with known temporal and spatial resolution in order to obtain flow
pattern. We consider visual movement and deformation of the picture as an atmospheric flow perturbation and in the
simplest situation in order to obtain flow pattern. We consider visual movement and deformation of the picture as an
atmospheric flow perturbation and in the simplest situation imply it to be identical. The identity of two points is determined
by their neighboring vicinity. Thus the method, analyzing some vicinity of point at the first image, is to find an identical
vicinity at the second one. It is natural that we can not find the exact transformation because of liquid particle deformation.
However, assuming that the above deformation is small enough we can disregard such a problem. Algorithm analyses the
couple of consecutive monochromic images in some spectral range and releases the field of deformation which translate the
first image into the second one. The field of deformation in hand is developed as a solution of a variation problem. In that
way we build an auxiliary space in which small displacement of original image point will be in correspondence with that of
auxiliary space point. Also two points with equal in the above sense vicinities have the same image point in the space. The
solution is developed by use of iterative method. A tangent hyperplane is considered and found is the optimum directions of
movement toward the solution. Further the iterative process is repeated, and as we can control both the movement of image
point and displacement itself, the error of iteration is not accumulated. Therefore the solution can be obtained with adequate
accuracy. Then it extracts regions free of clouds and calculates the velocity field. Numerical experiments show major
parameters affecting the solution to be the size of point vicinity and how thoroughly it is described, that can be determined
by number of reference points, and therefore by dimension of auxiliary space. It is interesting that both parameters have
their optimum values depending on image character. This technique is subjected to time resolution of input data, so there
are some restrictions for experiments methods. In analysis of cyclones in Earth atmosphere with the scale about thousand
kilometers it requires temporal resolution about some minutes while at the present time such resolution is about hour
depending on satellite orbit. The further possible development of the method may follow several directions, such as to take
into consideration the vertical movement of the cloud layer using stereoimages, to improve trajectory search algorithm in
case when exact coincidence is impossible, to automate evaluation of optimum algorithm parameters in accordance with
statistical properties of images being aligned.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 1149