MODELING AND VISUALIZATION OF CLOUDS FROM REAL WORLD DATA
A. Roditakis
ETHZ, Federal Institute of Technology Zurich, CH 8093 Hoenggerberg Zurich, Switzerland
(roditak@geod.baug.ethz.ch)
Commission V, WG V/6
KEY WORDS: Photogrammetry, Meteorology, Modeling, Visualization, Graphics, Hardware, Rendering, Point Cloud.
ABSTRACT:
Visualization of gaseous phenomena, such as smoke, fire or clouds has been a challenging issue in computer graphics due to the
complicated geometry of such natural objects. In the early 80's, approaches that faced this problem tried to simplify the
representation of such geometry, using simple particles or ellipsoids. During the last decade work was done on rendering fractal
volumes and random textures in order to create a realistic turbulent image of random generated volumes, and in parallel, modeling of
soft or 'blobby' objects (also known as 'metaballs") was combined with hardware accelerated volume rendering and gave some first
convincing images of 3D gas volumes in near realtime. There has been large technological developments in the computer graphics
hardware and a lot of work has been done from the computer graphics community on simulation of clouds, still there is not much
done in the direction of modeling and rendering of such objects from real world measurements, and work is lacking on the issues that
arise from datasets with incomplete spatial ot temporal resolution.
In the presented work we use the modeling technique of metaballs based on cloud bottom and top height measurements taken during
the CLOUDMAP2 project, in order to construct a 3D volume from point clouds. We present the problems that arise from fitting the
volume to the measured points, combination of cloud top and bottom height estimations with different resolutions and the generation
of large volume data. Solutions to these problems are presented which include the calculation of cloud volumes, interpolation with
cloud top height estimations from simultaneous ground based and satellite observations and finally we present various methods to
render such volumes using hardware and software assisted techniques.
1. INTRODUCTION
1.1 Motivation and aim of work
The work presented in this paper, was conducted in the frame
of a EU project called CLOUDMAP2. It targeted on the
parameterization of macro- and micro-physical properties of
clouds through the combination of space-bourne and ground
based sensors. This fusion of sensors helps to calibrate and
validate satellite products that could provide global coverage
for the cloud location and distribution, and thus help us
understand better the role of clouds in the interaction between
Earth's climate and anthropogenic inputs. The final aim of our
work was to create a processing chain that would take the cloud
top and bottom heights estimations as input, and deliver a three
dimensional visualization of cloud fields, that would provide us
with an instant image of the cloud shape and distribution over
measured areas. Furthermore we want to underline the
difficulties that come from using real world data for cloud
modelling and rendering, where we witness blunders, errors
and areas with missing values. These problems do not appear in
case studies for cloud rendering, since we notice that for the
demonstration of new modelling and rendering techniques,
authors use artificially generated data with uniform resolution
and coverage (perfect data).
1.2 Overview
Within the EU project Cloudmap2, several measurement
campaigns were conducted in the area of Switzerland, where
simultaneous measurements from ground digital cameras and
satellite sensors were conducted, in order to estimate cloud
bottom and top heights and winds. From the bottom side, a
configuration of digital cameras provides stereo cloud bottom
heights photogrammetrically, while at the same time a satellite
passes over the area aqcuiring images, which provide us with
stereo top heights. As the focus of the present work lies on the
modelling and visualization of these measurements into clouds,
details on the methods used in the aqcuisition step can be found
in the published work of other participants of the
CLOUDMAP2 project (Seiz, 2002), (Mayer et. al, 2003).
Starting from the cloud bottom heights we cleaned the dataset
from points that did not belong to clouds, and as a next step we
experimented with modelling the cloud bottom height, using
polygonal modelling and texturing to simulate the appearrance
of clouds. Similar work had been done by (Lee, 1996) and we
also cocnluded that this method presents limited options for 3D
cloud fields. We continued on the path of previous work done
on visualization of fluids and applied some basic volume .
rendering techniques, In the modeling aspect we tried two
different options for a definition of a 3D cloud field, the first
being a straightforward 3D regular grid calculation, and the
second influenced by the work from Nishita and others
(Nishita, 1996) on soft body modelling using 'meta-balls', also
known as 'soft-balls. These two methods interpolated the
distinct points over a 3D grid, with the second having the
advantage of smooth blending of neighbouring points.
Rendering of this three-dimensional cloud field was also
performed by utilizing two different techniques in parallel. We
tested the well known method for visualizing three-dimensional
iso-surfaces called 'marching cubes', which provided us with
rendering of a surface of the cloud field, with a chosen 1S0-
parametric value, and in parallel we implemented a hardware
accelerated volume rendering technique, using OpenGL
language and textured two-dimensional slices of the volume.
Finally we made use of an existing software library, used at the
National Center for Atmospheric Research, at Colorado, USA
(Lacroute, 1994), (NCAR URL) that although doesn't deliver
real-time rendering of the volume field, its high performance in
combination with the realistic lighting calculations, gave until
now the best results as far as visual quality is concerned. In
order to demosntrate the difference of an extrapolated 3D cloud
pe N) HH az ped pet (TY (05
(s
us iua died een i Ae SURE
