Sujet de Master 2008-2009
Simulating
an evolving cloudy sky
/ Simuler
un ciel nuageux évolutif
Responsible
Context
Weather
simulations (low resolution) as well as physical simulations of
isolated clouds are very expensive, and still, approximate and unable
to produce detailed realistic skies. In the context of
graphics applications (video games, simulators,
google earth -like exploration), one also wants real-time performance,
exploration of wide areas, but also control (i.e., predictability)
of the result by the graphist !
Fortunately,
the cloud
physics is not limited to the numerical solving of differential
equations
(Navier-Stokes): aerodroms and pilots
have simple methods for evaluating
the key parameters
(instability, cloud base and top height). Moreover, several
stereotypical
"patterns" rule phenomena (frontology, inversions, aerothermy due to
topography and soil type, ribbons and convective cells) ... Conversely, there
exist many CG methods able to combine
computer simulation and coarse procedural enrichment, and to generate
or
regenerate on the fly parameters required for a portion of landscape.
Thus,
past or ongoing works in our lab concern the coupling of models of
different
scales, the procedural
simulation of fluids, the real-time
exploration of vast landscapes, rendering real-time
clouds.
Description
The
aim of this study is to simulate plausibly and in
real-time a cloudy sky on a landscape portion (depending on the
exploration), seen from ground or above (as a first step we can focus
on cumuluniform clouds seen from not far). To
be able to regenerate any portion of the sky on demand and to adapt to
the needed
resolution, the idea is to couple coarse physical simulation and
proceduralism, and
to adopt a multi-scale strategy. The ingredients are:
the atmospheric 3D environment (state, movements), the thermals
emitted from the ground which moisture creates clouds when condensing,
and
the soil environment which modulates the emission of thermals and air
movements.
- On
the proceduralism
side (ie, parameters known at any point and time for the simulation),
there will be the whole environment: patterns of atmospheric movements
at large
scale (wind,
fronts) and mid-scale (cells),
sun illumination, ground (modifying local movements of air and sun
exposure). One has to define
variables (humidity and heat vs cloud base and top height) and
functions (constrained noise, analytic functions ...) so as to
facilitate
the simulation and the controllability by the graphics user.
- In
addition, we would define a deterministic spatio-temporal function of
thermal creation
on the ground. This
will allow us to regenerate on demand any portion of landscape
consistently.
- On
the simulation side, the idea is to simulate the evolution
of thermals: Newton's laws and thermodynamics can determine
trajectory,
evaporation and condensation of these blob-particles. However, this
changes
the local conditions (temperature and humidity) for next thermals: so
the
3D result must be stored, as for a classical simulation on grid.
As
one cannot store this on a landscape-wide area, we need to know how
to quickly resimulate a
portion of landscape so as to reconstruct the state of the atmosphere
(and sky) while exploring.
- For
this to be achievable
with good performance, the idea is to simulate this mechanism only
at coarse scale, then to consider more detailed thermals for the
visualization, influenced by but not influencing the state of the
atmosphere.
- For
the smallest details, procedural noise techniques allow to "dress"
coarse objects with appearant details, which can even be done in real
time (this is not a priority during the time of this study).
Références
References
- See
clickable links throughout the text.