About the project

The transport processes in the boundary layer of the atmosphere are very complicated due to the turbulent nature of air motion. It is the presence of cumulus clouds that makes these turbulent dynamics of the boundary layer even more complex. At the moment there is no consensus on which way particles (smoke, dust) are dispersed within and around a cumulus cloud. To understand more about the characteristics of a cumulus cloud during its entire lifecycle, the visualization of particles released within a cumulus cloud could lead to interesting discoveries.

Typical Cumulus Clouds

Particle tracing and visualization could give a qualitative comparison of direct measurements with the in-cloud thermodynamics and turbulence, as produced by Large-Eddy Simulations (LES). Large-Eddy Simulation is a computational method to simulate the dynamics of the Atmospheric Boundary Layer (ABL) and the dynamics of clouds formation. The results of numerical simulations such as LES or the measurements made of physical phenomena often consist of large datasets that represent velocity fields and other attributes such as position, temperature and moisture. Particle tracing can be an excellent method for the visualization of time-varying flow fields produced by measurements or simulations.

However, particle tracing can be a very computationally demanding task when one wants to visualize a large dataset of time-varying flow fields in real time. The rapidly increasing computational power of graphics hardware could make the visualization of many particles with interactive rates possible. Interactive rates are needed to manually release particles at a certain point in space and to immediately observe in real-time how these particles behave within a (time-dependent) flow field.

An implementation of an algorithm used for particle tracing on graphics hardware has shown to outperform a conventional CPU implementation considerably, given that certain bottlenecks have been overcome. One of the largest bottlenecks for a graphics hardware implementation is the transfer of data from system memory to the graphics hardware. During computation, the results of every pass should be stored in graphics memory, so that the computation can benefit from the very high bandwidth of the graphics hardware. Many high-level languages and algorithms for different purposes have been developed to exploit the parallelism and high bandwidth of the graphics hardware. These languages and algorithms make use of the ever growing programmability of the Graphics Processing Unit (GPU).

Particle tracing and other computational demanding, scientific visualization applications performed on graphics hardware are called "General Purpose computations on the Graphics Processing Unit" (GPGPU). GPGPU has become a larger field of study since the accelerated growth of computational power of graphics hardware, pushed for a great deal by the computer games industry.