Environmental modeling using graphical processing unit with CUDA

Modeling transport and deposition processes of toxic materials in the atmosphere is one of the most challenging environmental tasks. These numerical simulations with dispersion models are very time consuming, therefore, their acceleration is extremely important. One possible, effective solution for increasing the computational time can be the parallelization of the source codes. At the same time, the technological improvement of graphics hardware created a possibility to use desktop video cards to solve numerically intensive tasks. In this study, we present a new and powerful parallel computing structure for solving different environmental model simulations using the graphics processing units (GPUs) with CUDA (compute unified device architecture). Two different types of dispersion models were developed and applied based on this technology: a stochastic Lagrangian particle model and an Eulerian model. We present and discuss the results and advantages of both methods. A Lagrangian particle model was applied to simulate the transport of radioactive pollutants from a point source after a hypothetical accidental release at local scale. In addition, an Eulerian model was used to simulate sulfur dioxide transport and transformation in the troposphere at regional scale. Moreover, in both cases, CPU and GPU computational times were also compared. We can achieve typical acceleration values in the order of 80-120 times in case of Lagrangian model and 30-40 times in case of Eulerian model using this new parallel computational framework compared to CPU using a single-threaded implementation. Next to the obvious advantages, the barriers of this new method are also discussed in this study.