Improving BettAir Maps

Presentation of the problem and objective of the experiment

Nowadays, the distribution of pollutants at street and urban level is not completely understood because the sources of the emissions of the gas concentrations may change fast at a given location and between nearby sites. In this context, HPC and Computational Fluid Dynamics (CFD) are key tools for tracking the dispersion of pollutants with high resolution. The goal of this experiment is to train Generative Adversarial Networks that mimic the output of HPC-CFD simulations at an affordable cost and to add them to Bettair’s map generation pipeline.   

Short description of the experiment

A dataset of thousands of real urban configurations (building shapes and street geometry) will be created along with its micro-meteorology and dispersion patterns using CFD over HPC. These geometries will be sampled from different European cities in order to have a heterogeneous dataset. Then, generative models will be built from the CFD results for micro-meteorological variables and for the spatial distribution of pollutants, capable of generating the distribution of the dispersion variables in urban geometries. These models will compress the information contained in the CFD simulations avoiding the need for HPC, enabling the generation of affordable high-resolution heat maps, once integrated into the map generator service.

Organisations involved:

HPC Provider: Barcelona Supercomputing Center