4 Aug 2021Faster results using spatial splitting: modelling the West Midlands with ADMS-Urban

The spatial splitting option introduced in ADMS-Urban version 5.0 has been used to speed up run times for modelling of the UK West Midlands region. Both turnaround time and total CPU time for modelling large regions with ADMS-Urban can be significantly reduced by judicious use of spatial splitting in combination with task farming.

The effect on run time and computational expense when using spatial splitting over increasing numbers of cores is investigated in a newly published paper co-authored by CERC researchers, and shows substantial reductions in both metrics compared to a single processor run. The reduced run times allowed the production of high-resolution contour plots and averages over different spatial areas and temporal subsets, which may be relevant for exposure and population health calculations and can be repeated for multiple emissions scenarios. Validation statistics and plots generated using the Model Evaluation Toolkit are also presented for key regulated pollutants.

This work is part of the Natural Environment Research Council (NERC) funded WM-Air (West Midlands Air Quality Improvement Programme) project led by the University of Birmingham, which aims to improve understanding of pollution sources and levels in the West Midlands region, alongside the development of tools for assessing the air quality, health and economic impacts of potential policy measures. CERC were involved in emissions processing using the Emissions Inventory Toolkit (EMIT) and also supported model evaluation for this project. The images show the variable-sized spatial splitting domains used to optimise run time, with smaller domains used in areas with a higher density of explicitly modelled road sources, and the resulting high-resolution contour plot of the annual average NO₂ concentrations across the region.