Using emerging computing hardware to tackle the problem of complexity in air-quality and climate modelling studies

  • Supervisors: Dr David Topping
    Dr Paul Connolly


  • External Supervisors: Dr Michael Bane, High End Compute, Manchester

  • Contact:

    Dr David Topping, david.topping@manchester.ac.uk

  • CASE Partner: Yes - Paratools Inc, Sci-Tech, Darebsury UK

Application deadline: 3 February 2017

Introduction:

Aside from directly and indirectly influencing the earths climatic system, atmospheric aerosol particles are the single most important agent in determining air quality and is widely viewed as the most damaging of the regional pollutants.  Aerosol particles sit in a complex ‘soup’ of gas phase products in the atmosphere. As our knowledge of key processes improves, or simply grows, mechanistic models become much more complex. Typically, models ranging from the single particle scale to models of laboratory chambers and then the atmosphere decrease in complexity, be it chemical and/or numerical. Propagating full process and chemical complexity in geophysical models is currently not possible with existing computational resource. This is important since we have recently demonstrated the danger of not capturing all processes necessary for predicting cloud droplet number with associated impact on radiative forcing (Topping et al.2013).

Attempts to address the big questions of climatic and health impacts implies improving the knowledge on aerosol composition and properties yet, sooner or later, we must take decisions on what to do with the complexity of both. Should we care about linking climate and health impacts to a chemical ‘complexity’? Presently, we do not have appropriate technologies to address this. This is the focus of this project by utilizing exciting and novel software and compute hardware.

Project Summary:

How will regional atmospheric aerosol models of the future maintain pace with computational hardware developments and will this enable us to embed more detailed aerosol representations in regional models? The answer is not entirely clear, even though the landscape of HPC provision is changing rapidly with emphasis on external accelerators (Graphics cards, Intel XeonPhi) and low energy computing. Most regional or national HPC facilities will embed such computing technology in the immediate future, shifting the required skill set of model designers. Chemical kinetics models may be responsible for over 90% of an atmospheric model’s computational time even without gas-particle partitioning. However, including gas-aerosol partitioning, even without condensed phase processes, in regional models increases costs by up to a factor of 9 (Archer- Nicholls et al 2014). This presents a timely challenge. The Kinetic PreProcessor (KPP) is a general analysis tool that enables the rapidly generation of correct and efficient chemical kinetics code and is used widely. KPPA on the other hand (the Kinetics PreProcessor: Accelerated) (Linford et al 2011), is the next generation KPP tool that generates OpenMP, CUDA and OpenCl code for external accelerators. In this project you will assist in the development of a new generation of gas-aerosol schemes that includes gas-aerosol partitioning alongside gas phase kinetics in KPPA for regional models via managing work carried out through Paratools. 

The aims of the project are to:

  • Identify where the computational bottlenecks are in existing gas-aerosol schemes.
  • Work with our CASE partner to apply new software and hardware tools to mitigate these bottlenecks.
  • Assess the improvement in science deliverables by this move to using emerging technology. 

The candidate will work with academics and CASE partners to learn the programming and software tools necessary to complete this project. You should have some level of comfort with programming already. This could include familiarity with Python, Fortran, Matlab, etc. The modeling tools offer the opportunity to develop new skills that are valuable to employers, particularly as the landscape of computing hardware changes. Also, you will spend time with our CASE partners at their base in Daresbury, UK, or in the US.  Paratools have access to hardware and software resources at the Performance Research Lab, U. Oregon in the US for performance engineering tasks including development, performance, evaluation, and optimization. ParaTools has computer resource facilities conducive to performing software development, performance monitoring, and parallel runtime studies using multi-core CPUs and acceleration hardware devices in its offices in Bruyères-le-Châtel, France, and its parent company ParaTools, Inc.’s offices in Boulder, Eugene, and Baltimore in the US. 

This project will appeal to a candidate with an enthusiasm for truly understanding the importance of computing, is happy to travel and engage with international partners, and who has strong interests in cross-disciplinary science.

References:

Archer-Nicholls, S., et al:  Geosci. Model Dev. Discuss., 7, 6061-6131, 2014.

Linford, J. C.  et al  IEEE TPDS: Special Issue on High-Performance Computing with Accelerators, Vol. 22, No. 1, p. 119-131, 2011.

Topping, D et al: Nature Geoscience 6, 443–446, 2013.

Apply Now