- Formulate and test filtering and data assimilation methods for dispersion modelling with a view to incorporating innovations in to operational models (e.g. those used by the Met. Office);
- Develop recommendations for integrating such methods into now-casting tools.
Atmospheric, near-wall, turbulent flows through vegetation or urban environments are highly complex due to their spatial variability, temporal unsteadiness and multiscale nature. However, modelling them effectively is important for a number of domains of application including carbon dioxide fluxes over forests, seed dispersion and particulate and pollutant dispersion in cities. (See Fig. 1 for an example of our recent work in the latter domain, involving numerical modelling and experiment).
Averaging methods over different spatial domains and time scales are a way to separate out the effects of time-dependent and space-dependent processes in a mathematically consistent fashion, leading to practical models for canopy flow and dispersion (Coceal and Belcher, 2004; Belcher et al., 2015; Goulart et al., 2018; Hertwig et al., 2018). Alternative approaches include the use of modal decomposition of the flow field (e.g. Higham et al., 2018) to statistically separate coherent activity from stochastic motions, to study how coherent structures drive dispersion processes (e.g. Beaumard et al., accepted), or to decompose the flow field in to local and non-local components to examine these dynamics from a more fundamental perspective (Keylock, 2018,2019).
In both cases, an open research question of significant practical importance concerns the use of dispersion models for now-casting; specifically, the manner by which obtaining data (from a formal measurement, or from Instagram videos of an explosion) is used to guide an updated forecast. Our goal is to formulate a statistical data assimilation framework within which is embedded an existing dispersion modelling tool. The assimilation framework will be designed so that data of varying uncertainty can be incorporated appropriately, to produce a decision tool of use for our colleagues in the Met. Office.
Our methodology is primarily theoretical and computational, with data and codes from NCAS and the Met. Office available to serve for testing our methods and for validation. The statistical dimension will build on well-known techniques for forecasting under uncertainties (Bayesian methods). By utilising high quality simulations of the salient physics (large-eddy simulations or LES) we will also be able to determine the innate uncertainties in the model formulation and test alternate decompositions for developing enhanced canopy dispersion models (see Hertwig et al. (2018) for an example of a comparison of a LES and a number of dispersion models). The ensemble of dispersion model simulations under different parameterizations will be assigned varying probabilities of likelihood based on the observational data (with its own precision). This will result in a “pruning” of the model parameter space to narrow the confidence limits on a now-casting tool for decision makers.
Training and Skills
The student will participate in seminars and reading groups put on both for and by the PhD students in Loughborough. They will also be able to access courses of technical relevance at Loughborough provided by the high-performance computation team and the CDT for Future Propulsion and Power. We will host a CENTA training session on modal decomposition methods. NCAS offers a range of training for PhD students, including: “Introduction to Atmospheric Science” and “Introduction to Scientific Computing”. The Met. Office will provide in-kind support to help train the student in the use of their codes (e.g. the NAME dispersion model).
Year 1: Training and familiarisation with statistical theory, applied dispersion modelling and computational fluid dynamics. Literature review on dispersion methods and the decomposition methods.
Year 2: Development of the integrated framework and the assignation of probabilities to simulations. Formal introduction of data as a means to update these probabilities and refine the simulation in real time.
Year 3: Insertion of new dispersion model into the framework based on alternate averaging philosophy. Running of the simulation formalism on different test cases, and comparison of the effectiveness of different dispersion models.
Partners and collaboration (including CASE)
The primary collaboration is between Loughborough and NCAS with very positive discussions about in-kind support from the Met. Office. Chris Keylock at Loughborough is co-lead of the EPSRC U.K. Fluids Network special interest group on Non-equilibrium Turbulence, of which Omduth Coceal at NCAS is a member. Both have theoretical and applied research experience in the field. In addition, the team will exploit their existing links to develop further collaborations with Ecole Centrale de Lyon, Los Alamos and Tokyo Tech., who bring expertise on the use of reduced order and high-resolution modelling for decision support, and urban planning.
Please add project/institutional contact details including a link to the application website if applicable:
e-mail for contact: email@example.com
Institutional homepage for second supervisor:
Webpage for related research: http://www.nonequilibrium-turbulence.org.uk