Parallelization Performances of PMSS Flow and Dispersion Modeling System over a Huge Urban Area

Abstract

The use of modeling as a support tool for crisis management and decision planning requires fast simulations in complex built-up areas. The Parallel Micro SWIFT SPRAY (PMSS) modeling system offers a tradeoff between accuracy and fast calculations, while retaining the capability to model buildings at high resolution in three dimensions. PMSS has been applied to actual areas of responsibilities of emergency teams during the EMERGENCIES (very high rEsolution eMERGEncy simulatioN for citIES) and EMED (Emergencies for the MEDiterranean sea) projects: these areas cover several thousands of square kilometers. Usage of metric meshes on such large areas requires domain decomposition parallel algorithms within PMSS. Sensitivity and performance of the domain decomposition has been evaluated both for the flow and dispersion models, using from 341 up to 8052 computing cores. Efficiency of the Parallel SWIFT (PSWIFT) flow model on the EMED domain remains above 50% for up to 4700 cores. Influence of domain decomposition on the Parallel SPRAY (PSPRAY) Lagrangian dispersion model is less straightforward to evaluate due to the complex load balancing process. Due to load balancing, better performance is achieved with the finest domain decomposition. PMSS is able to simulate accidental or malevolent airborne release at high resolution on very large areas, consistent with emergency team responsibility constrains, and with computation time compatible with operational use. This demonstrates that PMSS is an important asset for emergency response applications.Atmosphere 2019, 10, 404 2 of 17 Gaussian models can provide simulation results in a short time and partly account for the effects of buildings. Nonetheless, they cannot take into account the unsteady flow in and above the urban canopy and the complex dispersion pattern between the buildings.Contrary to simplified models, computational fluid dynamics (CFD) models solve the Navier-Stokes equations. CFD models are categorized, according to the scale of the range of the spatio-temporal scales that are resolved, in Reynolds averaged Navier-Stokes (RANS) [11,12] and large eddy simulation (LES) [13][14][15]. While they provide reference solutions for complex flows in a built-up configuration, they are most often limited to academic configurations and/or small built-up domains: they indeed require very long computational times by default, making them unsuitable for live operational use.As a trade-off between computational time and accuracy of the solution, Röckle [16] first suggested an original approach. This approach uses mass consistency with observations to solve for the mean flow. Vortex flow structures are analytically defined around and between buildings and within street canyons, and then a mass consistency scheme is applied. Kaplan and Dinar [17] improved the approach by coupling the flow with a Lagrangian particle dispersion model to compute dispersion in an urban area. The PMSS modeling system uses this approach and integrates subsequent modifications using more recent urban tra...

Accelerated Time and High-Resolution 3D Modeling of the Flow and Dispersion of Noxious Substances over a Gigantic Urban Area—The EMERGENCIES Project

“…The PMSS modeling system [17,[19][20][21] is a parallelized flow and dispersion modeling system. It is constituted of the Parallel-SWIFT (PSWIFT) flow model and the Parallel-SPRAY (PSPRAY), an LPDM, both used in small-scale urban mode.…”

“…The PMSS modeling system has been validated both in scalar mode [22,26,27] and regarding parallel algorithms [17,20]. The parallel testing was performed in computational environments ranging from a multicore laptop to several hundred cores in a highperformance computing center.…”

Hazardous atmospheric dispersion in urban areas: A Deep Learning approach for emergency pollution forecast

“…PMSS modelling system has been thoroughly validated against wind tunnel and full-scale experimental results [29,30]. The parallelism was shown to be very efficient, from a multicore laptop up to clusters with several hundreds or thousands of cores in the case of a high-performance computing center [31,32]. In terms of computational time, to give an order of magnitude considering a spatial domain of 1 km 3 with a horizontal and vertical (close to the ground) resolution of 2 meters, PMSS computational time is around 5 minutes for the flow and 10 more minutes for tracking a plume for 15 minutes using an optimal number of cores on a standard work station.…”

High-Speed Visualization of Very Large High-Resolution Simulations for Air Hazard Transport and Dispersion

“…The PMSS modeling system (see [17][18][19][20]) has been used to model the flow and dispersion on domain with extension of several tens of kilometers and horizontal grid resolution of 3 m. The modeling system consists of the individual PSWIFT and PSPRAY models. The PSWIFT model [21][22][23][24] is a flow and turbulence parallel, 3D, mass-consistent, and terrain-following diagnostic model that uses a Röckle-type [25] approach to take into account buildings.…”