Source Inversion for contaminant plume dispersion in urban environments using building-resolving simulations
Flow in urban environments is complicated by the presence of buildings, which divert the flow into often unexpected directions. Contaminants released at ground level are easily lofted above tall ({approx} 100 m) buildings and channeled through urban canyons that are perpendicular to the wind direction (see e.g., IOP 9 in Chan, 2005). The path of wind and scalars in urban environments is difficult to predict even with building-resolving computational fluid dynamics codes, due to the uncertainty in the synoptic wind and boundary conditions and other errors in the models. Given the difficulties due to the complexity of urban flows, solving an inverse problem becomes quite challenging. That is, given measurements of concentration at sensors scattered throughout a city, is it possible to detect the source of the contaminant? The ability to locate a source and determine its characteristics in a complex environment is necessary for emergency response for accidental or intentional releases of contaminants in densely-populated urban areas. The goal of this work is to demonstrate a robust statistical inversion procedure that performs well even under the complex flow conditions and uncertainty present in urban environments. Much work has previously focused on direct inversion procedures, where an inverse solution is obtained using an adjoint advection-diffusion equation. The exact direct inversion approaches are strictly limited to processes governed by linear equations. In addition, they assume the system is steady-state and that the equations are linear (Enting, 2002). In addition to adjoint models, optimization techniques are also employed to obtain solutions to inverse problems. These techniques often give only a single best answer, or assume a Gaussian distribution to account for uncertainties. General dispersion related inverse problems, however, often include non-linear processes (e.g., dispersion of chemically reacting substances) or are characterized by non-Gaussian probability distributions (Bennett, 2002). Traditional methods also have particular weaknesses for sparse, poorly constrained data problems, as well as in the case of high-volume, potentially over-constrained and diverse data streams. We have developed a more general and powerful inverse methodology based on Bayesian inference coupled with stochastic sampling. Bayesian methods reformulate the inverse problem into a solution based on efficient sampling of an ensemble of predictive simulations, guided by statistical comparisons with observed data. Predicted values from simulations are used to estimate the likelihoods of available measurements; these likelihoods in turn are used to improve the estimates of the unknown input parameters. Bayesian methods impose no restrictions on the types of models or data that can be used. Thus, highly non-linear systems and disparate types of concentration,meteorological and other data can be simultaneously incorporated into an analysis. In this work we have implemented stochastic models based on Markov Chain Monte Carlo sampling for use with a high-resolution building-resolving computational fluid dynamics code, FEM3MP. The inversion procedure is first applied to flow around an isolated building (a cube) and then to flow in Oklahoma City (OKC) using data collected during the Joint URBAN 2003 field experiment (Allwine, 2004). While we consider steady-state flows in this first demonstration, the approach used is entirely general and is also capable of dealing with unsteady, nonlinear governing equations.
- Research Organization:
- Lawrence Livermore National Laboratory (LLNL), Livermore, CA (United States)
- Sponsoring Organization:
- USDOE
- DOE Contract Number:
- W-7405-ENG-48
- OSTI ID:
- 886928
- Report Number(s):
- UCRL-CONF-216903; TRN: US200617%%389
- Resource Relation:
- Journal Volume: 47; Journal Issue: 6; Conference: Presented at: 86th American Meteorological Society Annual Meeting, Atlanta, GA, United States, Jan 29 - Feb 02, 2006
- Country of Publication:
- United States
- Language:
- English
Similar Records
Event Reconstruction for Atmospheric Releases Employing Urban Puff Model UDM with Stochastic Inversion Methodology
Dynamic Data-Driven Event Reconstruction for Atmospheric Releases