Jobs Button PSU Button        
PSU PSU ARL Logo PSU Logo
APPLIED RESEARCH LABORATORY AT THE PENNSYLVANIA STATE UNIVERSITY

FSMA | computational mechanics (cm) - Atmospheric Dispersion

Images | Animations (below)

Computational Mechanics (CM) has accomplished and is currently performing various projects related to modeling atmospheric transport and dispersion of a contaminant. Several of these projects are joint with the PSU Meteorology Department (http://www.met.psu.edu/). Related collaborations with Aerospace Engineering and Architectural Engineering are on-going. These projects include developing new ways of assessing uncertainty in dispersion models, assimilating sensor data into the models, characterizing the source of a contaminant and the associated meteorological conditions, doing high fidelity modeling o a potential release in an urban setting, and designing better ways of interfacing the dispersion models with the forcing numerical weather prediction models. The models range from basic Gaussian plume and puff models through full CFD in urban settings plus Lagrangian particle dispersion. Much of the work includes the DoD dispersion model, SCIPUFF.

Assessing Uncertainty in Dispersion Models
Computational Mechanics has been working with the Meteorology Department to parameterize Numerical Weather Prediction (NWP) ensemble-variability for use by the atmospheric transport and dispersion model, SCIPUFF, to estimate ATD uncertainty from meteorological sources. The effort investigates closure concepts for the length and velocity scale parameters used by SCIPUFF to estimate ATD uncertainty from meteorological sources. A primary outcome of this research will be best-practice recommendations for computing the SCIPUFF ensemble variability model metrics and documentation of model sensitivities for real-world environments that exhibit non-ideal weather conditions.

Figure 1 shows the increase in the mean plume footprint produced by computing an appropriate length scale based on analysis of an ensemble of meteorological conditions. The plume on the left is the result of a single base case run of SCIPUFF. The one on the right results from enhancing the data with the ensemble variability parameters.

Sensor Data Fusion for Dispersion
The goal of this project is to devise, test, and implement methods to assimilate monitored data into atmospheric transport and dispersion models. We would like to predict transport and dispersion of a hazardous contaminant in spite of inexact source information, inexact knowledge of meteorological conditions, imperfect models based on ensemble averages, the inherent uncertainties of turbulent diffusion, and monitored data with errors.

Figure 2a shows a plume of contaminant that is embedded in a meandering wind field. Figure 2b is a plume reconstructed from data samples every 125 m in space for a time sampling rate of 80 s apart. The wind direction had to be inferred from the concentration values. The plume is well reproduced without assuming known meteorological conditions.

Source Characterization
PSU has developed methods to characterize contaminant sources. The technique begins with with monitored concentrations and compares them to predicted concentration using an atmospheric transport and dispersion model. Then the model’s input parameters are modified to better match the measurements. The technique uses artificial intelligence techniques, specifically a Genetic Algorithm (GA). Figure 3 indicates the process. Given sensor readings, the GA back-calculates the appropriate input parameters. Since there is a large sensitivity to meteorological parameters, we also back-calculate meteorological data that is necessary to compute the correct concentration patterns. The following parameters have been back-calculated with the genetic algorithm model:

  • Source strength
  • Source location (3d)
  • Time of release
  • Wind direction
  • Wind speed
  • Stability class
  • Boundary layer height

Modeling a Hypothetical Chlorine Release
Computational Mechanics modeled a hypothetical Chlorine release using state-of-the-science techniques. Figure 4 shows the results using two different methods. The upper plot used the DoD Hazard Prediction and Assessment Capability package, including the Urban Wind Model, Urban Dispersion Model, and Second Order Closure Puff Model (SCIPUFF). The lower plot was produced using full CFD capabilities using the Acusolve model to do a Detached Eddy Simulation (DES) calculation. The general regions of the dispersed Chlorine are quite similar, but the DES calculation is able to produce a more detailed pattern.

Computational Mechanics personnel have been collaborating with Meteorology Department faculty to develop methods for high-resolution meteorological modeling for atmospheric transport and dispersion and probabilistic weather analysis. This project includes real-time meteorological modeling on Beowulf clusters, numerical weather prediction for the Torino Olympics, configuring new weather modeling systems for Chem-Bio modeling, accomplishing case studies, and calibrating the uncertainties to ensemble uncertainties. Figure 5 shows a scatter plot between the ensemble covariance and the mean error covariance after binning the data. The linear relationship implies that a calibration can be derived from such analyses.

PSU/ARL Computational Mechanics Division has high fidelity tools for CFD modeling and dispersion computation that are being used to compute the details of atmospheric transport and dispersion. Some of the techniques used include Unsteady Reynolds-averaged Navier-Stokes (URANS) Modeling, Detached-Eddy Simulation (DES) and its variants, Large-Eddy Simulations (LES), and Lagrangian Particle Trajectory Models. Figure 6a shows the streamlines about PSU west campus buildings as computed using DES techniques. Figure 6b shows the details of vorticity dynamics about a cube. The results of these DES calculations agreed well with measured data.

The PSU/ARL CM division has the capability to model particle paths through a simulation. This has been done within simple wind fields, numerical weather prediction models, and full CFD model wind fields. These Lagrangian particle methods go beyond simply following the field, but also include realistic amounts of modeled noise to stimulate separations of large clouds of particles. This Lagrangian method is an alternate method to compute dispersion that is capable of giving a detailed picture of dispersion near to structures. Figure 7 shows Lagrangian pathlines for 10 µm particles in the wake of a cylindrical column at moderate Reynolds number.

Images
image
View Image (183kb)

Figure 1:

This figure shows the increase in the mean plume footprint produced by computing an appropriate length scale based on analysis of an ensemble of meteorological conditions. The plume on the left is the result of a single base case run of SCIPUFF. The one on the right results from enhancing the data with the ensemble variability parameters.

image
View Image (110kb)

Figure 2a and 2b:

A plume of contaminant that is embedded in a meandering wind field. Figure 2b is a plume reconstructed from data samples every 125 m in space for a time sampling rate of 80 s apart.

image
View Image (56kb)
Figure 3:
This figure shows graphically how measured data taken by sensors is incorporated in order to make better concentration predictions.
image
View Image (96kb)

Figure 4:

The results of using two different methods to analyze a hypothetical Chlorine release. The left plot used the DoD Hazard Prediction and Assessment Capability package, including the Urban Wind Model, Urban Dispersion Model, and Second Order Closure Puff Model (SCIPUFF). The right plot was produced using full CFD capabilities using the Acusolve model to do a Detached Eddy Simulation (DES) calculation.

image
View Image (39kb)

Figure 5:

This figure shows a scatter plot between the ensemble covariance and the mean error covariance after binning.

image
View Image (256kb)

Figure 6a and 6b:

The streamlines about PSU west campus buildings as computed using DES techniques. Figure 6b shows the details of vorticity dynamics about a cube. The results of these DES calculations agreed well with measured data.

image
View Image (145kb)

Figure 7:

Lagrangian pathlines for 10 µm particles in the wake of a cylindrical column at moderate Reynolds number.


Animations
image
View Animation
(Flash Player 1.9mb)

Animation 1:

Data is assimilated into the one dimensional Shallow Water Equations using four different methods: nudging, three dimensional variational assimilation, extended Kalman filter, and the ensemble Kalman filter. The bottom green curve is the bottom topography and the up curve shows the free surface. Note that after the large perturbation, all methods are effective in helping the free surface recover and remain stable.

image
View Animation
(Flash Player 287kb)

Animation 2:

The detailed flow features that can be captured with Lagrangian particle methods. As 5000 particles are released from the plane behind a bluff body, they are captured within the flow field. Note the two re-circulation zones – one in the lee of the emission plane and one that forms in the cavity of the bluff body. Particles are colored by velocity.