event

PhD Defense by Daejin Kim

Primary tabs

School of Civil and Environmental Engineering

 

Ph.D. Thesis Defense Announcement

 

Large-scale, Dynamic, Microscopic Simulation for Region-wide Line Source Dispersion Modeling

 

By

 

Daejin Kim

 

Advisor:

 

Dr. Randall L. Guensler (CEE); Dr. Haobing Liu (CEE)

 

Committee Members:

 

Dr. Michael O. Rodgers (CEE); Dr. Angshuman Guin (CEE);
Dr. Fang (Cherry) Liu (COC), Dr. Catherine Ross (CRP)

 

Date & Time: March 25th, 3:00pm

Location: Mason Building, Room 2228

 

 

 Complete announcement, with abstract, is attached

 

Although a variety of modeling tools have been developed to predict potential public exposure to harmful transport emissions at regional and sub-regional scales, computational efficiency remains a critical concern in the design of modeling tools.  Microscale dispersion models run at high resolution often require extremely long runtimes for larger roadway networks and high-resolution receptor grids.  Motivated by the challenges encountered in the previous modeling efforts, this work develops an advanced modeling framework for region-wide applications of line source dispersion models that integrates a high-performance emission rate lookup system (i.e., MOVES-Matrix), link screening, and innovative receptor site selection routines to further accelerate model implementation within distributed computing modeling framework.  The case study in the 20-county metropolitan Atlanta area accounts for an extremely large number of link-receptor pairs (more than 160 thousand transportation links and 1.1 million receptors) demonstrates that the modeling system generates comparable concentration estimates to extremely-high-resolution processes, but with very high computational efficiency.  Using AERMOD, the regional analysis required only 10 days to complete the analyses, compared to a total runtime by traditional methods of more than one year.  The improvement in computational speeds are attributed to 1) the employment of supervised random forests machine learning model for link screening model, which objectively excludes transportation links that do not significantly affect receptor concentrations, and 2) the use of the supercomputing clustering system, where multiple AERMOD simulation jobs are split and simultaneously processed, thereby significantly reducing the total run-time.  The dynamic-grid-receptor model systematically generates the network of receptor sites, based upon road geometry and meteorological conditions, to help minimize model runtime without undermining pollutant concentration predictions.  The comprehensive modeling methodology presented in this work will make comparison of air quality impacts across complex project scenarios (and transportation development alternatives over large geographic areas) much more feasible.  All these aspects should be of interest to a broad readership engaged in near-road air quality modeling for transportation planning and air quality conformity and for environmental analysis under the National Environmental Policy Act.

Status

  • Workflow Status:Published
  • Created By:Tatianna Richardson
  • Created:03/10/2020
  • Modified By:Tatianna Richardson
  • Modified:03/17/2020

Categories

Keywords