Validation of a Real-Time CBRN Track Before Detection Model in a Highly Urbanized Environment Using Large Eddy Simulation

Monitoring, detecting, and responding to chemical, biological, radiological, and nuclear (CBRN) threats in real-time is a complex and multi-tiered process. Previous efforts often required analysts to fuse information coming from multiple sensors and modalities, define its context, and decide on an actionable path forward. Multi-Node Analytics (MNA) is an emerging tool that leverages advents in Internet of Things (IoT), low SWaP computing, and communications to reduce the minimum detectable signal, lower false alarm rate, and decreases time to detect (compared to traditional techniques). However, practical limitations associated with collecting high-fidelity space-time plume realizations in operational environments (e.g. heavily urban areas) under realistic geospatial layouts make testing MNA system response difficult. Therefore, a methodology for accurately characterizing a MNA algorithm’s performance is critical for evaluating system performance in new environments and driving system improvements.

In this presentation, a methodology for testing Areté’s MNA algorithm in a novel highly urban environment is presented. First, we will introduce Areté’s MNA algorithm which utilizes a simplified transport and dispersion model with a micro-scale wind field estimated from real-time wind and turbulence measurements to determine the degree to which CBRN sensor observations are correlated in a manner consistent with a plume. This algorithm has been successfully tested in rural environments and is currently used operationally in several rural and semi-urban environments, however, performance in highly urbanized environments remains elusive. Next, we verify the Large Eddy Simulations (LES) from the Joint Outdoor-indoor Urban Large Eddy Simulation (JOULES) are a means to create synthetic meteorological and CBRN sensor data for evaluating MNA. Using JOULES, a total of 210 tracers were run over seven meteorological environments consisting of 30 ensemble members each for a case study in Lower Manhattan (New York City, USA). The JOULES output data cube is sampled at three locations for meteorological data and biological particle data to create simulated sensor datasets which are then used as input into MNA. Results show that Areté’s MNA algorithm correctly detects 80.5% of the threat tracers simulated by JOULES. Further investigation reveals that 15.2% of the tracer simulations not detected by Areté’s MNA algorithm, are attributed to errors in estimating the micro-scale wind field used by the MNA algorithm. Additional analysis shows that the MNA wind field algorithm performs best in environments with positive heat flux and light winds and worst in environments with zero heat flux and strong winds. The presentation concludes with ideas for improving the methodology for generating wind field to improve the detection rate in an urban environment.