Comparative investigation of source term estimation algorithms using FUSION field trial 2007 data

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Tuesday, 19 January 2010: 3:45 PM
B308 (GWCC)
Nathan Platt, Institute for Defense Analyses, Alexandria, VA; and D. Deriggi

Presentation PDF (1.0 MB)

The release of hazardous materials into the atmosphere on the battlefield or in populated areas must be considered for future scenarios. Given a warning based on detections at a few sensors, it should be useful to rapidly provide an estimate of the location, time of release, and amount of material released. Such information could lead to refined predictions of the hazardous area and support follow-on actions to investigate the cause and nature of the hazardous release.

In September 2007, a short-range test – Fusing Sensor Information from Observing Networks (FUSION) Field Trial 2007 (FFT 07) – designed to collect data to support development of prototype source term estimation (STE) algorithms was conducted. A comparative investigation of STE algorithms began in 2008. The general method of this investigation was to first provide the participating developers with a subset of sensor data that was collected on selected FFT 07 trials. Next, developers provided “blind” predictions which could then be compared to the parameters of the actual release. Phase I of this investigation consisted of 104 individual cases of simulated sensor data that were distributed in September 2008. These cases simulated continuous streams of concentration data for ingestion by STE algorithms. A total of eight different STE algorithm developers participated in this exercise. Phase II of this exercise incorporates: a) lessons learned from Phase I, b) the addition of ‘bar-sensor' input data stream, and c) the use of a simulated environment to supplement field trial data and is planned to start in early FY10.

As of late July 2009, a total of fourteen full and partial sets of predictions were received with some exercise participants providing multiple sets of predictions based on different algorithms they have been developing. Preliminary analyses of the results including statistics on the differences between predicted and actual locations and masses of the releases were provided to STE participants at the end of July, 2009. These analyses considered several parameters that might influence results including the number of sensors (four versus sixteen), the release type (instantaneous versus continuous), the time of the release (day versus night), meteorological inputs (“scientific” quality met versus “simulated” operational met), and the number of sources (single versus double versus triple versus quad releases). In the presentation, we highlight our findings with an emphasis on describing the similarities and differences among STE algorithm performance. We also discuss potential implications of this work for the Joint Effects Model requirement for source term estimation.