Using regression-based source detection algorithm for source location with FFT-07 data

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Tuesday, 19 January 2010: 4:00 PM
B308 (GWCC)
Randolph J. Evans, ENSCO, Inc., Melbourne, FL; and S. E. Masters and M. A. Kienzle

The FUSION Field Trial 2007 (FFT-07) was a short-range (~500 meters) tracer field test conducted in September 2007 at Dugway Proving Grounds using tracer releases in a highly instrumented network of samplers and meteorological measurement sites. Data were collected for the purpose of evaluating different source term algorithms. Algorithm developers were provided limited data and were asked to determine source term characteristics such as location and release rates, without knowing those characteristics.

We developed a linear regression-based source identification model and ran it for 102 of the 104 cases derived from the FFT-07 data. The linear regression approach uses a transport and dispersion model utilizing turbulent wind measurements to generate emissions from a grid of virtual sources, and correlates the predicted signals to the observed signals across the array of digiPIDs. After the information from all virtual sources is processed using release time and location, a “weighted centroid” is calculated to identify a potential source location. Preliminary results indicate this method demonstrates considerable skill for single sources (burst or continuous), but currently will only identify the mean spatial location of multiple sources.

Source location and release rate results were tabulated and submitted to the FFT-07 independent analysis team. This team compared the model results with the actual data results and the comparison suggests that the linear regression model predicts average source locations within 100 meters. We are conducting further tests with an expanded virtual source domain. This presentation will present a discussion of the linear regression method along with updated results from the expanded grid network.