JSTO Sensor Data Fusion Program Goals Workshop on Chemical Data Assimilation John Pace DTRA/CBT 21 June 2005

JSTO Environmental Hazard Prediction Program Overview Background: Joint Science and Technology Office is part of the Chem-Bio Defense Program (CBDP) –Manage projects to develop S&T capabilities needed for operational CB modeling systems Environmental Hazard Prediction program has many elements, including SDF program –Improved dispersion models, urban dispersion models, coastal & littoral models, several others

JSTO Sensor Data Fusion Program Overview Original concept for JSTO program derived from weather data assimilation analogy –Combine first guess from model, valid at time t, with data from Chem and Bio sensors valid at approximately the same time –Resulting analysis should be more accurate than original model prediction or sensor data –Several key differences between CB data and weather data assimilation methodologies Differences on constraints on solutions No need to use combined solution as first guess for next calculation Instead, need to work backward to estimate source location and time

JSTO Sensor Data Fusion Program Overall JSTO program direction from DTRA/CBT –Program Manager, Principle Investigator from other organizations JSTO program coordinated with multi-national TP9 program –Coordinates several related US, UK, Canadian, Australian projects Principle objectives: –Blend CB sensor data with dispersion model –More accurate depiction of CB hazard area –Ability to “backtrack” to source location –Provide guidance about sensor performance –Support tool to place sensors for facility protection Status: –SDF Program established, Program Plan under development –Good progress on several projects –Several good leads for additional performers to add to program –But…always looking for good new ideas and approaches

What Sensor Data Fusion Program is NOT Considerable effort underway simply to acquire and display data from CB sensors in common format, on command and control display systems –Goal is to identify hazard areas, compare to model output Provides some limited capability if sensors are always accurate, if the sensor characteristics are understood by the user, if nearby sensor readings agree, and there aren’t more sensors data than the user can understand These assumptions are not usually true Beyond these limits, much more is achievable than this basic display capability can provide

"Model" Results

Sensor Data

Reality (?) based on sensor data

Data fusion?

Claude Monet The Seine at Giverny, 1897

Jackson Pollock Number 1, 1950 (Lavender Mist), 1950

Prototype Backtracking Capability UK developed STEM-II –Updates source-term estimates in real- time –Robust Bayesian approach –Handles real-time chemical sensor data and observer data –Demonstration version available –Further development required, including Multiple sources Continuous and line releases Biological releases

Question Can our program benefit from idea exchange with air quality chemical data assimilation work?