Real-time fusion of sensor data to achieve improved situational awareness

In hazard release incidents it is particularly important to have good situational awareness in order to make the most effective decisions and save lives. Chemical and biological and other sensors can be deployed to measure the presence of toxic airborne contaminants; however it is necessary and a significant challenge to make best use of the information provided to determine how a hazard will evolve and warn people. This is an area of significant research around the world.

There are several approaches to the problem, including fusing data to determine a source term that best matches the observations and data assimilation that uses observations to refine a modelled representation of hazard. The output of a source estimation algorithm can be used to make a hazard prediction using a standard dispersion model, whereas the data assimilation method produces a “now-cast” of the hazard.

We have developed a rapid data assimilation approach that fuses real-time sensor observations. It represents the hazard cloud as a collection of Gaussian puffs and updates their position, size and mass using the expectation-maximisation algorithm. This internal hazard representation provides a hazard now-cast that can be easily displayed, and is also in a form that can is compatible with forward dispersion models thereby providing improved hazard prediction. Details of the model will be given together with recent developments that are being carried out. These range from simple statistical spatial-domain fitting, through to Bayesian techniques and Markov Chain Monte Carlo approaches to give spatio-temporal fitting to sensor data. We are also developing an approach to estimate the local meteorology based on the sensor measurements, which may provide significant operational benefits in some circumstances. We will also provide validation results using the FUSION Field Trial 07 (FFT07) dataset.