Optimizing Sensor Placement for Nuclear Forensics with Multiple Data Types

The goal of nuclear detonation forensics is to rapidly determine detonation parameters such as explosion yield and material type to aid in accurate and definitive attribution. Early-time signatures recorded on networks of seismic, infrasound, optical and electromagnetic pulse sensors can provide high value information to forensic analysts if clear signals can be observed. This leads to the problem of determining optimal locations for network sensors so as to maximize signal quality, as predicted by physical propagation models and stochastic models of background noise. The sensor placement problem becomes complex when the network is tasked with observing multiple detonation sites. We are developing a physics-based computational platform to evaluate predicted network performance and solve for sensor locations that optimize the performance.

Our approach combines random sampling and gradient search techniques, applied to maximizing the probability of observing high signal-to-noise signals emanating from any of several designated detonation sites. The inputs to the search algorithm include signal amplitudes predicted with complex propagation models and empirically derived estimates of noise, which are used to compute signal-to-noise ratios from all prospective detonation sites to a grid of possible sensor locations. We have applied the new computational framework to the monitoring of hypothetical explosions in the United States with a network of seismic and infrasound sensors. Our initial results are promising and illustrate the importance of refined network performance metrics, data fusion to combine observations from different sensor types, and improved physics-based models of observables.