Bayesian analysis of time-interval data for environmental radiation monitoring

Time-interval (time difference between two consecutive pulses) analysis based on the principles of Bayesian inference was compared with frequentist methods to determine the method with the highest detection probability and the shortest average run length. Using experimental and simulated data, Bayesian analysis of time-intervals (Bayesian (ti)) was compared with Bayesian and frequentist analyses of counts in a fixed count time (Bayesian (cnt) and 1.65σ, respectively). Experimental data were acquired with DGF-4C (XIA, Inc) system in list mode. Simulated data were obtained using Monte Carlo techniques to obtain a random sampling of the Poisson distribution. All statistical algorithms were developed using R (R Core Development Team, 2010). Detection probabilities and average run lengths for the three methods were compared. Bayesian analysis of time-interval information provided a similar detection probability as Bayesian analysis of count information, but was able to make a quicker decision with fewer pulses at relatively higher radiation levels. The relationships of the source time, change points and modifications to the Bayesian approach for increasing detection probability are presented.