Smoothing and Time Series Modeling of Nuclear Material Accounting Data for Protracted Diversion Detection

The time series of material balances in nuclear material accounting (NMA) is also known as the material unaccounted for (MUF) sequence. This paper applies a joint cusum test to residual time series from NMA that arise from either of two options. The first residual series is the standardized, independently transformed MUF (SITMUF) sequence that relies on an estimate of Σ, the MUF covariance matrix. The second residual series arises from using either time series modeling or nonparametric smoothing on the MUF sequence and ignores the estimate of Σ. Assuming that the MUF sequence is multivariate Gaussian and ignoring estimation error in Σ, we find the anticipated result that the first option is superior to the second option. In addition, we find that the SITMUF scheme in the first option is robust to modest estimation error in Σ over a large number of idealized facilities, but not necessarily so for any specific idealized facility. These two findings provide a perspective on previous literature that addressed a perceived weakness in NMA.