Application of Latin Hypercube Sampling to RADTRAN 4 Truck Accident-Risk Sensitivity Analysis

The transportation risk analysis code, RADTRAN 4 (Neuhauser and Kanipe 1992), computes estimates of incident-free dose consequence and accident dose-risk. The output of the code includes a tabulation of sensitivity of the result to variation of the input parameters for incident-free analysis. Values are calculated using closed mathematical expressions derived from the constitutive equations, which are linear. However, the equations for accident risk are not linear, in general, and a similar tabulation has not been available. Because of the importance of knowing how accident-risk estimates are affected by uncertainties in the input parameters, a direct investigation was undertaken of the variation in calculated accident dose-risk with changes in individual parameters. A limited, representative group of transportation scenarios was used, initially, to determine which of 23 accident-risk parameters affect the calculated accident dose-risk significantly (Mills et al. 1995). Many ofthe parameters had minimal effect on the output, and others were \"fixed\" either by regulation, convention, or standards. The remaining 5 input arrays were selected for further study through Latin Hypercube Sampling (LHS) (Iman and Shortencarier 1984). The use of LHS yields statistical infonnation about risk calculations by providing multiple input-parameter sets, compiled from \"random\" sampling of parameter distributions, for multiple RADTRAN calculations. The LHS method requires fewer observations than classical Monte Carlo methods to yield statistically significant results. This paper summarizes the preliminary parameter study and LHS application results to date, in addition to presenting the results of subsequent studies of RADTRAN input parameter distributions and their effects on risk estimate uncertainty.