Propagating transport-code input parameter uncertainties with deterministic sampling

A novel approach to uncertainty quantification in codes simulating fusion plasma, deterministic sampling (DS), is evaluated. This method uses a few carefully selected samples and can be used to propagate input parameter uncertainties through calculations where other sampling methods are unmanageable due to time constraints. The primary analysis is performed on the transport code TRANSP, but another faster code is also tested where a comparison with Monte Carlo sampling is made. The tests, performed with two pulses at the Joint European Torus (JET), show that even lower order DS will give a reliable estimation of the standard deviation of the calculated neutron rate. However, a higher order DS can give information about higher output moments, such as skewness an kurtosis. The TRANSP-simulated neutron rates of both examined pulses are found to have an uncertainty with an upward skewness, meaning input parameter uncertainties are can better explain an underestimation of the neutron rate than an overestimation. This information can, for example, be lead to a better benchmarking comparison between the measured and calculated neutron rates.