Uncertainty analysis of an aviation climate model and an aircraft price model for assessment of environmental effects

Abstract

Estimating, presenting, and assessing uncertainties are important parts in assessment of a complex system. This thesis focuses on the assessment of uncertainty in the price module and the climate module in the Aviation Environmental Portfolio Management Tool (APMT). The aircraft price module is a part of the Partial Equilibrium Block (PEB) and the climate module is a part of the Benefits Valuation Block (BVB) of the APMT. The PEB estimates a future fleet and flight schedule and evaluates manufacturer costs, operator costs, and consumer surplus. The BVB estimates changes in health and welfare for climate, local air quality, and noise from noise and emissions inventories output from the Aviation Environmental Design Tool (AEDT). The assessment was conducted with various uncertainty assessment and sensitivity analysis methods: the nominal range sensitivity analysis (NRSA), the hybrid Monte Carlo sensitivity analysis, the Monte Carlo regression analysis, the vary-all-but-one Monte Carlo analysis, and the global sensitivity analysis with Sobol' indices and total sensitivity indices. Except the NRSA, all other analysis methods are based on the Monte Carlo simulation with random sampling. All uncertainty assessment methods provided the same ranking of significant variables in both APMT modules. Two or three significant variables are clearly distinguished from other insignificant variables. In the price module, seat coefficients are the most significant parameters, and age is an insignificant factor between input variables of the regression model. In the climate module, statistical analyses showed that climate sensitivity and short-lived RF are most significant variables that contribute the variability of all three outputs. However, the HMC analysis suggested that discount rate is the most sensitive factor in the NPV estimation.

(cont.) Comparing the Sobol's indices with the total sensitivity indices showed that there are no significant interactions to change the ranking of significant variables in both modules.