Page 22 - EER-1-1

Explora: Environment
and Resource WTW emissions of road and rail transport

each predictor variable. The selection of these methods was The models also included a real-world emissions
guided by the available information and data. correction factor (σ). However, in all cases, this was set to
Each model variable was defined as a parametric unity because the models, and the data from the literature
distribution which represented the probability of possible used in the analysis were already designed to reflect real-
13
values. The definition of input variables is discussed word operation, or else the emission factors were based on
in Section 2.5 for direct (or TTW) emissions, and real-world fuel use and activity.
in Section 2.6 for indirect (or WTT) emissions. This 2.5. Simulation of direct emissions
relied on statistical analysis of empirical data, software
simulation, findings from peer-reviewed scientific studies, The simulation of direct emissions from road transport,
consultation, or expert judgment where applicable. The rail passenger transport, and rail freight transport is
method for determining total annual emissions for the described in the following sections. In each case, the input
Brisbane-Melbourne route is discussed in Section 2.7. variable definitions – including parametric distribution
types, typical values, and plausible ranges – are given in
Quantitative data were used, where available, to Tables S5-S9.
develop the input distributions, supplemented with the
results from peer-reviewed scientific studies and other 2.5.1. Road transport
information from the literature. Statistical techniques were Fleet-average emission factors for road transport
applied to develop the input distributions, namely, Monte were derived from a recent and detailed study of the
Carlo simulation, bootstrap analysis, and parametric Australian road transport sector over the period 2019 –
15
14
distribution fitting. 2050. Two software tools were used to estimate direct
5
The R packages “fitdistr,” “fitdistrplus,” “extraDistr,” emissions, namely, the Australian Fleet Model (AFM)
“sn,” and “truncdist” were used to optimize the fitting of and the Net Zero Vehicle Emission Model (n0vem). 16,36
simulated emissions data to pre-defined types of statistical Short descriptions of these models are included in
distribution. The most appropriate theoretical distribution Supplementary File B (Software). For this study, direct
was, then, determined by comparing all fitted parametric TTW emission factors (g/vehicle-km) were extracted
distributions with the simulated input values. This was from a recent study, and the relevant statistics are shown
5
done visually using quantile-quantile plots for all fitted in Table 2. These reflected fleet-average and real-world
distributions, and by applying the Cramer-Von Mises test emissions from conventional ICEVs (petrol, diesel,
and minimizing fitting errors. 8,9 and liquefied petroleum gas), hybrid EVs (HEVs), and
plug-in HEVs (PHEVs) in the 3 years of interest. Zero
The following candidate distribution types were
considered in the fitting process: Uniform (U: a, b), direct emissions were assumed for BEVs and fuel-cell
8
triangular (T: a, b, c), normal (N: m, s), log-normal (L: m, s), EVs (FCEVs).
weibull (W: s, s), gamma (G: s, r), exponential (E: s), non- In addition to direct emissions, n0vem also estimates
standard beta distribution (B: s, s), the location-scale t (O: electricity and H consumption for various types of EV,
2
m, s, df), skewed t (S: m, s, a, df), and Dirac delta function year of manufacture and driving conditions, and these
(D: m). Truncation was applied to the fitted distributions estimates were extracted for this study. The results showed
by setting lower and upper limits (a, b). The plausible the following:
range for each input was defined as the 99.7% confidence • Fuel cell electric PVs (H ) were expected to have
2
interval (CI, equivalent to ±3 SD in a normal distribution), insignificant penetration in the on-road fleet and were
which prevented the use of unrealistic values. The input ignored for all years. The contributions of BEVs and
8
variable definitions, including parametric distribution PHEVs to total travel were negligible in 2019, but
definitions, typical values, and plausible range, are given increased to 6% in 2030 and 74% in 2050.
in Tables S5-S9. • The additional fleet-average electricity requirement
for electric PVs on the route was predicted to be
In the Monte Carlo simulation, random samples were
drawn from input distributions with one million iterations. 0.019 kWh/km in 2030 and 0.196 kWh/km in 2050.
These samples were then propagated through the assessment The associated triangular distributions (Wh/km) were
8
model to generate probability output distributions. This adopted from another study, i.e., T: 18, 21, 19 for 2030
approach not only allowed for the estimation of expected and T: 186, 217, 196 for 2050.
values (e.g., fleet-average emissions) but also gave a EVs (battery and H ) were expected to have an
2
reasonably accurate depiction of the associated variability insignificant amount of travel in the freight sector in
and uncertainty. 2019 and 2030 (<1%) and could be ignored. The situation

Volume 1 Issue 1 (2024) 6 doi: 10.36922/eer.3470