addressing uncertainty in oil and natural gas industry greenhouse

Additionally, Monte Carlo simulation can also be computationally intensive, with 10,000 simulations being
the norm. The bigger difficulty/barrier is the need to build a simulation model in Excel for uncertainty
analysis that reflects the complexity of the emissions estimation model and allows generation of the
variables required for the Monte Carlo simulation software.
To perform a Monte Carlo simulation, one must first determine the distribution of the data for each of the
uncertain variables used in the emissions model estimate. Ideally, each of these distributions should be
derived from data and knowledge of the underlying process. It is helpful in many instances to first graph
the data, and using the shape of the graph to determine the underlying distributions. The parameters that
define the distribution could be derived from the data. For example, a normal distribution is defined by its
mean and variance. If data are limited, one may have to rely on expert judgment to determine the
underlying distribution.
It is then necessary to statistically test the hypothesis that the data follow a certain distribution. The test
will vary based on the hypothesized distribution. For example, the Shapiro-Wilks test is often used to test
if the data are normal or lognormal. Options to test for other distributions include Empirical Distribution
Functions (Singh and Singh, 2006).
Once distributions are determined for all of the data sources, the Monte Carlo simulation will proceed by
randomly sampling each of the distributions that describe the data used for estimating emissions. As many
as 10,000 replicate samples are typically taken, with the total emissions being estimated for each replicate.
These repeated determinations of emission are used to generate a distribution of the total emissions with its
mean being the estimate of total emissions, and its uncertainty determined by its variance.
4.6.2 Comparison of Uncertainty Propagation and Monte Carlo
Section 6.3.1 of the IPCC Good Practices document compares the uncertainty propagation method and the
Monte Carlo simulations (IPCC, 2006). It notes that the uncertainty propagation method’s assumption of
normality leads to symmetric 95% confidence intervals whereas the Monte Carlo method can take into
account the fact that emissions are bounded below by zero to fit an asymmetric (and thus narrower)
confidence interval. If the data are skewed and one transforms the data (discussed earlier), one could
achieve the asymmetric confidence intervals using uncertainty propagation, as well.
Since the Monte Carlo simulations can assume a truncated distribution, the lower confidence limits tend to
be closer to the mean than the upper confidence limits. The IPCC Good Practices document goes on to
state that the two methods produce results that are fairly comparable. It recommends that countries report
the results of the uncertainty propagation method and those countries with “sufficient resources and
expertise” report Monte Carlo results as well.
Pilot Version, September 2009 4-35