Quantifying uncertainty of emission estimates in National Greenhouse Gas Inventories using bootstrap confidence intervals

Greenhouse gas (GHG) emissions have exacerbated global warming, and consequently are the focus of worldwide reduction efforts. Reducing emissions involves accurately estimating GHG emissions and the uncertainty associated with such estimates. The uncertainty of GHG emission estimates is often assessed using the 95% confidence interval. Given a small sample size and non-normal distribution of the underlying population, the uncertainty estimate obtained using the 95% confidence interval may lead to significant bias. Bootstrap confidence interval is an effective means of reducing bias. This work presents a procedure for estimating the uncertainty of GHG emission estimation using bootstrap confidence intervals. Numerical simulation is performed for GHG emission estimates under three distributions (namely normal, log-normal and uniform) to find the 95% confidence intervals and bootstrap confidence intervals. Finally, the accuracy and sensitivity of the uncertainty of various interval estimations are examined by comparing the coverage performance, interval mean and interval standard deviation. Simulation results indicate that the bootstrap intervals are more applicable than the 95% confidence interval given non-normal dataset and small sample size. Moreover, when sample size n is less than 30, the bootstrap confidence interval has a smaller interval length with a smaller deviation than that of the classical 95% confidence interval regardless of whether the data distribution is normal or non-normal. This study recommends a sample size greater than or equal to 9 for estimating the uncertainty of emission estimates. When the sample size n exceeds 30, either the normality-based 95% confidence interval or bootstrap confidence intervals may be used regardless of whether the data distribution is normal or non-normal. A case study of carbon stock from Taiwan demonstrates the feasibility of the proposed procedure.

► Emission estimates data generated by Monte Carlo Method.
► Bootstrap Samples are selected by bootstrap Simulation.
► Computed Emission estimates sorted to find BCa Confidence Interval.
► Various parameter combinations and Distributions are considered.
► Validity and Sensitivity are tested by coverage percentage.