Comparison of pooled standard deviation and standardized-t bootstrap methods for estimating uncertainty about average methane emission from rice cultivation

Abstract

The general sample standard deviation and the Monte-Carlo methods as an estimate of confidence interval is frequently being used for estimates of uncertainties with regard to greenhouse gas emission, based on the critical assumption that a given data set follows a normal (Gaussian) or statistically known probability distribution. However, uncertainty estimated using those methods are severely limited in practical applications where it is challenging to assume the probability distribution of a data set or where the real data distribution form appears to deviate significantly from statistically known probability distribution models. In order to solve these issues encountered especially in reasonable estimation of uncertainty about the average of greenhouse gas emission, we present two statistical methods, the pooled standard deviation method (PSDM) and the standardized-t bootstrap method (STBM) based upon statistical theories. We also report interesting results of the uncertainties about the average of a data set of methane (CH4) emission from rice cultivation under the four different irrigation conditions in Korea, measured by gas sampling and subsequent gas analysis. Results from the applications of the PSDM and the STBM to these rice cultivation methane emission data sets clearly demonstrate that the uncertainties estimated by the PSDM were significantly smaller than those by the STBM. We found that the PSDM needs to be adopted in many cases where a data probability distribution form appears to follow an assumed normal distribution with both spatial and temporal variations taken into account. However, the STBM is a more appropriate method widely applicable to practical situations where it is realistically impossible with the given data set to reasonably assume or determine a probability distribution model with a data set showing evidence of fairly asymmetric distribution but severely deviating from known probability distribution models. (C) 2015 Elsevier Ltd. All rights reserved.