## Procedures Of Approach

The principle of Monte Carlo analysis is to select random values of emission factor, activity data and other estimation parameters from within their individual probability density functions, and to calculate the corresponding emission values. This procedure is repeated many times, using a computer, and the results of each calculation run build up the overall emission probability density function. Monte Carlo analysis can be performed at the category level, for aggregations of categories or for the inventory as a whole. Statistical software packages are readily available - some of which include Monte Carlo algorithms that are very user-friendly12.

Like all methods, Monte Carlo analysis only provides satisfactory results if it is properly implemented. This requires the analyst to have scientific and technical understanding of the inventory. Of course, the results will only be valid to the extent that the input data, including any expert judgements, are sound.

The Monte Carlo approach consists of four clearly defined steps shown in Figure 3.7. Only the first of these requires effort from the user, the remainder being handled by the software package. The emission inventory calculation, the PDFs, and the correlation values should be set up in the Monte Carlo package. The software performs the subsequent steps. In some cases, the inventory compiler may decide to set up its own programme to run a Monte Carlo simulation; this can be done using statistical software. The Section, 'Choosing a simulation technique and sample size' below contains a short discussion of various software packages.

12 Winiwarter and Rypdal (2001), Eggleston et al. (1998) and Monni et al. (2004) provide examples of Monte Carlo analysis applied to national greenhouse gas inventories to estimate uncertainties both in overall emissions and emissions trends. Another example of the use of Monte Carlo analysis is given in McCann et al. (1994). More detailed descriptions and applications of this method are presented in Bevington and Robinson (1992), Manly (1997), Morgan and Henrion (1990), and Cullen and Frey (1999). A brief example of the application of Monte Carlo analysis is provided in Box 3.2 based on Ogle et al. (2003).

Figure 3.6