Uncertainty: Normal distribution vs Lognormal distribution

Question

Hi,

I created an process in openLCA. I use ecoinvent processes as inputs. There is a certain range in the input quantities. EG.  for one important parameter, values can range from 1 gram to 4 gram, so I use 2,5 grams for the baseline scenario, but I want to perform a Monte Carlo simulation.

My question is: Can I simply calculate a MEAN and STANDARD DEVIATION and choose Normal distribution to input the uncertainty into my process? Or is it necessary to choose Lognormal distribution with  GEOMEAN and Geometrical Standard Deviation? Unfortunately, excel does not calculate geometrical standard deviation (and I am not an expert in statistics)

Because in the precursor processes (Ecoinvent) uncertainty is always defined as lognormal distribution.

Is it problematic to "mix" normal and lognormal distribution in one system?

Answer 1

Lognormal is often used as default in LCA because it is strictly positive.  Random sampling from a lognormal distribution will not return zero or negative numbers.  Normal distributions are evenly distributed on either side of your mean, and the left-hand tail can dip into negative territory, which means some of your samples will have impossible or unintended values.  That said, you can choose normal if you want, you need to determine whether it is an accurate model for your scenario.