We consider a nearly Gaussian random variable X (see \cite{Lefebvre}) that, after a power transformation, the variable Xc where c={(2k+1)/(2j+1)}, k,j={0,1,…}, is approximately Gaussian. This transformation is useful to model errors in temperature forecasts.
@article{arxiv.1308.0248,
title = {Nearly Gaussian random variables and application to meteorology},
author = {Rui Gonçalves},
journal= {arXiv preprint arXiv:1308.0248},
year = {2013}
}