For some variables, for instance precipitation, the distribution is closer to a normal distribution after taking te logarithm or square root of the data. When the operation assumes a normally distrbuted variable (e.g., correlation, regression) the results may be better after such a transformation. For plots the variables are transformed back to the original scale, so that a straight-line regression becomes an exponential or parabolic line.

Negative data are replaced by undefined. In the case of a logarithmic transformation, zeros are arbitrarily replaced by 1/5 of the lowest positive value encountered.