Instead of taking log, take log(y+1), such that zeros turn out to be ones and may then be kept within the regression. This biases your model a bit and is considerably frowned upon, but in practice, its negative unwanted side effects are typically pretty minor. If it’s not too many rows of information which have a zero, and those rows aren’t theoretically necessary, you’ll be able to determine to go ahead with the log and lose a couple of rows from your regression. After transforming a variable, observe how its distribution, the r-squared of the regression, and the patterns of the residual plot change.
A high correlation doesn’t necessarily indicate that there’s good settlement between the two methods. Not to be confused with Estimation statistics § Gardner–Altman plot. The other piece of knowledge plotted on a Skew-T is the wind pace and course.