I think a log transformation of the original non-normal data might help you assuming you don't have any zero value in your observations.

Please avoid transformations and normalizations when performing inferences with natural variables (such as concentrations of heavy metals in biotic or abiotic matrices). This is because such procedures alter and deeply bias the behavior and nature of your data, often driving to misleading and generally erroneous results. As you may have noticed, very little of what occurs naturally exhibits normal behavior; phenomena in nature are largely characterized by noise and entropy. As advice, I recommend performing non-parametric or semi-parametric analyses using the actual variables' distributions you measure, thus obtaining more realistic estimates

Bingyao Huo, Raw data does not need to be normally distributed for ANOVA, residuals should be. If non-normality of residuals is expected, don't perform any transformations, they alter the interpretation. Instead use a method to, calculate the standard errors / p-values that do not assume normality, for example, permutation testing.