Dataset where svm performance is significantly different from random forest

Geographical datasets (e.g. when predicting population density from the built environment characteristics in my case) are, at least from my experience, one of the cases where SVM performs consistently worse than random forests. Although I have not investigated the issue in detail, it seems that random forests are more resistant to high amounts of noise and can learn to apply different rulesets in different regions or environments, something which SVM (SVR) fails to do.