Land abandonment is a global phenomenon whose environmental consequences are difficult to assess. The Murcia region is one of the most arid regions in southern Europe and also one of the most prone to land abandonment. This study researches which environmental features are more relevant to explain abandonment at the agricultural plot scale. Geomorphometric features were measured at different scales to investigate which scales could be more relevant. Two different models have been used: logistic regression, a statistical model that allows the interpretation of the involved features, and Random Forest, a machine learning model with a higher predictive power but lower interpretability. The combined use of both such models allows a set of predictors to be selected, which, when used with Random Forest, produce a map that is highly accurate for predicting abandonment and, when used with logistic regression, produce an interpretable model. The main conclusion is that climate is the most relevant factor to explain land abandonment. Copyright © 2015 John Wiley & Sons, Ltd.