Variable importance for the random forest models for the county and prefecture scale (top axis) and the 5×5 km grid cells scale (bottom axis).

<p>The variable importance is calculated by comparing the mean squared error from models with the original dataset with the mean squared error from models with an altered dataset where the predictor variable is randomly permuted. Acronyms: AET = actual evapotranspiration, PC00-10 = Change in population density between 2000 and 2010, GDP/Area = gross domestic product per km², HII = Human Influence Index.