Disaggregating Census Data for Population Mapping Using Random Forests with Remotely-Sensed and Ancillary Data
Fig 3
Variable importance for Random Forest regression, presented as the mean decrease in residual sum of squares when the variable is included in a tree split.
The model including variables here was used to produce the density weighting layer for the dasymetrically distributed population map in Fig. 2. Variable names are defined and described in Table 1.