Disaggregating Census Data for Population Mapping Using Random Forests with Remotely-Sensed and Ancillary Data
Fig 5
A visual comparison of Kenyan population maps for census data in 1999 produced at coarser administrative unit (Level 4) and finer-scale administrative unit (Level 5).
The difference illustrates the finer gradations of RF model predictions for the density weighting layer when there are larger ranges of observed population densities present in training data (N = 505 for Level 4, N = 6622 for Level 5).