Madagascar Population Map Metadata Report

Prediction Weighting Layer Used in Population Redistribution

The data presented below represent the predicted number of people per ~100 m pixel as estimated using the random forest (RF) model as described in Stevens, et al. (2015). The following pages contain a description of the RF model and its covariates, their sources and any metadata collected for each covariate. The prediction weighting layer is used to dasymetrically redistribute the census counts and project counts to match estimated populations based on UN estimates for the final population maps provided by WorldPop.

Stevens, F. R., Gaughan, A. E., Linard, C., & Tatem, A. J. (2015). Disaggregating Census Data for Population Mapping Using Random Forests with Remotely-Sensed and Ancillary Data. PLOS ONE, 10(2), e0107042. doi:10.1371/journal.pone.0107042

plot of chunk predict_density

Madagascar Census Data and Observed Population Density

These data are the population density values used to estimate the RF model used to create the prediction weighting layer you see above. Values represent population density as measured by people per hectare and calculated from population counts within each census unit. These values are used as the dependent variable during model estimation.

Madagascar Census Data, 2006, Admin-level 2

Folder: Census
File Name: MDG_adm4_2006.shp
Source: Estimates derived from RGPH (Recensement Général de la Population et de l'Habitat) 1993, Institut National de la Statistique, Madagascar
Description: These census data were extracted from GeoHive (www.geohive.com). Required fields for map production are ADMINID and ADMINPOP.
Class: polygon
Derived Covariates:
area, buff, zones,

class       : SpatialPolygonsDataFrame 
features    : 17459 
extent      : 311938, 1089711, 7167963, 8675263  (xmin, xmax, ymin, ymax)
coord. ref. : NA 
variables   : 22

plot of chunk census_data

Random Forest Model and Diagnostics

These output and figures outline the estimated RF model that is used to predict the population density weighting layer. The model is fitted to the population density values for the preceding census data using covariates aggregatedfrom the ancillary data sources summarized following the model diagnostics.


Call:
 randomForest(x = x_data, y = y_data, ntree = popfit$ntree, mtry = popfit$mtry,      nodesize = length(y_data)/1000, importance = TRUE, proximity = TRUE) 
               Type of random forest: regression
                     Number of trees: 500
No. of variables tried at each split: 22

          Mean of squared residuals: 0.44
                    % Var explained: 86

plot of chunk random_forest

Covariate Metadata

Folder: Landcover
File Name: mdg_lc.tif
Source:
Description:
Class: raster
Derived Covariates:
cls011, dst011, cls040, dst040, cls130, dst130, cls140, dst140, cls150, dst150, cls160, dst160, cls190, dst190, cls200, dst200, cls210, dst210, cls230, dst230, cls240, dst240, cls250, dst250, clsBLT, dstBLT,

class       : RasterBrick 
dimensions  : 15210, 7952, 120949920, 1  (nrow, ncol, ncell, nlayers)
resolution  : 100, 100  (x, y)
extent      : 302929, 1098129, 7157914, 8678914  (xmin, xmax, ymin, ymax)
coord. ref. : +proj=utm +zone=38 +south +datum=WGS84 +units=m +no_defs +ellps=WGS84 +towgs84=0,0,0 
data source : D:\APRF\RF\data\MDG\Landcover\Derived\landcover.tif 
names       : landcover 
min values  :        11 
max values  :       250