Fig 1.
Study area administrative units.
NUTS (Nomenclature des unités territoriales statistiques) and LAU (Local Administrative Units) in the study area. LAU delineations within Berlin correspond to local planning areas, Proj: ETRS-89. Inset A illustrates different higher and lower average LAU sizes across federal states. Administrative boundaries from [51] under dl-de/by-2-0 license (https://www.govdata.de/dl-de/by-2-0).
Table 1.
List of acronyms.
Fig 2.
Earth Observation-based covariate layers.
(1A) Imperviousness and (1B) infrastructure density from OSM to generate (1C) building density. (2) Building Height. Projection: ETRS-89.
Fig 3.
Workflow: Top-down redistribution of census population.
Workflow of the data-driven redistribution of census data using different dasymetric mapping approaches. Refer to text for details about BD-BUILD, BD-RESI, WD-DENS, WD-VOL, WD-VOLADJ.
Fig 4.
Workflow: Bottom-up gridded population.
Building fraction, building height and floor area per capita for bottom-up mapping. Refer to text for details about BU-LFA.
Fig 5.
Top-down and bottom-up gridded population mapping quality.
Mean absolute percentage error (MAPE), Slope and R2 of redistribution models at different spatial validation scales (NUTS-1 to LAU and BPA). All numbers used to create this figure in SI 8 in S1 File.
Fig 6.
Population mapping quality in relation to population density and spatial resolution.
Population density (top) and spatial resolution (bottom) of all NUTS-1, NUTS-3 and LAU areas related to their respective relative estimation error (REE) in the different redistribution models. Black dots: Mean value of y-axis bins. Grey bars: Mean values of bins +/- one standard deviation. |skew.| = absolute skewness of the means.
Fig 7.
Distribution of Local Administrative Units and reference population by error range.
Histogram of LAU validation units (left) and census reference population (right) within REE bins for each model. Orange colors represent underestimation, purple colors represent overestimation, grey color represents accurate predictions (REE ranges from -10% to +10%). All numbers used to create this figure in SI 7 in S1 File.
Fig 8.
Quality of gridded population models by LAU (spatial representation).
Spatial distribution of REE by LAU and model. Purple shades imply over-estimation, orange shades imply under-estimation, grey shades imply accurate predictions (REE between -10% and +10%). Administrative boundaries from [51] under dl-de/by-2-0 license (https://www.govdata.de/dl-de/by-2-0).
Fig 9.
Gridded population product comparison.
Comparison of WorldPop (Left, Constrained/UN-Adjusted/100m, target year 2020, [27]), GHS-POP (center, target year 2015, [21]) and the best product map from this study (right).
Fig 10.
Quality of bottom-up gridded population mapping using floor area per capita from a different year or spatial subset.
(Left) Ratio of population estimates within all LAU when using LFA/cap (national average) from different years compared to LFA/cap from 2018. Blue line: Predicted Population Total. (Right) Ratio of population estimates within all LAU when using LFA/cap from a single NUTS-1 unit for the whole study area only compared to individual LFA/cap per NUTS-1 unit. Blue line: Predicted Population Total.