Enhancing GHSL Population Grids Using Hexagon KH-9 Built-up Data: Refining 1970s Rural and Peri-Urban Distributions in Istanbul

TL;DR

This paper tackles the bias and temporal gaps in historical gridded population data by combining declassified KH-9 Hexagon imagery with harmonised settlement-level census records to produce high-resolution population grids for 1975–1990 in Istanbul. By comparing a GHSL baseline, a Hexagon-enhanced workflow, and a Hexagon-plus-census approach, the authors demonstrate that contemporaneous, fine-grained built-up masks significantly reduce spurious population allocation to undeveloped land and better reflect intra-district settlement structure. The approach improves both pixel-level accuracy and the distribution of population across DoU classes, offering a reproducible method to reconstruct historical population dynamics in data-sparse regions. The method has broad applicability for historical urbanisation studies and climate-risk assessments where archived high-resolution imagery is available.

Abstract

Accurate reconstruction of historical population distributions from the 1970s to the 1990s remains a significant limitation in global gridded population products due to coarse built-up data and limited census records. This study is, to our knowledge, the first to integrate declassified Hexagon KH-9 reconnaissance imagery into gridded population mapping. We enhance the GHS-POP framework by combining segmented built-up land cover from the HexaLCSeg dataset, derived from 1977 KH-9 imagery, with geocoded settlement-level census data to construct high-resolution historical population grids. Applied to Arnavutkoy and Cekmekoy in Istanbul for the period 1975-1990, we evaluate three dasymetric approaches, including a standard GHSL baseline, a Hexagon-enhanced workflow, and a fully integrated model incorporating local census records. Pixel-wise and zonal analyses show that GHSL misallocates populations to historically undeveloped regions, while the Hexagon-derived dataset substantially improves the representation of fragmented rural and peri-urban areas often missing from global products. Incorporating settlement-level LAU-2 census data further refines spatial population distribution. The results demonstrate that combining historical reconnaissance imagery with high-resolution census data improves the accuracy of historical population grids, and given the global coverage of declassified missions, this methodology offers significant potential for reconstructing historical population patterns in data-scarce regions worldwide.

Figures (12)

• Figure 1: Coverage and count of Hexagon satellite imagery per 1-degree grid cell. Aggregated KH-9 satellite totals were obtained from the USGS website earth_resources_observation_and_science_eros_center_declassified_2017. Map data from OpenStreetMap and produced by the authors.
• Figure 2: Comparative accuracy of global gridded population products for both areas of interest in Istanbul. The heatmaps show the percentage error for GHS-POP, GRUMPv1, LandScan, and WorldPop, computed against official district census totals for benchmark years where each product is available. dark-red cells indicate higher overall percentage error.
• Figure 3: Overview of the study areas: Arnavutköy (west) and Çekmeköy (east), Istanbul. the background classification shows the HexaLCSeg historical landcover (LC) dataset, of which the Built-Up area was derived sertel_automatic_2024
• Figure 4: Figure showing the share of area that is attributed to different Degree of Urbanisation classes over time for both the areas of interest according to the GHS-SMOD_GLOBE_R2023A dataset schiavina_ghs-pop_2023
• Figure 5: Example of the resolution of Hexagon imagery for the study areas.