Multiple Global Population Datasets: Differences and Spatial Distribution Characteristics

Ruxia Chen, Huimin Yan, Fang Liu, Wenpeng Du, Yanzhao YangISPRS Int. J. Geo Inf 2020 · DOI

TL;DR

This paper compares four major global population datasets (HYDE, GPWv4, GHSL, WorldPop) to understand their differences in accuracy and spatial distribution. It reveals that while datasets using remote sensing and mobile data show higher accuracy in data-rich regions like the UK, areas with sparse data and challenging terrain (like Tibet) exhibit greater consistency across datasets despite lower absolute accuracy.

What Problem Does This Paper Address?

Accurate population data is crucial for understanding human impact on the environment, resource utilization, and socioeconomic development. However, the proliferation of global population datasets, each with different data sources and spatialization methods, creates a challenge for researchers in selecting the most appropriate dataset for their specific application. The differences in accuracy and spatial distribution characteristics between these datasets are not always clear, leading to potential uncertainties in research findings. This paper aims to bridge this gap by systematically comparing four widely used global population datasets across diverse geographical and socioeconomic contexts.

Methods

The study employs a multi-faceted approach to compare the four selected population datasets:

  • Accuracy Assessment: The datasets' population estimates are compared against official statistical data (World Bank and Statistical Yearbook) for the year 2015 to calculate relative errors.
  • Spatial Distribution Consistency Analysis: A pixel-by-pixel comparison is conducted to assess how consistently the four datasets represent population distribution. This involves reclassifying population density into nine levels using a natural breakpoint method and then analyzing the agreement between datasets.
  • Case Study Approach: Four distinct case areas were chosen to represent a range of geographical and demographic characteristics: the United Kingdom (high density, high urbanization), Argentina (low overall density, high urban concentration), Sri Lanka (flat terrain, high density), and the Tibet Autonomous Region of China (high altitude, sparse population, data acquisition challenges).
  • Analysis of Influencing Factors: The study investigates how factors such as data source, spatialization methods, geographical characteristics (terrain, altitude), urbanization level, and data acquisition difficulty influence the accuracy and consistency of the datasets.

Datasets

The following four global spatialized population datasets were used:

  • History Database of the Global Environment (HYDE) version 3.2.000: Known for its long historical time series.
  • Gridded Population of the World version 4 (GPWv4): Based on administrative units and statistical data.
  • Global Human Settlement Layer (GHSL): Utilizes remote sensing data to identify built-up areas.
  • WorldPop: Incorporates a variety of data sources including remote sensing, mobile communication, and census data, employing advanced modeling techniques.

Reference data for total population in 2015 was obtained from the World Bank and the Statistical Yearbook of the Tibet Autonomous Region of China.

Key Insights and Ideas

  • Data Source and Spatialization Matter: The fundamental differences in how datasets are constructed (e.g., reliance on census data vs. remote sensing, interpolation methods) directly impact their accuracy and spatial representation.
  • Data-Rich vs. Data-Scarce Regions: Datasets perform better in areas with abundant and high-quality data (like the UK), showing lower relative errors. Conversely, in regions with data limitations (like Tibet), while absolute accuracy might be lower, the datasets tend to show higher consistency with each other, suggesting a shared understanding of sparse distribution.
  • Urbanization's Dual Role: High urbanization levels can lead to more accurate population estimates due to better data availability and distinct built-up areas (benefiting GHSL). However, complex urban-rural gradients and medium population densities can also be areas of high inconsistency between datasets.
  • The Future is Multi-Source: Datasets like WorldPop, which integrate diverse data sources and advanced machine learning models, show promise for improving spatial accuracy and capturing complex population patterns.

Key Results

  • Accuracy: GPWv4 and GHSL exhibited the smallest absolute relative errors (within 3%) for total population estimation, likely due to their reliance on administrative unit data and refined spatial allocation. WorldPop showed significant relative errors (up to 20%) in Argentina, Sri Lanka, and Tibet, potentially due to its complex spatialization methods and data limitations in these regions. HYDE's accuracy varied by region, with a low error in the UK but higher errors in Tibet.
  • Consistency: The Tibet Autonomous Region showed the highest proportion of completely/highly consistent regions (97.01%), indicating that despite lower absolute accuracy, the datasets largely agree on the sparse distribution. The UK, with its complex spatial distribution, had the lowest consistency (66.75%).
  • Population Density and Consistency: Data consistency was generally high in extremely low population density areas across all case studies. In highly urbanized and densely populated areas, consistency was also high, particularly for WorldPop and GHSL. However, medium-density areas with high urbanization levels were the most prone to inconsistency between datasets.

Limitations and Future Work

  • Data Acquisition Difficulty: The study acknowledges that obtaining accurate and up-to-date demographic data, especially in high-altitude or remote regions, remains a significant challenge and directly impacts dataset accuracy.
  • Definition of Urban/Rural: Differences in how urban and rural populations are defined across countries (e.g., inclusion of small towns) can introduce subtle variations in population statistics.
  • Future Directions: The authors suggest that improving spatial population datasets in data-scarce regions will rely on refining demographic data and leveraging big data sources like remote sensing and mobile communication. For areas with medium population density, further research using advanced machine learning algorithms and the development of standardized experimental areas for verification are recommended to enhance reliability and consistency. The development of more accurate and detailed spatial population data is seen as a crucial direction for future research.

Summary updated

BibTeX
@article{chen2020multiple,
  author = {Ruxia Chen and Huimin Yan and Fang Liu and Wenpeng Du and Yanzhao Yang},
  title = {{Multiple Global Population Datasets: Differences and Spatial Distribution Characteristics}},
  journal = {ISPRS Int. J. Geo Inf.},
  year = {2020},
  doi = {10.3390/ijgi9110637},
}
Figure