Accurate knowledge of population distribution is vital for effective spatial analysis and policy support across numerous domains, including urban planning, disaster risk management, and infrastructure development. Traditional methods, relying primarily on place-of-residence statistics from official sources like censuses, offer a static view of population density, failing to capture the dynamic shifts caused by human mobility throughout the day and across seasons. While promising, mobile phone data suffers from accessibility limitations and inconsistencies. This research addresses these shortcomings by developing a novel, multi-layered dasymetric approach to generate high-resolution spatio-temporal population data for the entire European Union. The study aims to provide a more comprehensive and accurate representation of population distribution by integrating official statistics with diverse geospatial data sources, thereby addressing the limitations of existing methodologies and offering a valuable resource for researchers and policymakers.

Existing methods for mapping population distribution range from simple areal interpolation techniques to sophisticated dasymetric mapping approaches. Dasymetric mapping, which disaggregates population counts from administrative units to finer zones using covariates like land use/land cover, has proven effective in improving the geographical representation of population. However, most existing population grids focus solely on place-of-residence, representing nighttime population distribution. This static view ignores the significant daily and seasonal fluctuations in population density due to human activities. Previous attempts to map daytime population distribution have been limited in scope, typically focusing on smaller areas or individual countries and utilizing diverse methodologies with varying data sources. The emergence of unconventional big geospatial data sources, such as mobile phone records, offers exciting possibilities, but challenges related to data access, quality, and consistency hinder widespread use for large-scale applications. This paper builds upon the existing literature by proposing a novel data fusion method to overcome the limitations of existing approaches and produce a consistent and comprehensive spatio-temporal population dataset for the entire EU.

The researchers employed a multi-layered dasymetric approach to generate the spatio-temporal population grids. This methodology involved four main phases: 1. **Estimation of monthly and regional population stocks:** Sixteen population groups were identified based on their activity patterns (residents, workers in 11 economic sectors, students at two educational levels, tourists, and non-working/non-studying population). Monthly stocks for each group were estimated at the NUTS3 regional level for 2011 using official statistics from Eurostat and other sources, incorporating seasonal variations based on school calendars and tourism data. 2. **Mapping of land-use features:** A high-resolution land use/land cover (LULC) map was created by integrating data from CORINE Land Cover, OpenStreetMap, and TomTom Multinet, improving on the thematic and spatial detail of existing CLC data. Additional layers were created for various activity densities based on points of interest (POIs) from these sources, to reflect the locations of activities associated with each population group. 3. **Dasymetric disaggregation of population stocks:** A two-tiered approach was used to downscale population stocks from NUTS3 regions to 100m pixels and then to 1km² grids. The first step distributed population groups among relevant LU types proportionally to their occurrence within the region. The second step allocated populations to individual grid cells based on built-up density. Specific adjustments were made for residents and tourists. The procedure generated 204 intermediate population grids (12 months x 17 population groups), which were then aggregated to create daytime and nighttime grids for each month. 4. **Quality assessment:** The reliability of the grids was evaluated using allocation accuracy, comparing estimated populations against independent datasets from Italy, Portugal, Spain, and Belgium. The independent datasets included census data and mobile network operator data from Proximus in Belgium. This process allowed for an assessment of the accuracy of the methodology across different data types and geographical scales.

The study produced 24 population grids (12 months x 2 temporal frames: day and night) covering the 28 EU member states (as of 2019) at a 1 km² resolution. These grids represent a typical workday for each month, with nighttime grids assuming all individuals are at their place of residence, and daytime grids reflecting the distribution of individuals at their primary activity locations during core working hours. Analysis of the resulting data revealed significant differences in population distribution between day and night. In a three-dimensional rendering of Milan, substantial variations in population density were observed between day and night. For 34 major European cities (population > 1 million in 2011), daytime population density in city centers averaged 1.9 times higher than nighttime density, exhibiting an exponential decay with distance from the city center. The average day-to-nighttime population ratio across these cities was 1.097, with variations among cities reflecting differences in urban structures and activity patterns. A cluster analysis revealed different spatio-temporal population profiles among the cities, with some showing predominantly residential city centers, while others exhibited higher daytime densities in the center and surrounding areas. Validation using independent data sources from Italy, Portugal, and Spain showed high agreement with nighttime population data (99.2%-99.6% accuracy), reflecting the use of census data in the methodology. Daytime population grids exhibited a consistent allocation accuracy of approximately 93% in these countries. The comparison against mobile network operator data for Belgium revealed lower agreement (78-80%) due to the smaller spatial zones and different population concepts. However, the comparable accuracies for day and night suggest similar quality for both grids.

The findings address the research question by demonstrating the feasibility of generating accurate and consistent spatio-temporal population grids at a continental scale. The multi-layered dasymetric approach successfully integrated diverse data sources to overcome the limitations of using mobile phone data alone, creating a dataset richer in thematic detail. The high agreement between the generated grids and independent data in several EU countries validates the approach's accuracy. The observed patterns of daytime population density in large European cities confirm existing knowledge about the concentration of daytime activities within urban centers. The detailed spatio-temporal information provided by these grids offers significant advancements for diverse applications. This dataset is a valuable resource for researchers, policymakers, and practitioners, enabling improved analysis and planning across various sectors including urban planning, disaster risk management, environmental impact assessment, and transport planning.

This study provides the first EU-wide, high-resolution (1 km²) spatio-temporal population dataset, representing both daytime and nighttime distributions for each month of 2011. The multi-layered dasymetric approach successfully combined official statistics with diverse geospatial data, overcoming the limitations of existing methods. The dataset is publicly available and offers significant potential for advancing research and informing policy in various fields. Future work could focus on updating the dataset with more recent census data, incorporating additional demographic and socioeconomic attributes, and increasing temporal resolution by integrating data from mobile phone records. Further exploration of how these grids can inform applications in areas such as disaster risk management and urban planning should also be prioritized.

While the study presents a significant advance in spatio-temporal population mapping, some limitations exist. The reference year for population data is 2011, which could limit the immediate applicability for certain scenarios needing more recent data. The 1 km² resolution, chosen to balance detail and uncertainty, might be insufficient for applications requiring finer spatial granularity. The cross-comparison exercise, while providing validation, could not fully assess monthly variations due to a lack of sub-regional temporal data. Differences in population concepts between the generated grids and the mobile network operator data also affected the cross-comparison results.