High-Resolution dataset on elderly care facility accessibility and inequality in 21 Chinese cities (2020)

Abstract

Rapid population aging in China has created an urgent demand for equitable access to elderly care services, yet a notable data gap remains in quantifying and comparing accessibility across major cities. To address this, we present a comprehensive dataset on the spatial accessibility and inequality of elderly care facilities in 21 major Chinese cities circa 2020. The dataset was generated using the Gaussian Two-Step Floating Catchment Area (Ga2SFCA) method, which integrates facility capacity, population demand, and distance decay in travel behavior. Core data include 21 high-resolution (100 m) raster files measuring accessibility and 21 corresponding raster files (1 km) measuring spatial inequality using the Gini coefficient. Technical validation compared 4,099 model-derived network distances with commercial map service APIs, showing strong accuracy (R² > 0.94 for all cities). The full dataset, including raster and tabular files, is publicly available. This resource offers a foundation for research in urban planning, public health, transportation geography, and socioeconomics.

Introduction

Population aging is a defining global demographic trend with profound implications for economic and social development^1,2. In China, this trend is particularly acute; the population aged 60 and over reached 264 million (18.7% of the total) in 2020 and is projected to peak at 487 million by 2053³. This demographic shift has intensified the demand for community-based elderly care services. In response, the Chinese government has made substantial investments to expand the coverage of care facilities⁴. However, significant spatial disparities in the distribution and accessibility of these services persist, particularly within large urban areas, posing challenges to the well-being of older adults^5,6.

While the spatial inequality of elderly care is widely acknowledged, a critical data gap has hindered systematic, cross-city comparative analysis. High-resolution, methodologically consistent datasets are essential for quantitatively assessing accessibility, identifying underserved populations, and evaluating the effectiveness of policy interventions⁷. The absence of such data makes it difficult for researchers to conduct robust empirical studies and for policymakers to move from broad investment strategies to targeted, evidence-based planning^8,9.

To address this gap, we present a comprehensive dataset of elderly care facility accessibility and inequality for 21 major Chinese cities¹⁰. Using a validated geospatial modeling approach, this dataset provides a standardized, high-resolution snapshot of the urban service landscape circa 2020. The dataset is designed to be immediately useful for a wide range of interdisciplinary applications. It offers a quantitative measure of the spatial outcomes of decades of urban development, enables the precise identification of “service deserts” and “service islands,” and provides a robust baseline for future longitudinal studies^11,12. By making this dataset publicly available, we aim to facilitate new research into the complex interplay between urban form, public service provision, and social equality in one of the world’s most rapidly urbanizing regions¹³.

Methods

The dataset was generated using the Gaussian Two-Step Floating Catchment Area (Ga2SFCA) model to measure accessibility and the Gini coefficient to assess spatial inequality. The entire workflow was designed to be transparent and reproducible.

Data sources

Accessibility, defined as a quantitative expression of interaction opportunities among nodes within a transportation network, is typically measured by integrating factors such as local elderly population density, distribution and capacity of elderly care facilities (supply), and transportation infrastructure development^7,14. This study primarily employed relevant data on elderly care facilities, population, and road networks for the analysis.

Specifically, the elderly care facility data included location (latitude/longitude) and capacity (number of beds) for all registered elderly care facilities in the 21 cities for the year 2020, sourced from the National Platform for Social Service Information (Fig. 1) (https://mca.gjzwfw.gov.cn). Population data, comprising gridded estimates at a 100-meter spatial resolution with age-specific distributions for 2020, were sourced from the WorldPop open data portal¹⁵ (https://www.worldpop.org). A comprehensive digital road network for all 21 cities was extracted from the OpenStreetMap (OSM) project (https://www.openstreetmap.org), using a 2020 data snapshot¹⁶. This network formed the basis for all travel distance calculations. Additionally, municipal and district-level administrative boundaries were obtained from the National Geomatics Center of China (https://www.ngcc.cn). These administrative boundaries, specifically the built-up land area polygons derived from them, were utilized to constrain the analysis to populated urban regions.

Fig. 1.

Location of the 21 cities in China.

Accessibility measurement (Ga2SFCA)

Spatial accessibility was calculated using the Ga2SFCA method, an enhancement of the standard Two-Step Floating Catchment Area (2SFCA) model that incorporates a continuous Gaussian function for distance decay^3,16. This method was selected because its gradual decay function more realistically models the travel behavior of older adults, who often prefer local services but are not strictly limited by a binary distance cutoff^17,18. The Ga2SFCA model requires only the specification of a maximum search radius, minimizing subjectivity in parameterization¹⁸. However, since the model is based on the shortest-path distance along the road network, this setting primarily simulates travel conditions for motorized vehicles. We acknowledge that older adults in practice may rely more on walking, public transit, or non-motorized modes, for which impedance factors such as travel time, transfers, and walking distance differ from motorized network distance. The use of a unified road-network distance benchmark was adopted primarily to establish a methodologically consistent and comparable spatial accessibility assessment framework across the 21 cities, and to enable computation based on currently available, open, and comprehensive road network data.

The calculation proceeds in two steps. First, for each elderly care facility j, a supply-to-demand ratio (R_j) is computed within its catchment area, which is defined by a distance threshold l₀. This ratio is weighted by a Gaussian distance decay function:

$$R_j=\frac{S_j}{{\sum }_{k\in \left\{l_{jk}\in L_0\right\}}{D}_{k}g\left(l_{jk}\right)}$$

where S_j is the capacity of facility j (number of beds), D_k is the elderly population in grid cell k, l_kj is the shortest travel distance between facility j and grid cell k along the road network, and g(l_kj) is the Gaussian distance decay function:

$$\mathrm{g}\left(l_{ij}\right)=\frac{e^{-1/2\ast {(l_{ij}/\beta )}^2}-e^{-1/2\ast {(l_0/\beta )}^2}}{1-e^{-1/2\ast {(l_0/\beta )}^2}},l_{ij}\le L_0$$

In 2SFCA-based models, the form of the distance decay function directly influences the sensitivity of accessibility assessments. The Gaussian function decays relatively gradually over short to medium distances, which better captures the preference of older adults for nearby facilities; meanwhile, its rapid decay at longer distances aligns with the limited travel capacity typical of the elderly population.

The maximum search radius, l₀, was set based on facility size, following standards outlined by China’s Ministry of Civil Affairs to reflect hierarchical service ranges. The radii were set at 3 km for Small facilities, 5 km for Medium, 8 km for Large, and 10 km for Extra-large facilities, classified by bed capacity. A facility located near the official service threshold may still represent a viable option for many elderly individuals. Therefore, the bandwidth parameter β = l₀ was selected to avoid the cliff effect, a sharp drop in weight at the radius boundary, thereby better aligning with the perceptual tendency of older adults to consider facilities that are slightly farther but still potentially accessible. To verify the robustness of this parameter choice, three scenarios with β = 0.5 l₀, l₀, and 2 l₀ were compared. The results indicate that when β varies around l₀, the relative ranking of accessibility across cities remains stable.

In the second step, the accessibility index (A_i) for each population grid cell i is calculated by summing the weighted supply-to-demand ratios of all facilities j within the search radius l₀ from cell i:

$$A_i=\sum _{j\in (l_{ij}\le L_0)}{R}_{j}g\left(l_{ij}\right)=\sum _{j\in (l_{ij}\le L_0)}\frac{S_{j}g\left(l_{ij}\right)}{{\sum }_{k\in \left\{l_{jk}\in L_0\right\}}{D}_{k}g\left(l_{jk}\right)}$$

A higher value of A_i indicates better accessibility to elderly care facilities.

Spatial inequality measurement (Gini coefficient)

The Gini coefficient was employed to measure the inequality in the spatial distribution of accessibility¹⁹. This metric was chosen as it provides a standardized, scale-independent value ranging from 0 (perfect equality) to 1 (perfect inequality), allowing for robust comparisons across different cities^20,21. The Gini coefficient (G) was calculated based on the accessibility scores of the 100m-resolution grid cells, aggregated within a 1 km² grid to provide a localized measure of inequality²²:

$$G=1-\sum (y_i+y_{i-1})(x_i+x_{i-1})$$

where y is the cumulative proportion of accessibility and x is the cumulative proportion of grid cells, ordered by increasing accessibility.

Data processing workflow

The dataset was constructed through a sequential computational workflow encompassing data filtering and calibration, core metric calculation utilizing the Ga2SFCA method and Gini coefficient, and summary output generation. The specific methodological steps are detailed below.

Initial data preprocessing ensured analytical readiness: all facility locations were geocoded. Population data were projected to the appropriate local UTM zone for each city. The OSM road network was cleaned to ensure topological correctness for network analysis. Subsequently, network analysis was performed by constructing an origin-destination (OD) cost matrix. This critical step computed the shortest-path travel distance along the road network between the centroid of every 100-meter population grid cell and every elderly care facility.

Using these prepared datasets – the OD cost matrix, facility capacities, and population data – the accessibility calculation phase implemented the two-step Gaussian Two-Step Floating Catchment Area (Ga2SFCA) algorithm. This process generated high-resolution (100-meter) raster files quantifying elderly care facility accessibility for each of the 21 cities.

To assess spatial inequality, the resulting 100 m accessibility scores were aggregated within a standardized 1 km² grid. The Gini coefficient was then calculated for each 1 km² grid cell based on the distribution of accessibility values within its bounds, producing corresponding 1-km resolution inequality raster files. Finally, City-level summary statistics, such as the areal proportion of different accessibility tiers, were calculated for the tabular data files^23,24.

Data Records

The complete dataset is deposited in the Zenodo repository under a Creative Commons Attribution 4.0 International license²⁵. The dataset is organized into raster files describing spatial patterns and tabular files providing city-level summaries. A detailed description of all files is provided in Supplementary Table S1.

Spatial data files (Rasters)

The core of the dataset consists of 42 raster files in GeoTIFF (.tif) format. This includes 21 accessibility raster files quantifying elderly care facility accessibility scores and 21 corresponding spatial inequality raster files quantifying spatial inequality scores.

The accessibility raster files provide accessibility scores at a 100 m × 100 m spatial resolution. Pixel values are dimensionless, 32-bit floating-point numbers representing the Ga2SFCA accessibility index. Higher values indicate better accessibility. The inequality raster files provide Gini coefficients at a 1 km × 1 km spatial resolution. Pixel values are dimensionless, 32-bit floating-point numbers ranging from 0 to 1. Higher values indicate greater spatial inequality in accessibility within that 1 km grid cell. All raster files are projected in the WGS 84 datum with the appropriate UTM zone for each city’s location (e.g., UTM Zone 50 N for Beijing).

Tabular data files (CSV)

The dataset includes a summary table in Comma-Separated Values (.csv) format. This file, City_Summary_Statistics.csv, contains the data presented in Table 1, which summarizes the areal proportions of accessibility tiers and key economic and inequality indicators for each of the 21 cities.

Table 1.

Areal proportions of accessibility tiers and economic indicators for 21 major Chinese cities.

City	Per Capita GDP (yuan), 2020	Areal Proportion of Low-Accessibility Zones (%)	Areal Proportion of Medium-Accessibility Zones (%)	Areal Proportion of High-Accessibility Zones (%)
Harbin	54570	79.49	18.82	1.70
Dalian	70304	74.67	21.95	3.38
Chongqing	75828	71.19	25.13	3.68
Chengdu	84600	74.12	23.85	2.03
Kunming	85400	83.47	15.56	0.97
Xi’an	90203	90.33	7.91	1.76
Tianjin	90371	72.23	26.74	1.03
Dongguan	96501	83.37	15.36	1.27
Shenyang	97871	76.25	19.41	4.35
Foshan	108164	68.37	30.68	0.95
Jinan	110200	82.82	16.10	1.08
Zhengzhou	120030	80.11	16.69	3.20
Changsha	121425	77.02	18.47	4.51
Qingdao	124282	65.91	33.05	1.05
Guangzhou	133900	50.08	43.57	6.35
Wuhan	138800	70.48	26.85	2.67
Hangzhou	152465	70.41	27.22	2.38
Shanghai	157279	45.25	52.93	1.82
Shenzhen	157575	77.60	22.38	0.03
Beijing	164220	56.19	40.71	3.09
Nanjing	165681	53.80	41.61	4.59

The cities are ordered by ascending per capita GDP in 2020.

Data Overview

The data records reveal consistent and distinct spatial patterns that users of this dataset can explore. A dominant feature across all 21 cities is a pronounced core-periphery gradient structure in accessibility (Fig. 2). For example, in the Beijing accessibility raster (Beijing_Accessibility.tif), the highest index values are concentrated in the central Dongcheng and Xicheng districts, forming a contiguous high-accessibility zone. Moving outward from this core, the accessibility values systematically decline, with the lowest values found in the urban fringe and peri-urban areas. Similarly, the inequality rasters (Fig. 3) show a consistent pattern where core urban areas exhibit lower inequality (lower Gini coefficients), while peripheral zones demonstrate higher inequality (higher Gini coefficients). These clear spatial signatures provide a rich basis for quantitative analysis of urban structure and service provision.

Fig. 2.

Spatial distribution of accessibility to elderly care facilities in the study areas. Each panel (a–u) illustrates the detailed accessibility pattern for the corresponding city.

Fig. 3.

Spatial distribution of inequality in accessibility to elderly care facilities, measured by the Gini index. Each panel (a–u) depicts the specific pattern of inequality within the corresponding city.

Technical Validation

A rigorous validation process was implemented to ensure the reliability of the dataset. This focused on verifying the accuracy of the network distance calculations, which are a critical input to the accessibility model.

Validation of network distance calculation

An initial sample of 4,200 origin-destination (OD) pairs was generated by randomly sampling 200 origin points and 10 elderly care facilities as destinations within each of the 21 cities. After removing points in invalid areas (e.g., water bodies), a final validation set of 4,099 high-quality OD pairs was established. For each pair, the shortest travel distance calculated by our OSM-based network model was compared against the recommended driving distance obtained from a mainstream commercial map service API (Fig. 4).

Fig. 4.

Comparison of Shortest Travel Times Estimated via APIs and the OD Model Using Randomly Selected OD Samples at the City Level. (a–c) The shortest distance under real-time traffic conditions from Baidu, Gaode, and Tencent Map APIs versus the OD model estimates. (d) Discrepancy between API results and OD model estimates. (Illustrative comparison results for Beijing are shown).

Linear regression analysis was conducted for each city. The results, summarized in Table 2, show a strong, statistically significant positive linear correlation between the model-estimated distances and the API reference distances. The consistently high coefficients of determination (R² > 0.94, p < 0.01) across all 21 cities confirm that the underlying network analysis provides an accurate proxy for real-world travel distances. This robust agreement validates the primary spatial input for the Ga2SFCA model, ensuring the scientific reliability of the final accessibility and inequality data records.

Table 2.

Summary of Correlation Analysis for Technical Validation of Shortest Distance Estimates.

City	Number of Validated OD Pairs (n)	Coefficient of Determination (R2)	p-value
Harbin	195	>0.94	<0.01
Dalian	193	>0.94	<0.01
Chongqing	198	>0.94	<0.01
Chengdu	198	>0.94	<0.01
Kunming	196	>0.94	<0.01
Xi’an	195	>0.94	<0.01
Tianjin	196	>0.94	<0.01
Dongguan	193	>0.94	<0.01
Shenyang	194	>0.94	<0.01
Foshan	195	>0.94	<0.01
Jinan	192	>0.94	<0.01
Zhengzhou	198	>0.94	<0.01
Changsha	195	>0.94	<0.01
Qingdao	195	>0.94	<0.01
Guangzhou	193	>0.94	<0.01
Wuhan	198	>0.94	<0.01
Hangzhou	193	>0.94	<0.01
Shanghai	197	>0.94	<0.01
Shenzhen	195	>0.94	<0.01
Beijing	193	>0.94	<0.01
Nanjing	197	>0.94	<0.01

Discussion of model robustness

While the validation confirms the accuracy of the distance calculations, it is important for users to understand the model’s underlying assumptions. The Ga2SFCA model’s output is sensitive to the choice of the maximum search radius (l₀)^26,27. The radii used in this study are based on official Chinese government standards for facility service tiers, providing a strong, policy-relevant justification²⁸. However, actual travel behavior can be influenced by local culture, topography, and the availability of transportation options, which may cause localized deviations from these standardized radii^29,30. Furthermore, the population data represents a residential snapshot (circa 2020) and does not account for diurnal mobility patterns, which could affect demand distribution throughout the day^18,31. These factors represent potential sources of uncertainty and are important considerations for advanced modeling applications using this dataset.

Usage Notes

This dataset is designed to be a versatile resource for researchers and policymakers across multiple disciplines. The data are provided in standard formats (GeoTIFF, CSV) to ensure broad compatibility with GIS software (e.g., QGIS, ArcGIS) and statistical analysis packages (e.g., R, Python).

Potential research applications

The methodologically consistent, multi-city nature of this dataset enables a wide range of novel research applications³²:

Firstly, the dataset provides a quantitative baseline for 2020, against which the impact of future urban planning interventions can be measured. For example, researchers can assess how the construction of a new subway line or the targeted placement of new care facilities alters accessibility scores and inequality metrics over time. It can also be used to identify “elderly care deserts” to guide the optimal siting of new infrastructure³³.

Secondly, the high-resolution accessibility scores can serve as a critical exposure variable in epidemiological studies²⁶. Research questions could include investigating the spatial correlation between low accessibility to care and adverse health outcomes, such as hospitalization rates for chronic diseases, emergency service utilization, or mental health indicators among the elderly³⁴.

Thirdly, by integrating this dataset with socioeconomic data (e.g., housing prices, income levels, educational attainment), researchers can conduct powerful analyses of environmental justice and social stratification. For instance, one could quantitatively test the hypothesis that “service deserts” are disproportionately located in lower-income neighborhoods or areas with higher concentrations of migrant populations³⁵.

Finally, the standardized methodology applied across 21 cities facilitates robust cross-city comparative analyses. This allows researchers to explore how different municipal governance models, economic development levels, and urban forms (e.g., monocentric vs. polycentric) correlate with the provision and equality of essential public services for vulnerable populations³⁶.

Limitations of the dataset

Users should be aware of the following limitations when designing their analyses:

To begin with, the accessibility model is based on road network distance, which best represents travel by private vehicle or taxi. It does not explicitly model travel times or accessibility via public transportation (e.g., bus, metro), which is a primary mode of transport for many urban elderly residents. Future work could integrate General Transit Feed Specification (GTFS) data to create a multi-modal version of this dataset.

Furthermore, the model quantifies supply solely by the number of beds, treating all facilities as functionally equivalent. It does not capture important variations in service quality, specialization (e.g., dementia care, rehabilitation services), or cost, all of which are significant factors in real-world facility selection.

Additionally, demand is represented by the total elderly population count per grid cell. The model does not differentiate demand based on subgroups with varying needs or mobility constraints (e.g., “active” versus frail elderly) or by socioeconomic status, which can influence the ability to access services.

Ultimately, this is a cross-sectional dataset that provides a static snapshot of accessibility and inequality for the period around 2020. It does not capture the dynamic changes in urban infrastructure, population distribution, or service provision that occur over time.

Author contributions

X.H.: data collection and preprocessing, and writing – original draft; Y.W.: statistical analyses and visualization; Z.W.: data analyses; H.J. (corresponding author): conceived and designed the study, supervised the research process, writing – review & editing, and finalized the manuscript.

Data availability

The dataset described in this manuscript, including all 21 accessibility raster files, 21 inequality raster files, and summary tabular data, is publicly and freely available in the Zenodo repository under a Creative Commons Attribution 4.0 International license. The dataset can be accessed at: https://zenodo.org/records/17010522.

Code availability

All custom code used to process the primary data and generate the accessibility and inequality metrics presented in this study is available on GitHub (https://github.com/H97-art/A-high-resolution-spatial-accessibility-and-equity-dataset-code.git). The repository includes scripts written in Python and detailed documentation to ensure full reproducibility of our results.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41597-026-07014-8.