Research on multiscale OpenStreetMap in China: data quality assessment with EWM-TOPSIS and GDP modeling

Figures & data

Figure 1. Number of registered OSM users (https://wiki.openstreetmap.org/wiki/Stats).

Figure 2. Technical framework of this study.

Table 1. OSM data quality evaluation index system.

OSM data quality evaluation index system	Index layer	Index factor	Index type
	Completeness	Road network length completeness	+
		Relative linear density	-
	Attribute accuracy	Road name attribute completeness	+
		POI name attribute completeness	+
	Position accuracy	Road network accuracy	+
	Temporal accuracy	Road network update frequency	+
		POI update frequency	+
	Logical consistency	Topological consistency	-
		Directional similarity	-

“+” means that higher values are indicative of higher quality, and “-” means that lower values are indicative of higher quality.

Figure 3. Position accuracy evaluation.

Figure 4. (a) The vector boundary of the study area, (b) OSM road network data in 2020, (c) OSM POI data in 2020.

Table 2. Corresponding relationships between OSM data and the reference data.

OSM tags	Reference dataset
Motorway, Motorway_link	Expressway
Trunk, Trunk_link	National roads
Primary, Primary_link	Provincial roads
Secondary, Secondary_link	County roads
Tertiary, Tertiary_link, Service, Living_street, Residential	Township roads
Unclassified, Track, Track_grade1–5	Village roads

Table 3. The weights of OSM data quality indicators at different scales.

Index factor	National	Provincial	Municipal	Grid
Road network length completeness	9.45%	18.48%	16.74%	15.37%
Relative linear density	9.78%	5.79%	0.79%	0.05%
Road name attribute completeness	11.26%	10.60%	10.89%	10.00%
POI name attribute completeness	7.52%	7.99%	11.38%	13.01%
Road network accuracy	10.11%	8.44%	9.65%	45.40%
Road network update frequency	12.73%	21.70%	22.40%	6.31%
POI update frequency	20.25%	19.75%	24.42%	9.70%
Topological consistency	12.32%	1.68%	0.47%	0.14%
Directional similarity	6.59%	5.58%	3.26%	0.02%

Figure 5. The weights of quality dimensions at different spatial scales.

Figure 6. OSM data quality trends at the national scale in China from 2014 to 2020.

Table 4. OSM data quality at the national scale in China from 2014 to 2020.

Year	Road network length completeness	Relative linear density	Road name attribute completeness	POI name attribute completeness	Road network accuracy	Road network update frequency	POI update frequency	Topological consistency	Directional similarity	T
2014	21.73%	36.39%	20.79%	44.71%	76.37%	47.83%	47.94%	49.41	2.01	51.96%
2015	29.66%	33.54%	20.18%	48.17%	72.38%	35.35%	33.79%	45.55	1.06	44.54%
2016	35.85%	31.38%	20.54%	53.18%	72.44%	29.64%	35.57%	46.89	0.36	54.95%
2017	41.36%	29.16%	20.04%	54.97%	71.88%	29.65%	31.67%	32.37	0.10	57.44%
2018	49.62%	25.43%	19.07%	55.13%	69.39%	23.93%	20.81%	34.54	0.02	46.68%
2019	54.85%	23.58%	18.87%	53.54%	68.75%	21.12%	22.05%	32.87	0.06	47.55%
2020	57.76%	22.87%	18.56%	51.04%	66.90%	29.89%	20.86%	29.03	0.05	50.21%

Figure 7. OSM data quality at the provincial scale in China from 2014 to 2020.

Figure 8. OSM data quality at the municipal scale (taking provincial capitals as an example) in China from 2014 to 2020.

Figure 9. Spatial distribution of OSM data quality at the provincial scale in China in (a) 2014, (b) 2015, (c) 2016, (d) 2017, (e) 2018, (f) 2019, and (g) 2020.

Figure 10. Spatial distribution of OSM data quality at the municipal scale in China in (a) 2014, (b) 2015, (c) 2016, (d) 2017, (e) 2018, (f) 2019, and (g) 2020.

Figure 11. OSM data quality in six geographical regions in China at the (a) provincial scale and (b) municipal scale.

Figure 12. Spatial distribution of OSM data quality at the grid scale in China in (a) 2014, (b) 2015, (c) 2016, (d) 2017, (e) 2018, (f) 2019, and (g) 2020. (h) OSM data quality in six geographical regions in China at the grid scale.

Table 5. Global spatial autocorrelation results based on Moran’s I values from 2014 to 2020.

Spatial scale	Provincial			Municipal			Grid
	$I_{\mathit{Global}}$	Z-Score	P-Value	$I_{\mathit{Global}}$	Z-Score	P-Value	$I_{\mathit{Global}}$	Z-Score	P-Value
2014	0.081	1.503	0.133	0.249	7.697	0.000	0.258	27.060	0.000
2015	0.166	2.607	0.009	0.141	4.834	0.000	0.103	12.724	0.000
2016	0.277	2.949	0.003	0.115	3.891	0.000	0.308	32.944	0.000
2017	0.238	2.525	0.012	0.184	5.816	0.000	0.317	34.897	0.000
2018	0.203	2.178	0.029	0.104	3.346	0.000	0.331	35.054	0.000
2019	0.207	2.209	0.027	0.211	6.578	0.000	0.332	35.682	0.000
2020	0.440	4.325	0.000	0.249	7.782	0.000	0.382	40.206	0.000
Mean	0.230			0.179			0.290

Figure 13. Local spatial autocorrelation distribution of OSM data quality at the provincial scale in (a) 2014, (b) 2015, (c) 2016, (d) 2017, (e) 2018, (f) 2019, and (g) 2020.

Figure 14. Local spatial autocorrelation distribution of OSM data quality at the municipal scale in (a) 2014, (b) 2015, (c) 2016, (d) 2017, (e) 2018, (f) 2019, and (g) 2020.

Figure 15. Local spatial autocorrelation distribution of OSM data quality at the grid scale in (a) 2014, (b) 2015, (c) 2016, (d) 2017, (e) 2018, (f) 2019, and (g) 2020.

Table 6. Statistics of local spatial autocorrelation for different spatial agglomeration types and scales from 2014 to 2020.

Spatial scale	Spatial agglomeration type	2014	2015	2016	2017	2018	2019	2020
Provincial	HH	4	4	4	5	6	5	8
	HL	0	0	0	1	1	0	1
	LH	3	3	2	2	1	2	2
	LL	2	2	3	2	2	2	2
Municipal	HH	41	42	27	32	31	42	52
	HL	19	17	10	5	8	10	10
	LH	22	15	16	29	17	35	28
	LL	49	70	35	63	52	44	73
Grid	HH	145	120	184	249	189	148	201
	HL	30	46	44	43	42	43	44
	LH	77	49	41	37	27	20	33
	LL	272	348	392	423	402	372	398

Table 7. Collinearity statistical results.

Spatial scale	Independent variable	VIF	Spatial scale	Independent variable	VIF
Provincial	Road network length completeness	6.247	Municipal	Road network length completeness	1.743
	Relative linear density	5.758		Relative linear density	1.812
	Road name attribute completeness	2.506		Road name attribute completeness	1.206
	POI name attribute completeness	1.579		POI name attribute completeness	1.179
	Road network accuracy	1.925		Road network accuracy	1.168
	Road network update frequency	1.592		Road network update frequency	1.291
	POI update frequency	1.369		POI update frequency	1.283
	Topological consistency	1.369		Topological consistency	1.081
	Directional similarity	1.771		Directional similarity	1.074

Table 8. OSM data quality and GDP modeling results at the provincial and municipal scales in China.

Spatial scale	Model	R^2	Adjusted-R^2	AICc	RSS
Provincial	GWR	0.8475	0.8106	−85.9606	5.9436
	GTWR	0.8953	0.8585	−130.2082	4.0824
Municipal	GWR	0.8775	0.8518	−890.0355	63.2534
	GTWR	0.9063	0.8753	−1066.2704	48.3905

Figure 16. Local R² spatial distribution of the GTWR model at the (a) provincial scale and (b) municipal scale.

Figure 17. Spatial distribution of regression coefficients for the GTWR model at the municipal scale.

Table 9. Comparison of the accuracy of the GDP simulation results of different machine learning models.

Machine learning model	Dataset	R^2	RMSE	MAE
SVM	Training set	0.5926	0.9097	0.5561
	Test set	0.6286	0.7512	0.5030
RF	Training set	0.7902	0.7717	0.4521
	Test set	0.7472	0.6230	0.7902
XGBoost	Training set	0.9314	0.5770	0.2429
	Test set	0.7716	0.6278	0.4167
CatBoost	Training set	0.8768	0.6696	0.3226
	Test set	0.8152	0.5582	0.3961

Figure 18. (a) DEM of six geographical regions in China. (b) Correlation between OSM data quality and DEM at different spatial scales.

Data availability statement

The OpenStreetMap data were sourced from the official website of OSM (https://www.openstreetmap.org/). Reference road vector data were obtained from Baidu Maps (https://map.baidu.com/). The vector boundary of the research area was obtained from the Chinese Resource and Environment Science and Data Center (https://www.resdc.cn/). The actual road mileage data, Provincial and municipal GDP were obtained from the National Bureau of Statistics of China (http://www.stats.gov.cn/) and the China City Statistical Yearbook (https://www.yearbookchina.com/).