Towards POI-based large-scale land use modeling: spatial scale, semantic granularity, and geographic context

Figures & data

Figure 1. Spatial distribution of POIs from three selected geographic regions (a) South Africa (b) Facebook (c) 4square (d) Here (e) OSM (f) Google (g) TomTom (h) Vkonta (i) Wikimap (j) South Korea (k) England.

Figure 2. Framework for AOI embedding learning based on POI.

Table 1. Example semantic translation from POI category to OSM tags.

Category	Data source	osm tags
Mexican Restaurant	Facebook	amenity = restaurant; cuisine = mexican; building = retail
Yucatecan Restaurant	Foursquare
Mexican cuisine	Wikimapia

Figure 3. Hierarchical space partition of three study regions.

Table 2. Hierarchical space partition-based road network.

Level	Road network level
Level 1	road segments that are higher than tertiary level roads, including motorway, primary,trunk, secondary, and tertiary
Level 2	Level 1 + unclassified level roads
Level 3	Level 2 + residential level roads
Level 4	Level 3 + all the lower level roads

Table 3. Spatially explicit POI embedding results for different geographic regions.

	South Korea	South Africa	England
bar	biergarten nightclub pub internet_cafe cafe	pub nightclub biergarten cafe restaurant	pub nightclub restaurant biergarten cafe
college	campus high_school university humanities education	high_school campus nursery preschool school	adult education rnib campus school university
car_rental	fuel bicycle rental hertz rent car_repair	fuel restaurant truck_rental limo_service bicycle_rental	car_wash nightclub post_depot fuel cinema
clinic	doctors hospital yueoseu orakeul dentist	doctors library arts_centre post office building	arts_centre library theatre police fire_station
casino	cinema bar internet_cafe food_court nightclub	nightclub restaurant bar cinema pub	gambling nightclub cinema bar bicycle_rental
arts_centre	cultural_centre theatre animal_shelter planetarium events_venue	theatre library building cultural_centre police	theatre events_venue music_venue library cultural_centre
court_house	fire_station police policy closest_town public_bath	townhall police prison post_office fire_station	townhall library police arts_centre events_venue
playground	park pitch fitness_centre sports_centre garden	park garden common golf_course pitch	park common picnic_table pitch nature_reserve
grave_yard	bongeunsa hakdong seoripul nonhyeon eonju	church taxi tennis_court garden_centre convenience	church tennis_court grit_bin watering_place hunting_stand
mall	cosmetics shoes boutique jewelry games	general supermarket video mobile_phone shoes	variety_store houseware frozen_food electronics catalogue
church	cathedral denomination deokposijang christian chapel	religious religion service times temple denomination	chapel grave_yard kindergarten subdenomination childcare
industrial	warehouse man_made sagimakgol chemical wind_energy	mechanic solar panel antenna winery engineering	warehouse works crescnt science_park knowlhill factory

Table 4. Grouping of OSM land use classes (13 classes to 3 classes).

Grouped class	Original class
‘open space’	‘natural’ ‘agricultural’
‘residential’	‘residential’
‘non-residential’	‘industrial’ ‘retail’ ‘commercial’ ‘recreation’ ‘transportation’ ‘cemetery’ ‘religious’ ‘military’ ‘civic’ ‘utility’

Table 5. F-score of three-class land use classification models.

	South Africa			South Korea			England
	non-resi	open space	resi	non-resi	open space	resi	non-resi	open space	resi
all sample, no agg, poi	0.87	0.71	0.61	0.83	0.7	0.68	0.8	0.77	0.6
filtered sample, no agg, poi	0.9	0.61	0.71	0.84	0.71	0.74	0.81	0.8	0.74
all sample, agg30, poi	0.82	0.56	0.6	0.74	0.54	0.55	0.81	0.56	0.62
filtered sample, agg30, poi	0.8	0.37	0.63	0.76	0.51	0.63	0.8	0.56	0.73
all sample, no agg, all feature	0.84	0.9	0.68	0.84	0.78	0.79	0.72	0.82	0.68
filtered sample, no agg, all feature	0.87	0.82	0.76	0.84	0.78	0.82	0.76	0.77	0.76
all sample, agg30, all feature	0.78	0.71	0.64	0.82	0.67	0.72	0.71	0.63	0.68
filtered sample, agg30, all feature	0.82	0.66	0.74	0.79	0.61	0.76	0.74	0.6	0.73

Figure 4. Confusion matrix for 13-class classification models (a) South Africa (b) South Korea (c) England.

Table 6. Classification performance of land use classes of 13-class model.

	South Africa		South Korea		England
	Precision recall		Precision recall		Precision recall
agricultural	0.88	0.78	0.69	0.59	0.58	0.52
cemetery	0.54	0.58	0.37	0.09	0.66	0.33
civic	0.85	0.9	0.79	0.86	0.82	0.74
commercial	0.4	0.23	0.51	0.33	0.4	0.14
industrial	0.58	0.55	0.8	0.58	0.6	0.45
military	0.62	0.04	0.65	0.29	0.31	0.03
natural	0.56	0.59	0.7	0.73	0.6	0.69
recreation	0.65	0.65	0.69	0.69	0.56	0.4
religious	0.8	0.68	0.89	0.72	0.67	0.53
residential	0.71	0.81	0.8	0.88	0.74	0.87
retail	0.82	0.91	0.65	0.67	0.7	0.77
transportation	0.84	0.7	0.78	0.67	0.71	0.53
utility	0.93	0.86	0.37	0.09	0.68	0.32

Table A1. Grouping of OSM land use classes.

Grouped class	Original class
agricultural	agricultural allotments animal keeping aquaculture farmland garden greenhouse greenhouse horticulture orchard pasture plant nursery vineyard
cemetery	cemetery grave_yard grave graveyard
civic	civic civic_admin community community_centre community_centre conference_centre government governmental institutional library service social_facility townhall clinic doctors health hospital school college university
natural	natural bare bare_rock basin beach conservation field flowerbed forest forestry grass grassland greenfield meadow nature nature_reserve reservoir sand scree scrub village_green water wetland wood
industrial	industrial brownfield construction depot landfill plant quarry
recreation	recreation camp_site caravan_site common dog park fishing golf golf_course holiday park leisure park pitch playground recreation_ground resort shooting_ground sport sports sports_centre stadium swimming theme_park water_park zoo observatory
religious	religious church churchyard place_of_worship monastery
residential	residential apartments village
retail	retail mall market marketplace
transportation	aeroway^a public transport^a railway highway

^a aeroway and public_transport are OSM keys. OSM polygons that are tagged with aeroway = or public_transport = will also be included in the samples.

Data availability statement

The data that support the findings of this study are available from PlanetSense project. Restrictions apply to the availability of these data, which were used under license for this study. Data are available from the authors with the permission of PlanetSense project (https://planetsense.ornl.gov/webps/). The OSM data used in this study can be openly accessed at https://www.openstreetmap.org/export.