|

Awesome Geospatial Representation Learning

🌟A collection of papers, datasets, benchmarks, downstream tasks and codes for Geospatial Representation Learning Models. We are committed to consistently updating it to ensure it remains up-to-date and relevant.

📢 Latest Updates

• 2025.11: Latest update.

Related Surveys

• Self-supervised learning for geospatial ai: A survey. [paper]
  Information Fusion (2025)
• A review of location encoding for geoai: methods and applications. [paper]
  IJGIS (2022)
• Spatio-temporal graph neural networks for predictive learning in urban computing: A survey. [paper]
  IEEE TKDE (2023)
• GeoAI for large-scale image analysis and machine vision: recent progress of artificial intelligence in geography. [paper]
  ISPRS International Journal of Geo-Information (2022)
• Deep learning for spatio-temporal data mining: A survey. [paper]
  IEEE TKDE (2020)

Taxonomy Framework

We systematically review recent advancements in geospatial representation learning (i.e., geospatial embedding), analyzing their evolution and classifying methods into single-view, dual-view, and multiple-view approaches based on the complexity of views in learning geospatial representations

Data Modality Perspective

• An illustration of concepts including location, location embedding, region, and region embedding in geospatial representation learning.

• 

• The usage frequency of data modalities during learning stage across four categories in the survey.

• 

• The dataset usage frequency across cities / countries in relevant papers. We summarized the data separately by location and region.

• 

Methodology Perspective

Urban Region Representation Learning

Category	Abbreviation	Title	Publication	Paper	Modality (Coverage)	Methodology	Downstream Tasks	Code & Weights
Urban Region Representation Learning	HDGE	Region Representation Learning via Mobility Flow	CIKM2017	Paper	Mobility Data: Taxi data in Chicago Downstream Task Data: Demographics data, Crime data, House price data	graph, skip-gram	Crime rate prediction, Average personal income and house price prediction	-
	ZE-Mob	Representing Urban Functions through Zone Embedding with Human Mobility Patterns	IJCAI2018	Paper	Mobility New York	co-concurrence + word2vec	Identifying Functional Regions	-
	CDAE	Learning Urban Community Structures: A Collective Embedding Perspective with Periodic Spatial-temporal Mobility Graphs	ACM TITS2018	Paper	Mobility+POI data from Beijing	graph + AutoEncoder	Predicting Willingness to Pay Spotting Vibrant Urban Communities	-
	RegionEncoder	Unsupervised Representation Learning of Spatial Data via Multimodal Embedding	CIKM2019	Paper	Satellite Images+ Mobility Flow+ POI data. coordinates Chicago and NYC {Satellite Imagery: Google Static Maps API \ POI: Foursquare API \ Mobility: Open Data (Chicago and NYC)}	graph + Denoising AutoEncoder + Discriminator	Region Popularity Prediction House Price Prediction	Code
	-	Beyond Geo-First Law: Learning Spatial Representations via Integrated Autocorrelations and Complementarity	ICDM2019	Paper	Mobility + POI (Beijing) Mobility: A Beijing taxi company POI: A business review site in China (dianping.com)	Graph + Adversarial AutoEncoder (Inter+Intra)	house sale amount prediction	-
	Tile2Vec	Tile2Vec: Unsupervised Representation Learning for Spatially Distributed Data	AAAI2019	Paper	Satellite img America - USDA’s National Agriculture Imagery Program (NAIP) Cropland Data Layer (CDL)	Contrastive learning + triplet loss	Land cover classification using aerial imagery Predicting country health indices Poverty prediction Visual analogies	Code
	MV-PN	Efficient Region Embedding with Multi-View Spatial Networks: A Perspective of Locality-Constrained Spatial Autocorrelations	AAAI2019	Paper	POI + Mobility (Beijing) NYC open data Mobility: A Beijing taxi company POI: A business review site in China (dianping.com)	Multi-view graph + AutoEncoder	regional mobility popularity - check-in counts	Code
	-	Learning to Interpret Satellite Images in Global Scale Using Wikipedia	IJCAI2019	Paper	Satellite Images Wikipedia article Building Dataset: WikiSatNet, around the world, from wikipedia	DenseNet+Doc2Vec	classification temporal image classification Land Cover Classification Semantic Segmentation	Code
	CGAL	Unifying Inter-region Autocorrelation and Intra-region Structures for Spatial Embedding via Collective Adversarial Learning	KDD2019	Paper	POI + Mobility(Beijing) NYC open dataMobility: A Beijing taxi company POI: A business review site in China (dianping.com)	Multi-view Graph + Encoder-Decoder(GCN) + adversarial network	PREDICTING REGIONAL POPULARITY - the number of mobile check-in events	-
	-	Predicting Economic Development using Geolocated Wikipedia Articles	KDD2019	Paper	Wikipedia articles + Nightlight Images	Doc2Vec + CNN	Poverty Prediction, EDUCATION INDEX PREDICTION	Code
	-	Predicting Economic Growth by Region Embedding:A Multigraph Convolutional Network Approach	CIKM2020	Paper	demographic, social, economic and housing (America)	Multi-graph Convolution Network	Economic growth sector-by-sector predictions from ACS features of ZIP code areas	-
	Urban2Vec	Urban2Vec: Incorporating Street View Imagery and POIs for Multi-Modal Urban Neighborhood Embedding	AAAI2020	Paper	Street-view img (Google Street View Static API) + POI (Yelp Fusion API) (Bay Area + Chicago + New York)	Inception-V3 + Bag-of-Words	Predicting demographic and socioeconomic attributes	Code
	GMEL	Learning Geo-Contextual Embeddings for Commuting Flow Prediction	AAAI2020	Paper	urban indicators mobility NYC open data	Graph. (node: urban indicators)	commuting flow prediction	Code
	READ	Lightweight and Robust Representation of Economic Scales from Satellite Imagery	AAAI2020	Paper	Satellite img (Sparse labels)(South Korea) from DigitalGlobe	Mean Teacher + PCA	purchasing power estimation	Code
	-	Learning to Score Economic Development from Satellite Imagery	KDD2020	Paper	Imagenet on resnet18 for transfer learning Image： ESRI custom satellite images (South Korea, Malawi, and Vietnam) data: Census data (Population and Gross Floor Area). Nighttime satellite imagery (optional)	Clustering	Change detection Economic visual interpretation Application to developing countries	Code
	MVURE	Multi-View Joint Graph Representation Learning for Urban Region Embedding	IJCAI2020	Paper	mobility, Region Attributes (Check-in and POI)(New York)	multi-graph, including POI graph, check-in graph, source graph and destination graph	land usage classification and crime prediction	-
	-	Using publicly available satellite imagery and deep learning to understand economic well-being in Africa	Nature Communications 2020	Paper	Satellite img: Landsat (daytime+nighttime) Benin, Lesotho, Malawi, Rwanda, Sierra Leone, Senegal, Tanzania, and Zambia	Resnet-18 + asset wealth index	asset wealth estimation social protection program the relationship between temperature and wealth satellite-estimated wealth distribution	Code
	SceneParse	Predicting Livelihood Indicators from Community-Generated Street-Level Imagery	AAAI2021	Paper	mapillary data (geotagged SVIs)	Graph + cluster	Poverty Population Women’s BMI	Code
	M3G	Learning Neighborhood Representation from Multi-Modal Multi-Graph: Image, Text, Mobility Graph and Beyond	AAAI2021	Paper	SVIs, POI, mobility Chicago and New York City Street-view Imagery: Google Street View Static API \ POIs: Yelp Fusion API	Contrastive learning(Triplet loss)	Predicting Demographics and Economics Crime Prediction	Code
	HUGAT	Effective Urban Region Representation Learning Using Heterogeneous Urban Graph Attention Network (HUGAT)	arxiv2022	Paper	POI, mobility (trip), check-in activities, land usage distribution. Manhattan in NY	Supervised learning, multi-view graph: POIs, region, POI check-in time, hot hours for taxi departure, hot hours for taxi arrival	crime, average personal income, and bike flow prediction	-
	CcFTL	A Cross-City Federated Transfer Learning Framework: A Case Study on Urban Region Profiling	arxiv2022	Paper				-
	PG-SimCLR	Beyond the First Law of Geography: Learning Representations of Satellite Imagery by Leveraging Point-of-Interests	WWW2022	Paper	Satellite img + POI	Contrastive Learning + Attention (POI similar + geographically adjacent)	Region Similarity Analysis (socioeconomic indicators prediction)	Code
	-	Predicting Multi-level Socioeconomic Indicators from Structural Urban Imagery	CIKM2022	Paper	Satellite img + Street-view img Satellite Imagery: Eris; Street-view Imagery: Baidu Map	Graph + Contrastive learning	transfer to other cities of socioeconomic indicator prediction((a) Commercial Activeness. (b) Number of POIs. (c) Economic Activity. (d) Resident Consumption. (e) Population.)	-
	Region2Vec	Urban Region Profiling via Multi-Graph Representation Learning	CIKM2022	Paper	POI + mobility (taxi trip) NYC Open Data	multi-view graph, Encoder-Decoder (with attention)	region clustering, popularity prediction, crime prediction	-
	MGFN	Multi-Graph Fusion Networks for Urban Region Embedding	IJCAI2022	Paper	Custom dataset (Mobility) NYC Open Data	Graph, Cluster, Attention	crime count, check-in count, land usage type prediction.	Code
	URGENT	Urban Region Profiling With Spatio-Temporal Graph Neural Networks	IEEE TCSS 2022	Paper	POI, traffic-status (speed, flow, demand)（Beijing+Hangzhou） Mobility: GPS records in Beijing POI: GaoDe public map API (https://lbs.amap.com/)	Graph, Encoder-Decoder (Transformer)	traffic prediction	-
	ReMVC	Region Embedding with Intra and Inter-View Contrastive Learning	TKDE2022	Paper	POI + Mobility(New York)	Contrastive Learning (Inter and Intra view)	Land Usage Clustering Popularity Prediction	-
	-	Multi-modal Based Region Representation Learning Considering Mobility Data in Seoul	PCS2023	Paper	mobility, poi taxi trip subway ridership data District data Downstream Tasks: Socioeconomic data (South Korea) seoul	GAT, construct two region graph: trip + subway	the number of emplyees and mean house values prediction.	-
	KnowCL	Knowledge-infused Contrastive Learning for Urban Imagery-based Socioeconomic Prediction	WWW2023	Paper	POI mobility images (satellite images - street view images)(Beijing, Shanghai, New York)	Multi-view Graph, Contrastive Learning, ResNet spatial graph: POI + region mobility graph: flow function graph: check-in POI (similarity) business graph: POI	trasfer to other cities for socioeconomic prediction (population economy crime)	Code
	Geo-Tile2Vec	Geo-Tile2Vec: A Multi-Modal and Multi-Stage Embedding Framework for Urban Analytics	ACM TSAS 2023	Paper	SVIs POI mobility 5th Ring Road of Beijing, downtown of Nanjing, and downtown of Nanchang	Contrastive Learning Based on destination, constructing context corpus	main land use category classification task restaurant average price regression task firm number regression task main POI category classification task	-
	-	Point-to-Region Co-learning for Poverty Mapping at High Resolution Using Satellite Imagery	AAAI2023	Paper	satellite images Kisumu, Malindi and Nakuru (Kenya) - Sentinel - 2 labels (low-income label)	gate, attention, cnn	Poverty Prediction	-
	HREP	Heterogeneous Region Embedding with Prompt Learning	AAAI2023	Paper	POI + Mobility(New York)	Heterogeneous Graph: (node:region, edge: mobility (taxi trip) (source and target) + POI + geographic neighbor)	land usage classification	Code
	RegionDCL	Urban Region Representation Learning with OpenStreetMap Building Footprints	KDD2023	Paper	OSM building footprints + POI (Singapore + New York)	Contrastive learning + transformer	Land Use Population Density Estimation tranfer to other cities (Land Use and Population Density Estimation)	Code
	C-MPGCN	Learning Region Similarities via Graph-Based Deep Metric Learning	TKDE2023	Paper	POI mobility NYC open data	multi-graph based on distance categories	region similarity	-
	-	Urban visual intelligence: Uncovering hidden city profiles with street view images	PNAS2023	Paper	SVIs Downstream Tasks: POI, geographical features, population...（US）	LASSO regression, segmentation model	outperforms dynamic population and transfer to other cities crime, health, non-vehicle travel poverty, vehicle travel	Code
	HGI	Learning urban region representations with POIs and hierarchical graph infomax	ISPRS(JPRS)	Paper	Custom POI from baidumap (ShenZhen+Xiamen)	Manifold learning + Graph + Self-attention + Contrastive learning	estimating urban functional distributions, population density, and housing price.	Code
	MMGR	Geographic mapping with unsupervised multi-modal representation learning from VHR images and POIs	JPRS2023	Paper	very-high-resolution (VHR) image POI (Wuhan+Shanghai)	Contrastive learning + manifold learning + attention	-	Code
	ROMER	Region-Wise Attentive Multi-View Representation Learning for Urban Region Embeddings	CIKM2023	Paper	mobility, poi, check-in activities (New York)	Multi-view graph: mobility (Origin, Destination), POI, check-in graph (Dynamic Graph)	land usage classification task check-in prediction	-
	HAFusion	Urban Region Representation Learning with Attentive Fusion	ICDE2024	Paper	POI + Mobility + Land use (NewYork，Chicago，San Francisco, Propose additional datasets: Chicago, San Francisco）	multi-graph + self-attention	crime preiction check-in activity prediction	Code
	ReCP	Urban Region Embedding via Multi-View Contrastive Prediction	AAAI2024	Paper	POI + Mobility(New York)	Contrastive learning + autoencoder	Land Usage Clustering Region Popularity Prediction	-
	MuseCL	MuseCL: Predicting Urban Socioeconomic Indicators via Multi-Semantic Contrastive Learning	IJCAI2024	Paper	Satellite img + Street-view img + POI + Mobility (Beijing,Shanghai,New York) For street view imagery, we employ the Baidu Maps API1 for Beijing and Shanghai, and the Google Maps API2 for New York. High-resolution (3.6-meter) remote sensing images are acquired through ArcGIS for all three cities. The POI data for Beijing and Shanghai originates from Baidu Maps, while New York’s data is sourced from OpenStreetMap3 (OSM). Socioeconomic indicators, including population density from WorldPop4, housing data from Lianjia5, and crime data from NYC Open Data6, are also integrated.	Contrastive learning + skip-gram + attention fusion	-	Code
	CGAP	CGAP: Urban Region Representation Learning with Coarsened Graph Attention Pooling	IJCAI2024	Paper	POI mobility (taxi trip) (New York-NYC open data)	Graph + Self-attention	crime prediction check-in prediction	-
	UrbanCLIP	UrbanCLIP: Learning Text-enhanced Urban Region Profiling with Contrastive Language-Image Pretraining from the Web	WWW2024	Paper	Satellite img + Text (generated)(Beijing,Shanghai,Guangzhou,Shenzhen)	Contrastive learning + integrate text modality into urban region profiling	1. Carbon 2. GDP 3. Population	-
	UrbanVLP	UrbanVLP: Multi-Granularity Vision-Language Pretraining for Urban Region Profiling	arxiv2024	Paper	Satellite img + Street-view img + Text (generated)(Beijing,Shanghai,Guangzhou,Shenzhen)	macro (satellite) and micro (street-view) levels	-	-
	CityFM	City Foundation Models for Learning General Purpose Representations from OpenStreetMap	CIKM2024	Paper	OSM: POI+building shape+road segments OSM data (Node,Way,Relation)(tags as textual annotations) (traffic speed inference) (New York, Seattle) building functionality classification (Singapore) population density estimation (Singapore, New York)	Contrastive learning Visual: resnet18-->visual embedding Text: contrastive learning use POI Road: positional encoding	traffic speed inference;Building Functionality Classification Population Density Estimation	-
	MTE	Urban region representation learning with human trajectories: a multi-view approach incorporating transition, spatial, and temporal perspectives	GISR2024	Paper	Trajectory（Shenzhen） Custom dataset from anonymous mobile phone location data	graph + contrastive learning	Similar location search Land use classification Population density estimation land use classification	Code
	ReFound	ReFound: Crafting a Foundation Model for Urban Region Understanding upon Language and Visual Foundations	KDD2024	Paper	Satellite img + POI (beijing，shangha，guangzhou，shenzhen，suzhou)	Contrastive learning + knowledge distillation + self-attention	1. Urban Village Detection 2.Commercial Activeness Prediction 3.Population Prediction	-
	USPM	Profiling Urban Streets: A Semi-Supervised Prediction Model Based on Street View Imagery and Spatial Topology	KDD2024	Paper	Street labels SVIs +(text description generated) Downstream Task: Socioeconomic Indicator Prediction (Wuhan)	Contrastive learning + semi-supervised learning + graph	Street Function Prediction Socioeconomic Indicator Prediction	-
	-	secomanas2021seasonal	JATM2024	Paper	Street-view img + (Social media) Check-in (+ Road Network)(Wuhan)	urban street model: intra street + inter street;	Traffic estimation Traffic speed prediction Traffic volume prediction	Code
	GeoHG	Learning Geospatial Region Embedding with Heterogeneous Graph	arxiv2024	Paper	geo-entity segmentation + point-of-interest（POI）integration, heterogeneous graph	-	-	-
	EUPAC	Enhanced Urban Region Profiling with Adversarial Self-Supervised Learning for Robust Forecasting and Security	arxiv2024	Paper	mobility, POI, NYC open data	multi-graph: POI, mobility, geographical neighbor	Check-in and crime prediction. Land usage classification.	-
	-	Explainable Hierarchical Urban Representation Learning for Commuting Flow Prediction	SIGSpatial2024	Paper	grid: Road Densities POI grid Population' Railway Users 43 cities around the world.	graph + self-attention	Commuting Flows Reconstruction	-
	Demo2Vec	Demo2Vec: Learning Region Embedding with Demographic Information	GeoAI2024	Paper	population mobility (taxi trip) {NYC open data} POI	multi-view graph: Mobility (origin destination) POI, geographical neighbor, population GCN+ self-attention + attentive-fusion --> embedding	crime prediction housing price prediction check-in activity prediction	-

Location Embedding

Category	Abbreviation	Title	Publication	Paper	Modality (Coverage)	Methodology	Dataset	Coverage	Downstream Tasks	Code & Weights
Location Embedding	Place2Vec	From ITDL to Place2Vec – Reasoning About Place Type Similarity and Relatedness by Learning Embeddings From Augmented Spatial Contexts	ACM SIGSPATIAL 2017	Paper	POI (location + type)	ITDL augmented spatial context （distance binned+ information theory）	build dataset: Binary-based + Ranking-based	-	POI type	Code
	Loc2Vec	Loc2Vec:Learning location embeddings with triplet-loss networks	Blog2018	Paper	location	12-channel tensor, Spatial similarity + Self-supervised, Contrastive learning (triplet loss)	1.5 million locations	(data source：1 million locations that were visited by users on our platform, approximately 500,000 location fixes that were captured while users where in transport.)	visualization	Code
	Space2Vec	MULTI-SCALE REPRESENTATION LEARNING FOR SPATIAL FEATURE DISTRIBUTIONS USING GRID CELLS	ICLR2020	Paper	location + features(air quality, POI, elevation, mineral)	self-supervised, (Point Feature Encoder(air quality, POI, elevation, mineral)、Point Space Encoder(multi-scale)、 Location Decoder、Spatial Context Decoder)	open-source dataset published by Yelp Data Challenge (partial)	Las Vegas downtown area	POI type	Code
	GPS2Vec	GPS2Vec: Pre-Trained Semantic Embeddings for Worldwide GPS Coordinates	IEEE TMM2021	Paper	Geo-tagged image / Check-ins / Tweets + location	two-level grid-based framework	venue semantic annotation & the geotagged image classification: Foursquare check-in records, one million Flickr images, (evaluate using NUS-WIDE) next location prediction: -	Flickr: global, 145 000 distinct users in over 100 countries Foursquare check-in records: global, trajectory：regional, New York and Los Angeles	venue semantic annotation, the geotagged image classification, next location prediction	-
	GPS2Vec+	Learning Multi-context Aware Location Representations from Large-scale Geotagged Images	ACMMM2021	Paper	Geo-tagged image + location	multi-context fusion, location context(2-level grid) + visual context (RGB)	image classification: Flickr, (evaluate using NUS-WIDE) venue annotation: Foursquare check-ins	Flickr: global, 145 000 distinct users in over 100 countries Foursquare check-in records: global	venue semantic annotation, geotagged image classification	Code
	Geo-SSL	Geography-Aware Self-Supervised Learning	ICCV2021	Paper	Satellite image + location (as supervision signal)	self-supervised, Contrastive learning, temporal positive pairs	fMoW, GeoImageNet	fMoW: global GeoImageNet: global,	geotagged image classification (Classifying Single Images + Classifying Temporal Data)	Code
	TALE	Pre-Training Time-Aware Location Embeddings from Spatial-Temporal Trajectories	TKDE2022	Paper	location + time(Trajectories)	Spatial-Temporal, hierarchical tree structure	Foursquare-NYC (Foursquare check-ins in New York), Foursquare-TKY (Tokyo) and Foursquare-JKT (Jakarta), Mobile-PEK(mobile phone signaling data in Beijing)	New York, Tokyo and Jakart, Mobile-PEK: Beijing	Location Classification, Location Visitor Flow Prediction, Next Location Prediction	Code
	-	Pre-training Contextual Location Embeddings in Personal Trajectories via Efficient Hierarchical Location Representations	ECML PKDD2023	Paper	location + time(Trajectories)	Geo-Tokenizer (Multi-scale), Hierarchical Auto-regressive ( incorporated information from the lower-level hierarchies into the upper-level hierarchies)	Mobile-T, Geo-Life	Mobile-T:(collected by the base stations of the major cellular network operator） Geo-Life：182 users over a period of five years in Microsoft Research (Asia)	Next Location Prediction, LandUsage Classification, Transportation Mode Classification	-
	MGeo	MGeo: Multi-Modal Geographic Language Model Pre-Training	SIGIR2023	Paper	location + graphic object	Geographic Encoder (contains features of the surrounding geographic objects)	GeoTES (Geographic Textual Similarity)	GeoTES: (Hangzhou) POI数据： obtained from the open-source OSM Queries: manually generated by annotators	Query-POI Matching (Ranking task + Retrieval task )	Code
	Sphere2Vec	Sphere2Vec: Multi-Scale Representation Learning over a Spherical Surface for Geospatial Predictions	ISPRS2023	Paper	Geo-tagged image + location	encodes point coordinates on a spherical surface	BirdSnap, BirdSnap+, Nabird+，iNat2017, iNat2018	iNat2017: species classification, 5089 categories iNat2018: species classification, 8142 categories BirdSnap: no metadata, 500 bird species BirdSnap+: supplemental metadata, 500 bird species NABirds+: supplemental metadata,555 bird species	geotagged image classification	Code
	CSP	CSP: Self-Supervised Contrastive Spatial Pre-Training for Geospatial-Visual Representations	ICML2023	Paper	Satellite image / Geo-tagged image + location	self-supervised, Contrastive learning	iNat2018, fMoW	iNat2018: species recognition, 8142 categories fMoW: Satellite Image Classification, over 1 million images from over 200 countries	Species Recognition, Satellite Image Classification	Code
	GeoCLIP	GeoCLIP: Clip-Inspired Alignment between Locations and Images for Effective Worldwide Geo-localization	NIPS2023	Paper	Geo-tagged image + location	contrasive learning, CLIP(Geo-tagged image - location), Random Fourier Features	training: MediaEval Placing Tasks 2016 (MP-16), testing: Im2GPS3k, Google World Streets 15K dataset (GWS15k), YFCC26k	MP-16: 4.72 million geotagged images from Flickr Im2GPS3k: 3000 images from Im2GPS dataset GWS15k: Global, (15,000 images from Google Street Map) YFCC26k:26,000 images random selected from YFCC100M dataset	image-to-GPS，limited data，Qualitative Results	Code
	SatCLIP	SatCLIP: Global, General-Purpose Location Embeddings with Satellite Imagery	arxiv2023	Paper	Satellite image + location	contrasive learning, CLIP(Satellite image - location)	pretrain: S2-100K Downstream Tasks： 1) Air Temperature 2) Elevation 3) Median Income: U.S. election maps 4) California Housing: California Housing 5）Population Density 6） Biomes, Ecoregions 7) Countries 8) iNaturalist: iNat2017	S2-100K: 100,000 multi-spectral satellite images and their corresponding locations (sampled approximately uniformly over landmass) 2）Elevation：obtained from the Google Static Maps API 3） U.S. election maps: including median income attribute 4) California Housing: obtained from the StatLib repository (20,640 observations on 9 variables) 6) Biomes, Ecoregions：global (all 846 terrestrial ecoregions, 14 terrestrial biomes) 7) Countries: self-built dataset 8) iNat2017: species classification, 5089 categories	Regression (Air Temperature, Elevation, Median Income, California Housing, Population Density), Classification(Countries, iNaturalist, Biome, Ecoregions)	Code
	GeoLLM	GEOLLM: EXTRACTING GEOSPATIAL KNOWLEDGE FROM LARGE LANGUAGE MODELS	ICLR2024	Paper	OSM + text prompt (Address + Nearby Places)	LLM, Address + Nearby Places ---> prompt	1) Population: WorldPop 2) AssetWealth: DHS 3) Women Edu: DHS 4) Sanitaiton: DHS 5) Women BMI: DHS 6) Population: USCB 7) Mean Income: USCB 8) Hispanic Ratio: USCB 9) Home Value: Zillow	1) WorldPop: global 2) DHS: ( Demo graphic and Health Surveys) across 48 countries 3) USCB: (United States Census Bureau), America only 4) Zillow: (Zillow Home Value Index) only covers the US	Regression (Population, Asset Wealth, Women Edu, Sanitation, Women BMI, Population, Mean Income, Hispanic Ratio, Home Value)	Code
	LLMGeovec	Geolocation Representation from Large Language Models are Generic Enhancers for Spatio-Temporal Learning	AAAI 2025	Paper	location+OSM data (addresses,nearby places)	LLM, Address + Nearby Places ---> prompts	1) AnnualAirTemperature: Chelsa 2) Annual Precipitation: Chelsa 3) Monthly Climate Moisture: Chelsa 4) Population Density: WorldPop 5) Nighttime Light Intensity: EOG 6) Human Modification Terrestrial: SEDAC 7) Global Gridded Relative Deprivation: SEDAC 8) Ratio of Built-up Area to Non-built Up Area: SEDAC 9) Child Dependency Ratio: SEDAC 10) Subnational Human Development: SEDAC 11) Infant Mortality Rates: SEDAC 12) Asset Index: DHS 13) Sanitation Index: DHS 14) Women BMI: DHS 15) Poverty Rate: DHS (Country) 16) Population Density: FaceBook(Country) 17) Women BMI: DHS(Country) 18）Population Density: NYC 19) Education Level: NYC 20) Income Level: NYC 21) Crime Rate: NYC	1) Chelsa: Global 2) WorldPop：Global 3）EOG：Global 4) SEDAC: Global 5) DHS: Global 6) NYC: City (train1k, test424)	geographic prediction, long-term time series forecasting, graph-based spatio-temporal forecasting	-

Related Links

• Awesome-Spatio-Temporal-Representation-Learning: Summary of Spatio-Temporal Representation Learning Models

• Awesome-Multimodal-Urban-Computing

• Awesome-Urban-Foundation-Models