pith. sign in

arxiv: 2605.26036 · v1 · pith:JKIX5S63new · submitted 2026-05-25 · 💻 cs.AI · cs.LG

CITYREP: A Unified Benchmark for Urban Representations Across Cities, Tasks, and Modalities

Junyuan Liu , Xinglei Wang , Zichao Zeng , Jiazhuang Feng , Quan Qin , Ilya Ilyankou , Guangsheng Dong , Tao Cheng This is my paper

Pith reviewed 2026-06-29 21:44 UTC · model grok-4.3

classification 💻 cs.AI cs.LG
keywords urban representation learningbenchmarkspatial splitscross-city generalizationmodel evaluationurban foundation modelsspatial leakage
0
0 comments X

The pith

Urban representation evaluations using random splits overestimate model performance and fail to support cross-city generalization.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces CityRep to standardize how urban embeddings are tested across cities, tasks, and data modalities by replacing random train-test splits with block-based spatial splits. It shows that random splits create spatial leakage, which inflates scores and changes which models rank highest. When 11 models are run on 8 cities and 8 tasks spanning regression, classification, and distribution prediction, results vary sharply by location and task. This matters because urban foundation models are meant to work in new places, yet current tests do not check for that. The benchmark supplies datasets, pipelines, and tools so future work can compare models on equal terms.

Core claim

CityRep supplies a spatial unit-agnostic alignment module, a block-based spatial split protocol that removes leakage, and an extensible suite covering 8 cities and 8 tasks; when 11 models are evaluated under this protocol, scores drop and rankings shift relative to random splits, while performance also differs markedly across cities and tasks.

What carries the argument

Block-based spatial splits that partition cities into contiguous blocks for train-test division, preventing spatial leakage while remaining compatible with heterogeneous urban data representations.

If this is right

  • Random splits produce higher scores than block-based spatial splits.
  • Model rankings reverse or shift when spatial splits replace random ones.
  • Performance differs substantially from city to city and task to task.
  • Generalization-aware evaluation protocols are required for credible claims about urban representations.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Model designers may need to add explicit spatial-awareness constraints during training to maintain rankings under block splits.
  • The benchmark could be used to identify which data modalities transfer most reliably across cities.
  • Real-world urban applications would benefit from testing new models on held-out cities before deployment.

Load-bearing premise

The choice of eight cities, eight tasks, and block-based splits is broad enough to support claims about cross-location generalization without creating its own selection biases or coverage gaps.

What would settle it

If additional experiments on new cities using both random and block splits produce identical scores and unchanged model rankings, that would show the claimed sensitivity does not hold.

Figures

Figures reproduced from arXiv: 2605.26036 by Guangsheng Dong, Ilya Ilyankou, Jiazhuang Feng, Junyuan Liu, Quan Qin, Tao Cheng, Xinglei Wang, Zichao Zeng.

Figure 1
Figure 1. Figure 1: Framework of CityRep Benchmark. CityRep standardizes the evaluation of heteroge￾neous urban representations by aligning different spatial supports to common downstream task units, evaluating them across eight cities and eight tasks, and using spatial block splits to mitigate leakage. representation transfers across urban contexts or across qualitatively different prediction problems. This limitation is esp… view at source ↗
Figure 2
Figure 2. Figure 2: GDP and LST prediction performance across cities. The figure reports R2 , MAE, and RMSE for each model and city. competitive on LST. These results indicate that benchmark performance is shaped not only by input modality, but also by model architecture, pretraining objective, spatial support, and alignment strategy. 4.2 Performance across Cities and Tasks The final row of [PITH_FULL_IMAGE:figures/full_fig_… view at source ↗
Figure 4
Figure 4. Figure 4: Raw primary-metric changes under random splits compared with spatial splits. Each cell reports random split minus spatial split for the primary evaluation metric. Positive values indicate higher raw metric values. gaps among models change across cities. Thus, results from a single city would give an incomplete picture of model quality [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Land-use (LUC) labels across the eight benchmark cities. [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Road-density (RDE) labels across the eight benchmark cities. [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Population (POP) labels across the eight benchmark cities. [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Age-distribution (AGE) labels across the eight benchmark cities. [PITH_FULL_IMAGE:figures/full_fig_p017_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Gross Domestic Product (GDP) labels across the eight benchmark cities. [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Nighttime lights (NTL) labels across the eight benchmark cities. [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: PM2.5 labels across the eight benchmark cities [PITH_FULL_IMAGE:figures/full_fig_p018_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Land-surface-temperature labels across the eight benchmark cities. [PITH_FULL_IMAGE:figures/full_fig_p018_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Example 10-by-10 spatial block split assignments for AETHER under seed 42. Blocks, [PITH_FULL_IMAGE:figures/full_fig_p022_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Test-block frequency across the five spatial split seeds. The figure shows how often each [PITH_FULL_IMAGE:figures/full_fig_p023_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Task-level rank changes between random and spatial splits by model and task. Negative [PITH_FULL_IMAGE:figures/full_fig_p024_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Overall mean-rank change between random and spatial splits across tasks. Negative values [PITH_FULL_IMAGE:figures/full_fig_p024_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Spatial-split primary metric values and rank diagnostics for all evaluated models. [PITH_FULL_IMAGE:figures/full_fig_p030_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Random-split primary metric values and rank diagnostics for all evaluated models. [PITH_FULL_IMAGE:figures/full_fig_p030_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Land-use classification: full metric bars under spatial block splits. [PITH_FULL_IMAGE:figures/full_fig_p031_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Land-use classification: full metric bars under random splits. [PITH_FULL_IMAGE:figures/full_fig_p031_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Road-density regression: full metric bars under spatial block splits. [PITH_FULL_IMAGE:figures/full_fig_p032_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: Road-density regression: full metric bars under random splits. [PITH_FULL_IMAGE:figures/full_fig_p032_22.png] view at source ↗
Figure 23
Figure 23. Figure 23: Population regression: full metric bars under spatial block splits. [PITH_FULL_IMAGE:figures/full_fig_p032_23.png] view at source ↗
Figure 24
Figure 24. Figure 24: Population regression: full metric bars under random splits. [PITH_FULL_IMAGE:figures/full_fig_p032_24.png] view at source ↗
Figure 25
Figure 25. Figure 25: Age-distribution prediction: full metric bars under spatial block splits. [PITH_FULL_IMAGE:figures/full_fig_p032_25.png] view at source ↗
Figure 26
Figure 26. Figure 26: Age-distribution prediction: full metric bars under random splits. [PITH_FULL_IMAGE:figures/full_fig_p033_26.png] view at source ↗
Figure 27
Figure 27. Figure 27: GDP regression: full metric bars under spatial block splits. [PITH_FULL_IMAGE:figures/full_fig_p033_27.png] view at source ↗
Figure 28
Figure 28. Figure 28: GDP regression: full metric bars under random splits. [PITH_FULL_IMAGE:figures/full_fig_p033_28.png] view at source ↗
Figure 29
Figure 29. Figure 29: NTL regression: full metric bars under spatial block splits. [PITH_FULL_IMAGE:figures/full_fig_p033_29.png] view at source ↗
Figure 30
Figure 30. Figure 30: NTL regression: full metric bars under random splits. [PITH_FULL_IMAGE:figures/full_fig_p033_30.png] view at source ↗
Figure 31
Figure 31. Figure 31: PM2.5 regression: full metric bars under spatial block splits [PITH_FULL_IMAGE:figures/full_fig_p034_31.png] view at source ↗
Figure 32
Figure 32. Figure 32: PM2.5 regression: full metric bars under random splits [PITH_FULL_IMAGE:figures/full_fig_p034_32.png] view at source ↗
Figure 33
Figure 33. Figure 33: Land-surface-temperature regression: full metric bars under spatial block splits. [PITH_FULL_IMAGE:figures/full_fig_p034_33.png] view at source ↗
Figure 34
Figure 34. Figure 34: Land-surface-temperature regression: full metric bars under random splits. [PITH_FULL_IMAGE:figures/full_fig_p034_34.png] view at source ↗
Figure 35
Figure 35. Figure 35: Raw primary metric change from the main 10 [PITH_FULL_IMAGE:figures/full_fig_p036_35.png] view at source ↗
Figure 36
Figure 36. Figure 36: Exploratory 26-city extension using AlphaEarth and TESSERA. Bars report the mean [PITH_FULL_IMAGE:figures/full_fig_p037_36.png] view at source ↗
Figure 37
Figure 37. Figure 37: Exploratory correlation analysis between city-level factors and downstream performance [PITH_FULL_IMAGE:figures/full_fig_p038_37.png] view at source ↗
read the original abstract

Urban representation learning encodes complex urban environments into general-purpose embeddings for diverse downstream tasks and emerging urban foundation models. However, current evaluations are limited, typically focusing on one or two cities and tasks and relying on random splits that introduce spatial leakage, leading to inflated performance and weak support for cross-location generalization and fair comparison. To address this, we propose CityRep, a unified benchmark that evaluates urban representations across data modalities, cities, and tasks using spatially structured splits. CityRep consists of three key components: (1) a spatial unit-agnostic evaluation framework that supports heterogeneous urban representations through a standardized alignment module; (2) a unified evaluation protocol using block-based spatial splits to mitigate spatial leakage and enable rigorous model comparison; and (3) an extensible multi-city, multi-task benchmark suite spanning 8 cities and 8 tasks across regression, classification, and distribution prediction. We evaluate 11 representative urban representation models. Results show that performance is highly sensitive to the split protocol, with random splits inflating scores and altering model rankings. We also observe substantial variability across cities and tasks, underscoring the need for generalization-aware evaluation. CityRep is released as a reproducible benchmark with datasets, evaluation pipelines, and diagnostic tools to facilitate fair comparison and support future research in urban representation learning towards urban foundation models.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper introduces CityRep, a benchmark for urban representation learning that spans 8 cities, 8 tasks (regression, classification, distribution prediction), and multiple modalities. It provides a spatial unit-agnostic alignment framework, a block-based spatial split protocol intended to reduce leakage, and an evaluation of 11 models. The central empirical claims are that random splits inflate scores and alter model rankings while performance varies substantially across cities and tasks, motivating generalization-aware evaluation protocols and the release of the benchmark suite.

Significance. If the empirical results on split sensitivity and cross-city variability hold under scrutiny, the work would be significant for the urban ML community by establishing a reproducible, multi-city benchmark that discourages over-optimistic evaluation and supports development of urban foundation models. The release of datasets, evaluation pipelines, and diagnostic tools is a concrete strength for reproducibility.

major comments (3)
  1. [Evaluation protocol] Evaluation protocol section: The assertion that block-based spatial splits eliminate leakage without introducing new artifacts (e.g., block-size interactions with urban density or adjacency correlations) is load-bearing for the sensitivity-to-split-protocol result, yet no sensitivity analysis to block size, no comparison against alternative spatial partitioning methods, and no quantification of residual spatial autocorrelation are reported.
  2. [Benchmark suite] Benchmark suite description: The claim that observed variability across cities and tasks supports the broader recommendation for generalization-aware evaluation rests on the 8-city/8-task selection being representative; however, no coverage analysis, diversity metrics, or justification for city/task selection is provided to rule out selection bias as the source of the reported variability.
  3. [Results] Results section on model rankings: The finding that random splits alter model rankings is central, but without reporting per-city/task variance estimates, statistical significance tests on ranking changes, or controls for the number of runs, it is unclear whether the ranking shifts exceed what would be expected from sampling variability alone.
minor comments (2)
  1. [Introduction] The abstract and introduction use the term 'spatial unit-agnostic' without a precise definition or pseudocode for the alignment module; a short formal description would improve clarity.
  2. [Experiments] Table or figure captions for the 11 models should explicitly list the modalities each model was originally trained on to allow readers to assess cross-modality generalization.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their thoughtful comments, which highlight important aspects of the evaluation protocol, benchmark design, and statistical analysis. We provide point-by-point responses below and commit to revisions that enhance the rigor of our claims without altering the core contributions.

read point-by-point responses

  1. Referee: [Evaluation protocol] Evaluation protocol section: The assertion that block-based spatial splits eliminate leakage without introducing new artifacts (e.g., block-size interactions with urban density or adjacency correlations) is load-bearing for the sensitivity-to-split-protocol result, yet no sensitivity analysis to block size, no comparison against alternative spatial partitioning methods, and no quantification of residual spatial autocorrelation are reported.

    Authors: We agree that additional validation of the block-based spatial split protocol would strengthen the manuscript. In the revised version, we will add a sensitivity analysis varying block sizes (e.g., 100m, 500m, 1km) and report effects on performance metrics and estimated leakage. We will also include a comparison to alternative methods such as random grid partitioning and administrative unit splits. To quantify residual spatial autocorrelation, we will compute Moran's I statistic on the training and test sets for each split. These additions will be presented in a new subsection of the evaluation protocol and an appendix. revision: yes

  2. Referee: [Benchmark suite] Benchmark suite description: The claim that observed variability across cities and tasks supports the broader recommendation for generalization-aware evaluation rests on the 8-city/8-task selection being representative; however, no coverage analysis, diversity metrics, or justification for city/task selection is provided to rule out selection bias as the source of the reported variability.

    Authors: The cities were chosen to represent a range of urban scales, climates, and data availability across continents, and tasks cover regression, classification, and distribution prediction with varying spatial granularities. However, we acknowledge the lack of explicit justification and metrics in the original manuscript. We will revise the benchmark suite section to include a justification based on data coverage and add diversity metrics such as city population variance, geographic spread, and task type distribution to demonstrate representativeness and mitigate concerns of selection bias. revision: yes

  3. Referee: [Results] Results section on model rankings: The finding that random splits alter model rankings is central, but without reporting per-city/task variance estimates, statistical significance tests on ranking changes, or controls for the number of runs, it is unclear whether the ranking shifts exceed what would be expected from sampling variability alone.

    Authors: We appreciate this point on statistical robustness. The original experiments were run with a fixed seed for reproducibility, but to address this, we will re-run all experiments with multiple random seeds (at least 5) and report mean and standard deviation per city and task. We will also apply statistical tests, such as the Friedman test for overall ranking significance and post-hoc tests for pairwise model comparisons, to confirm that observed ranking changes due to split protocol are statistically significant. These results will be added to the results section and tables. revision: yes

Circularity Check

0 steps flagged

Empirical benchmark paper with no derivation chain or self-referential reductions

full rationale

The paper proposes CityRep as a benchmark consisting of a spatial unit-agnostic framework, block-based splits, and a multi-city/multi-task suite; it then reports direct empirical observations on 11 models. No equations, predictions, or uniqueness claims appear that reduce by construction to fitted parameters, self-citations, or ansatzes defined inside the work. The sensitivity findings are measurements on the benchmark itself rather than outputs derived from its own inputs. This is a self-contained empirical contribution.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The work rests on domain assumptions about spatial autocorrelation in urban data and standard machine-learning evaluation practices; no free parameters or invented entities are introduced.

axioms (1)
  • domain assumption Random splits in spatial urban data introduce leakage that inflates performance and prevents fair cross-location comparison.
    This premise motivates the entire block-based split design and is stated in the abstract motivation.

pith-pipeline@v0.9.1-grok · 5788 in / 1178 out tokens · 37121 ms · 2026-06-29T21:44:08.353634+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

48 extracted references · 4 canonical work pages · 2 internal anchors

  1. [1]

    Satclip: global, general-purpose location embeddings with satellite imagery

    Konstantin Klemmer, Esther Rolf, Caleb Robinson, Lester Mackey, and Marc Rußwurm. Satclip: global, general-purpose location embeddings with satellite imagery. InProceedings of the Thirty-Ninth AAAI Conference on Artificial Intelligence and Thirty-Seventh Conference on Innovative Applications of Artificial Intelligence and Fifteenth Symposium on Educationa...

    2025

  2. [2]

    TESSERA: Temporal Embeddings of Surface Spectra for Earth Representation and Analysis

    Zhengpeng Feng, Clement Atzberger, Sadiq Jaffer, Jovana Knezevic, Silja Sormunen, Robin Young, Madeline C Lisaius, Markus Immitzer, Toby Jackson, James Ball, et al. Tessera: Temporal embeddings of surface spectra for earth representation and analysis.arXiv preprint arXiv:2506.20380, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  3. [3]

    AlphaEarth Foundations: An embedding field model for accurate and efficient global mapping from sparse label data

    Christopher F Brown, Michal R Kazmierski, Valerie J Pasquarella, William J Rucklidge, Masha Samsikova, Chenhui Zhang, Evan Shelhamer, Estefania Lahera, Olivia Wiles, Simon Ilyushchenko, et al. Alphaearth foundations: An embedding field model for accurate and efficient global mapping from sparse label data. arXiv preprint arXiv:2507.22291, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  4. [4]

    Urban2vec: Incorporating street view imagery and pois for multi-modal urban neighborhood embedding

    Zhecheng Wang, Haoyuan Li, and Ram Rajagopal. Urban2vec: Incorporating street view imagery and pois for multi-modal urban neighborhood embedding. InProceedings of the AAAI Conference on Artificial Intelligence, volume 34, pages 1013–1020, 2020

    2020

  5. [5]

    Musecl: predicting urban socioeconomic indicators via multi-semantic contrastive learning

    Xixian Yong and Xiao Zhou. Musecl: predicting urban socioeconomic indicators via multi-semantic contrastive learning. InProceedings of the Thirty-Third International Joint Conference on Artificial Intelligence, IJCAI ’24, 2024

    2024

  6. [6]

    From itdl to place2vec: Reasoning about place type similarity and relatedness by learning embeddings from augmented spatial contexts

    Bo Yan, Krzysztof Janowicz, Gengchen Mai, and Song Gao. From itdl to place2vec: Reasoning about place type similarity and relatedness by learning embeddings from augmented spatial contexts. InProceedings of the 25th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems, SIGSPATIAL ’17, New York, NY , USA, 2017. Association ...

    2017

  7. [7]

    Estimating urban functional distributions with semantics preserved poi embedding.International Journal of Geographical Information Science, 36(10):1905–1930, 2022

    Weiming Huang, Lizhen Cui, Meng Chen, Daokun Zhang, and Yao Yao. Estimating urban functional distributions with semantics preserved poi embedding.International Journal of Geographical Information Science, 36(10):1905–1930, 2022

    1905

  8. [8]

    Multi-modal contrastive learning of urban space representations from poi data.Computers, Environment and Urban Systems, 120:102299, 2025

    Xinglei Wang, Tao Cheng, Stephen Law, Zichao Zeng, Lu Yin, and Junyuan Liu. Multi-modal contrastive learning of urban space representations from poi data.Computers, Environment and Urban Systems, 120:102299, 2025

    2025

  9. [9]

    City foundation models for learning general purpose representations from openstreetmap

    Pasquale Balsebre, Weiming Huang, Gao Cong, and Yi Li. City foundation models for learning general purpose representations from openstreetmap. InProceedings of the 33rd ACM international conference on information and knowledge management, pages 87–97, 2024

    2024

  10. [10]

    Enriching location representation with detailed semantic information

    Junyuan Liu, Xinglei Wang, and Tao Cheng. Enriching location representation with detailed semantic information. In13th International Conference on Geographic Information Science (GIScience 2025), volume 346 ofLeibniz International Proceedings in Informatics (LIPIcs), pages 3:1–3:15, 2025

    2025

  11. [11]

    Roberts, V olker Bahn, Simone Ciuti, Mark S

    David R. Roberts, V olker Bahn, Simone Ciuti, Mark S. Boyce, Jane Elith, Gurutzeta Guillera-Arroita, Severin Hauenstein, José J. Lahoz-Monfort, Boris Schröder, Wilfried Thuiller, David I. Warton, Brendan A. Wintle, Florian Hartig, and Carsten F. Dormann. Cross-validation strategies for data with temporal, spatial, hierarchical, or phylogenetic structure.E...

    2017

  12. [12]

    Lahoz-Monfort, and Gurutzeta Guillera-Arroita

    Roozbeh Valavi, Jane Elith, José J. Lahoz-Monfort, and Gurutzeta Guillera-Arroita. blockCV: An R package for generating spatially or environmentally separated folds for k-fold cross-validation of species distribution models.Methods in Ecology and Evolution, 10(2):225–232, 2019

    2019

  13. [13]

    Importance of spatial predictor variable selection in machine learning applications: Moving from data reproduction to spatial prediction.Ecological Modelling, 411:108815, 2019

    Hanna Meyer, Christoph Reudenbach, Stephan Wöllauer, and Thomas Nauss. Importance of spatial predictor variable selection in machine learning applications: Moving from data reproduction to spatial prediction.Ecological Modelling, 411:108815, 2019

    2019

  14. [14]

    Poi2vec: geographical latent representation for predicting future visitors

    Shanshan Feng, Gao Cong, Bo An, and Yeow Meng Chee. Poi2vec: geographical latent representation for predicting future visitors. InProceedings of the Thirty-First AAAI Conference on Artificial Intelligence, AAAI’17, pages 102–108. AAAI Press, 2017

    2017

  15. [15]

    Learning urban region representations with pois and hierarchical graph infomax.ISPRS Journal of Photogrammetry and Remote Sensing, 196:134–145, 2023

    Weiming Huang, Daokun Zhang, Gengchen Mai, Xu Guo, and Lizhen Cui. Learning urban region representations with pois and hierarchical graph infomax.ISPRS Journal of Photogrammetry and Remote Sensing, 196:134–145, 2023. 10

    2023

  16. [16]

    Beyond alphaearth: toward human-centered spatial representation via poi-guided contrastive learning

    Junyuan Liu, Quan Qin, Guangsheng Dong, Xinglei Wang, Jiazhuang Feng, Zichao Zeng, and Tao Cheng. Beyond alphaearth: toward human-centered spatial representation via poi-guided contrastive learning. arXiv preprint arXiv:2510.09894, 2025

    work page arXiv 2025

  17. [17]

    Multi-scale representation learning for spatial feature distributions using grid cells

    Gengchen Mai, Krzysztof Janowicz, Bo Yan, Rui Zhu, Ling Cai, and Ni Lao. Multi-scale representation learning for spatial feature distributions using grid cells. InInternational Conference on Learning Representations, 2020

    2020

  18. [18]

    H3 indexes points and shapes into a hexagonal grid, 2026

    Uber Technologies, Inc. H3 indexes points and shapes into a hexagonal grid, 2026. https://h3geo.org/. Data available under the Apache 2.0 License. Accessed 06/05/2026

    2026

  19. [19]

    Stewart, Caleb Robinson, Isaac A

    Adam J. Stewart, Caleb Robinson, Isaac A. Corley, Anthony Ortiz, Juan M. Lavista Ferres, and Arindam Banerjee. Torchgeo: Deep learning with geospatial data.ACM Transactions on Spatial Algorithms and Systems, 11(4), 2025. Article 15

    2025

  20. [20]

    Geo-bench: toward foundation models for earth monitoring

    Alexandre Lacoste, Nils Lehmann, Pau Rodriguez, Evan David Sherwin, Hannah Kerner, Björn Lütjens, Jeremy Irvin, David Dao, Hamed Alemohammad, Alexandre Drouin, Mehmet Gunturkun, Gabriel Huang, David Vazquez, Dava Newman, Yoshua Bengio, Stefano Ermon, and Xiao Xiang Zhu. Geo-bench: toward foundation models for earth monitoring. InProceedings of the 37th In...

    2023

  21. [21]

    Satlaspretrain: A large-scale dataset for remote sensing image understanding

    Favyen Bastani, Piper Wolters, Ritwik Gupta, Joe Ferdinando, and Aniruddha Kembhavi. Satlaspretrain: A large-scale dataset for remote sensing image understanding. In2023 IEEE/CVF International Conference on Computer Vision (ICCV), pages 16726–16736, 2023

    2023

  22. [22]

    Pangaea: Assessing geospatial foundation models capabilities through a global and inclusive benchmark.IEEE Geoscience and Remote Sensing Magazine, 14:245–285, 2026

    Valerio Marsocci, Yuru Jia, Georges Le Bellier, Dávid Kerekes, Liang Zeng, Sebastian Hafner, Sebastian Gerard, Eric Brune, Ritu Yadav, Ali Shibli, Heng Fang, Yifang Ban, Maarten Vergauwen, Nicolas Audebert, and Andrea Nascetti. Pangaea: Assessing geospatial foundation models capabilities through a global and inclusive benchmark.IEEE Geoscience and Remote ...

    2026

  23. [23]

    Torchspatial: A location encoding framework and benchmark for spatial representation learning

    Nemin Wu, Qian Cao, Zhangyu Wang, Zeping Liu, Yanlin Qi, Jielu Zhang, Joshua Ni, Xiaobai Yao, Hongxu Ma, Lan Mu, Stefano Ermon, Tanuja Ganu, Akshay Nambi, Ni Lao, and Gengchen Mai. Torchspatial: A location encoding framework and benchmark for spatial representation learning. InAdvances in Neural Information Processing Systems, volume 37, pages 81437–81460, 2024

    2024

  24. [24]

    OBSR: Open benchmark for spatial representations

    Julia Moska, Oleksii Furman, Kacper Kozaczko, Szymon Leszkiewicz, Jakub Polczyk, Piotr Gramacki, and Piotr Szyma ´nski. OBSR: Open benchmark for spatial representations. InProceedings of the 33rd ACM International Conference on Advances in Geographic Information Systems, SIGSPATIAL ’25, 2025

    2025

  25. [25]

    MoRA: Mobility as the backbone for geospatial representation learning at scale

    Ya Wen, Jixuan Cai, Qiyao Ma, Linyan Li, Xinhuan Chen, Chris Webster, and Yulun Zhou. MoRA: Mobility as the backbone for geospatial representation learning at scale. InInternational Conference on Learning Representations, 2026

    2026

  26. [26]

    Dormann, Guillaume Cornu, Gaëlle Viennois, Nicolas Bayol, Alexei Lyapustin, Sylvie Gourlet- Fleury, and Raphaël Pélissier

    Pierre Ploton, Frédéric Mortier, Maxime Réjou-Méchain, Nicolas Barbier, Nicolas Picard, Vivien Rossi, Carsten F. Dormann, Guillaume Cornu, Gaëlle Viennois, Nicolas Bayol, Alexei Lyapustin, Sylvie Gourlet- Fleury, and Raphaël Pélissier. Spatial validation reveals poor predictive performance of large-scale ecological mapping models.Nature Communications, 11...

    2020

  27. [27]

    Primary Land Use Tax Lot Output (PLUTO), version 25v4

    New York City Department of City Planning. Primary Land Use Tax Lot Output (PLUTO), version 25v4. NYC Open Data / BYTES of the BIG APPLE, 2026. CSV tax-lot records; current version 25v4; last modified 2026-02-20; accessed 2026-05-06

    2026

  28. [28]

    Master Plan 2019 Land Use layer (2025)

    Urban Redevelopment Authority. Master Plan 2019 Land Use layer (2025). data.gov.sg, 2023. Indicative land-use polygon layer; data from Dec. 2019; extracted from the Master Plan 2019 approved on 2019-11-18; last updated 2025-12-05; accessed 2026-05-06

    2019

  29. [29]

    Environmental Planning Instrument – Land Zoning

    NSW Department of Planning, Housing and Infrastructure. Environmental Planning Instrument – Land Zoning. NSW Planning Portal Open Data, 2022. Vector land-zoning dataset; update frequency weekly; Creative Commons Attribution; accessed 2026-05-06

    2022

  30. [30]

    City of Cape Town. Zoning. Cape Town Open Data Portal feature service, 2019. Zoning / split-zoning feature service; published 2019-02-10; dataset-specific or custom licence; accessed 2026-05-06

    2019

  31. [31]

    Peraturan Gubernur Nomor 31 Tahun 2022 tentang Rencana Detail Tata Ruang Wilayah Perencanaan Provinsi Daerah Khusus Ibukota Jakarta

    Pemerintah Provinsi DKI Jakarta. Peraturan Gubernur Nomor 31 Tahun 2022 tentang Rencana Detail Tata Ruang Wilayah Perencanaan Provinsi Daerah Khusus Ibukota Jakarta. JDIH Provinsi DKI Jakarta, 2022. Peraturan Gubernur No. 31; ditetapkan 2022-06-27; BD Provinsi DKI Jakarta Tahun 2022 Nomor 63003; accessed 2026-05-06. 11

    2022

  32. [32]

    Development Plan for Greater Mumbai 2014–2034: Existing Land Use Maps and Report

    Municipal Corporation of Greater Mumbai. Development Plan for Greater Mumbai 2014–2034: Existing Land Use Maps and Report. Municipal Corporation of Greater Mumbai, 2012. Existing Land Use 2012 source; cited via UDRI/Login Mumbai data-source register; accessed 2026-05-06

    2014

  33. [33]

    Nairobi, Kenya Land Use 2010

    World Bank Data Catalog. Nairobi, Kenya Land Use 2010. World Bank Data Catalog, 2010. Land-use shapefile; source: Columbia University’s Center for Sustainable Urban Development; shapefile last updated 2020-06-25; metadata last updated 2023-01-19; Creative Commons Attribution 4.0; accessed 2026-05-06

    2010

  34. [34]

    OpenStreetMap road-network extract for the CityRep benchmark cities

    OpenStreetMap contributors. OpenStreetMap road-network extract for the CityRep benchmark cities. OpenStreetMap database, 2026. Drivable highway ways extracted for London, New York, Singapore, Sydney, Mumbai, Nairobi, Jakarta, and Cape Town; licensed under the Open Data Commons Open Database License (ODbL) 1.0; accessed/extracted 2026-05-06

    2026

  35. [35]

    Bondarenko, R

    M. Bondarenko, R. Priyatikanto, N. Tejedor-Garavito, W. Zhang, T. McKeen, A. Cunningham, T. Woods, J. Hilton, D. Cihan, B. Nosatiuk, T. Brinkhoff, A. Tatem, and A. Sorichetta. Constrained estimates of 2015–2030 total number of people per grid square at a resolution of 3 arc (approximately 100m at the equator), R2024B version v1. WorldPop, School of Geogra...

    2015

  36. [36]

    Bondarenko, R

    M. Bondarenko, R. Priyatikanto, N. Tejedor-Garavito, W. Zhang, T. McKeen, A. Cunningham, T. Woods, J. Hilton, D. Cihan, B. Nosatiuk, T. Brinkhoff, A. Tatem, and A. Sorichetta. Constrained estimates of 2015–2030 total number of people per grid square broken down by gender and age groupings at a resolution of 3 arc (approximately 100m at the equator), R2024...

    2015

  37. [37]

    Downscaled gridded global dataset for gross domestic product (GDP) per capita PPP over 1990–2022.Scientific Data, 12:178, 2025

    Matti Kummu, Maria Kosonen, and Sina Masoumzadeh Sayyar. Downscaled gridded global dataset for gross domestic product (GDP) per capita PPP over 1990–2022.Scientific Data, 12:178, 2025

    1990

  38. [38]

    VIIRS Nighttime Lights Annual V2.2.https://eogdata.mines.edu/products/vnl/, 2024

    Earth Observation Group, Payne Institute for Public Policy, Colorado School of Mines. VIIRS Nighttime Lights Annual V2.2.https://eogdata.mines.edu/products/vnl/, 2024. Accessed: 2026-05-06

    2024

  39. [39]

    Elvidge, Mikhail Zhizhin, Tilottama Ghosh, Feng-Chi Hsu, and Jay Taneja

    Christopher D. Elvidge, Mikhail Zhizhin, Tilottama Ghosh, Feng-Chi Hsu, and Jay Taneja. Annual time series of global VIIRS nighttime lights derived from monthly averages: 2012 to 2019.Remote Sensing, 13(5):922, 2021

    2012

  40. [40]

    van Donkelaar, M

    A. van Donkelaar, M. S. Hammer, L. Bindle, M. Brauer, J. R. Brook, M. J. Garay, N. C. Hsu, O. V . Kalashnikova, R. A. Kahn, C. Lee, R. C. Levy, A. Lyapustin, A. M. Sawyer, and R. V . Martin. Global Annual PM2.5 Grids from MODIS, MISR, SeaWiFS and VIIRS Aerosol Optical Depth (AOD), 1998–2022, V5.GL.04, Version 5.04. NASA Socioeconomic Data and Applications...

    1998

  41. [41]

    MODIS/Terra Land Surface Temperature/Emissivity 8-Day L3 Global 1km SIN Grid V061

    Zhengming Wan, Simon Hook, and Glynn Hulley. MODIS/Terra Land Surface Temperature/Emissivity 8-Day L3 Global 1km SIN Grid V061. NASA Land Processes Distributed Active Archive Center, 2021. Data set; accessed 2026-05-06

    2021

  42. [42]

    FSQ OS Places

    Foursquare Labs, Inc. FSQ OS Places. Foursquare Open Source Places data product, 2026. Open POI dataset; Apache License 2.0; preserve Foursquare attribution and NOTICE file; accessed 2026-05-06

    2026

  43. [43]

    and Mapillary

    Meta Platforms, Inc. and Mapillary. Mapillary Open Street-Level Imagery. Mapillary platform / API,

  44. [44]

    Street-level imagery and map-data platform; accessed via Mapillary API on 2026-05-06; images are shared under CC BY-SA and individual image/user attribution should be preserved where required

    2026

  45. [45]

    A review of location encoding for geoai: methods and applications.International Journal of Geographical Information Science, 36:639–673, 2022

    Gengchen Mai, Krzysztof Janowicz, Yingjie Hu, Song Gao, Bo Yan, Rui Zhu, Ling Cai, and Ni Lao. A review of location encoding for geoai: methods and applications.International Journal of Geographical Information Science, 36:639–673, 2022

    2022

  46. [46]

    Gengchen Mai, Yao Xuan, Wenyun Zuo, Yutong He, Jiaming Song, Stefano Ermon, Krzysztof Janowicz, and Ni Lao. Sphere2vec: A general-purpose location representation learning over a spherical surface for large-scale geospatial predictions.ISPRS Journal of Photogrammetry and Remote Sensing, 202:439–462, 2023

    2023

  47. [47]

    Land based classification standards

    American Planning Association. Land based classification standards. https://www.planning.org/ lbcs/, 2000. Accessed 2026-05-05

    2000

  48. [48]

    resolution/support

    James R. Anderson, Ernest E. Hardy, John T. Roach, and Richard E. Witmer. A land use and land cover classification system for use with remote sensor data. Professional Paper 964, U.S. Geological Survey, 1976. 12 Appendix Contents A Discussion 14 A.1 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 A.2 Broader Impa...

    work page arXiv 1976