Pith. sign in

REVIEW 2 major objections 5 minor 49 references

Do Maps Still Matter for Machines: Revisiting the Role of Choropleth Maps in Foundation Model Spatial Understanding

T0 review · 2 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

Pith's one-line read This paper claims that pairing a choropleth map image with symbolic GeoJSON data consistently improves foundation-model spatial reasoning, with the largest gains on global pattern-recognition tasks.

desk verdict Useful benchmark with a plausible result, but the Data+Map vs Data Only comparison needs to rule out GeoJSON truncation/parsing confounds before the headline claim is clean. read the letter →

arxiv 2607.17999 v2 pith:LB2KA2KZ submitted 2026-07-20 cs.AI cs.CV

classification cs.AIcs.CV
keywords choroplethmapsfoundationmodelsspatialreasoningGeoJSONmultimodalcartographicabstractionbenchmarkcognitivecompression
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper asks whether a centuries-old human tool—the choropleth map—still improves machine reasoning in an era when foundation models can ingest raw structured geodata directly. To answer it, the authors built a controlled benchmark of 2,400 synthetic choropleth maps, their matching GeoJSON representations, and 12,000 multiple-choice questions spanning five cognitive levels from simple region identification to global pattern delineation. Across 22 open-source and proprietary multimodal models, giving the model both the GeoJSON data and the map image (Data + Map) produced the highest accuracy for 20 of the 22 models, beating data alone by about 6 percentage points on discrete maps and by double digits over maps alone. The gains were largest on tasks requiring global pattern recognition such as counting clusters or trends, suggesting maps serve as spatial anchors that compress reasoning burden. The paper concludes that maps remain valuable external representations for machines, not obsolete artifacts.

What carries the argument

The load-bearing object is ChoroplethMap-Bench, a controlled benchmark in which every rendered choropleth map has a one-to-one symbolic twin in compact GeoJSON, so the same 12,000 questions can be posed under three input conditions—Data Only, Map Only, and Data + Map—while holding task content fixed. The benchmark deliberately varies map type (discrete 4-class versus continuous color encoding), 12 hue settings, and four spatial structures (cluster, trend, structure, random), and organizes questions into five hierarchical cognitive dimensions. The comparison logic does the work: any accuracy difference across conditions is attributable to representation format rather than task difficulty, let

What would settle it

Re-run the benchmark while logging the token count of every GeoJSON and comparing it to each model's context window, and record parse-failure and retry rates per condition. If, after ensuring full legibility (or padding Data+Map to equalize token load), Data-Only accuracy still trails Data+Map, the map-benefit claim stands; if the gap vanishes under equalized legibility, the conclusion is confounded by input truncation. A second check: swap the map for an information-equivalent non-cartographic image; if that produces the same boost, the effect is not specifically choropleth cartography.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central discovery is that cartographic abstraction is not redundant for foundation models: when a choropleth map image is added to the exact same GeoJSON data, 20 of 22 evaluated models improve, with an average gain of 6.0% on discrete maps and 3.9% on continuous maps over data alone, and 14.8% and 16.3% over map alone. The effect is strongest in the highest-level task dimension, Delineate, where models must recognize clusters, gradients, and ring structures across the whole map, and weakest for simple point-lookup tasks like Identify. Discrete class-based maps consistently beat continuous gradient maps, while hue variation had almost no effect, and organized sp

Load-bearing premise

Every model actually receives and parses the complete compressed GeoJSON in the Data-Only and Data+Map conditions—the paper never reports token counts, context-window limits, or parse-failure rates, so a Data-Only shortfall could stem from truncated or mangled text input rather than from the missing map.

Editorial extensions

If this is right

  • If the finding holds, map images should be treated as an input modality to engineer for in geospatial AI systems rather than a legacy human interface.
  • Discrete classification and high-contrast, hue-neutral color schemes are better for machine readers, giving cartographers a concrete, model-facing design role.
  • Because Map Only lags far behind Data + Map for current models, purely visual map reasoning remains unsolved; grounding map reading in symbolic data is the practical recipe until vision encoders improve.
  • Benchmarking geospatial reasoning should include representation-format comparisons; accuracy alone on a single input format can understate a model's true spatial competence.
  • The largest gains on global-pattern tasks imply that applications like regional disparity analysis, climate-pattern detection, or election-outcome mapping will benefit most from map-plus-data pipelines.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The unstated legibility premise is the most serious unaddressed confound: the paper compresses GeoJSON to keep it compact but never reports token counts or parse-failure rates, and its limitations section does not mention this risk.
  • A testable extension: replace the choropleth with a non-cartographic but information-equivalent visual (e.g., a grid heatmap or ordered scatter) under Data + Visual. If that also beats Data Only, the effect is generic visual anchoring, not specifically cartographic abstraction.
  • The map-type and hue results suggest a broader design principle the authors only hint at: AI-oriented cartography may diverge from human-oriented cartography—machine-readable maps could emphasize discrete classes and luminance contrast in ways that differ from human aesthetic norms.
  • The small human-validity study (four participants, all from one demographic) limits the claim that the benchmark is clear and solvable at 92.9%; a larger, more diverse validation would firm up that baseline.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper introduces ChoroplethMap-Bench, a controlled synthetic benchmark of 2,400 choropleth maps with paired GeoJSON data and 12,000 multiple-choice questions spanning five task dimensions (Identify, Spatial Recognition, Compare, Rank, Delineate). It evaluates 22 open-source and proprietary foundation models under three input conditions: Data Only, Map Only, and Data + Map. The central claim is that maps remain valuable for machine spatial understanding: Data + Map outperforms Data Only for 20 of 22 models, and the difference is statistically significant by Friedman/Nemenyi tests. The paper reports human validation (92.93% accuracy), robustness checks on temperature, quantization, classification, prompting, language, and repeated runs, and interprets the gains as evidence that maps act as cognitive compression interfaces for spatial reasoning.

Significance. If the central comparison is valid, this is a useful and timely controlled benchmark. The strengths are concrete: deterministic ground truth generated from map geometry, human validation confirming solvability, a broad model suite, multiple robustness analyses, and open data/code (GitHub, CC BY 4.0). The task hierarchy from local lookup to global pattern recognition is sensible, and the finding that gains concentrate in D5 Delineate is a credible, falsifiable pattern that supports the cognitive compression hypothesis. The main threat to significance is that the Data+Map versus Data Only comparison may be confounded by input legibility or information-superiority; if that is resolved, the benchmark could become a standard reference for evaluating cartographic representations in foundation models.

major comments (2)
  1. [§3.2.2 Step 4; §3.3.1; §4.1 (Table 3)] The headline Data+Map>Data Only comparison is potentially confounded by unverified GeoJSON legibility. §3.2.2 Step 4 compresses GeoJSON 'to keep the symbolic input compact,' but no token counts, context-window utilization, or parse/truncation success are reported for any model. Since Data+Map contains the same JSON plus a map, a model with truncated Data Only can fall back on the image; Data Only cannot. The 4–6% gain in Table 3 may thus reflect input formatting rather than cartographic abstraction. This is sharpened by Qwen3-VL-2B-Instruct, the only chance-level case (28.11%, p=0.092, §4), which matches the expected truncation pattern. §5.6 does not acknowledge this threat. Please report token statistics, check for truncation, and add a concise non-spatial control (CSV/table) to separate 'map helps' from 'extra channel helps'.
  2. [§3.3.1; §5.5; §6] Even with fully legible input, Data+Map is information-superior to Data Only by construction: it contains the complete GeoJSON plus a map. The paper's statement that differences 'can be directly attributed to representational format' (§3.3.1) therefore overstates the design. Part of the gain is expected from information quantity/redundancy. The stronger interpretation of maps as 'cognitive compression interfaces' (§5.5, §6) needs a control that adds a non-cartographic but equally informative channel (e.g., a table or scatterplot of the same values), or a per-model analysis showing that map-only visual structure predicts the residual gain after controlling for Data Only accuracy. Without this, the result supports 'adding a map to GeoJSON helps,' but not specifically 'maps are uniquely valuable spatial abstractions.'
minor comments (5)
  1. [Table 3] For Qwen3.6-plus and Claude Opus 4.8, Data Only is slightly higher than Data+Map (68.9 vs. 68.3 and 73.8 vs. 73.6). The text says '20 out of 22,' which is accurate, but the two exceptions are within the repeated-run noise reported in §5.4 (≤0.5 pp); this should be acknowledged when claiming 'consistently.'
  2. [Table 8] The header lists FP16 precision, but the rows and text compare NF4, INT8, and BF16. Make the table header consistent with the body.
  3. [§3.2.3] Human validation uses four participants only, all from the same demographic; report per-question variance or confidence intervals, and consider a second validator group. This does not affect the main conclusion, since 92.93% overall accuracy is far above chance.
  4. [§5.2.1] The language robustness experiment uses only two models and no significance tests; the claim of 'cross-lingual robustness' would be stronger with per-model confidence intervals or a paired test.
  5. [§1; §5.6] The abstract and Section 5.6 describe the benchmark as the first of its kind; given MapQA [49], clarify that the novelty is the controlled three-condition representation comparison rather than the first choropleth QA dataset.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found; benchmark comparison is empirical and self-contained.

full rationale

The paper's central claim is an empirical benchmark result, not a derivation from fitted parameters or self-referential definitions. Data+Map, Data Only, and Map Only are independently constructed input conditions over the same tasks, and the observed accuracy differences are measured rather than entailed by construction. The fact that Data+Map contains strictly more information than Data Only is not circular: a superset input can in principle hurt performance, and indeed Map Only often underperforms Data Only, showing that additional information does not trivially yield gains. No parameter is fitted to a subset and then renamed as a prediction; no uniqueness theorem or ansatz is imported from the authors' prior work; and the self-citations (e.g., [1], [44]) are contextual and not load-bearing for the main result. A legitimate validity concern about unverified GeoJSON token counts, truncation, or parse success is a correctness/confounding risk rather than a circularity, so it does not change the circularity score.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

No physical or mathematical entities are invented. 'Cognitive compression' is interpretive language for the role of maps, not an independently falsifiable mechanism. The free parameters listed are hand-set benchmark design constants, not fits to model performance, but they define the scope of the central claim. The axiomatic assumptions are mostly standard benchmark-domain assumptions; the weakest is the unverified full-legibility of GeoJSON inputs and the independence assumption in the statistical tests.

free parameters (6)
  • discrete_classes = 4
    Hand-selected number of ordered classes for discrete choropleth maps (Section 3.2.2); central result may change with class count.
  • hue_settings_count = 12
    Twelve evenly spaced HSV hues cycled across maps to reduce color bias; affects the hue-invariance analysis.
  • vertex_limit = 8
    TopoJSON simplification caps exterior vertices per polygon at 8 (Section 3.2.2 Step 2); affects visual legibility and Map-Only results.
  • compactness_threshold = 0.1
    Polygons with compactness below 0.1 are discarded to remove edge fragments; a hand-set filter that shapes the benchmark geometry.
  • region_count_window = 15-40
    Clipping windows retained only if they contain 15-40 regions; controls task difficulty.
  • rendering_resolution = 640x640 @ 300 DPI
    Fixed rendering resolution; visual decoding and OCR of map labels depend on this choice.
assumptions (5)
  • domain assumption The simplified synthetic maps preserve the spatial information needed for all five task types, and anonymized labels remove semantic priors.
    Central to a fair Map-vs-Data comparison; if simplification removes relevant cues, Map Only is artificially weak. Invoked in Section 3.2.2 Steps 1-5.
  • domain assumption Every evaluated model can consume the entire compressed GeoJSON without truncation or parsing failure in Data Only and Data+Map conditions.
    No token counts, context windows, or truncation checks are reported. If the JSON is truncated, Data Only is handicapped relative to Data+Map. Relevant to Appendix A.1-A.3.
  • domain assumption Ground-truth answers are deterministically recoverable from the map image alone, including legend legibility and color contrast.
    Supported only by a small human validation study (Section 3.2.3): four GIS undergraduates on 25% of maps. Not verified for every map or every model's visual decoder.
  • standard math The 12,000 questions are treated as independent Bernoulli trials in the above-chance testing procedure.
    Poisson-binomial tests in Section 4 assume independence, but five questions share each map and all questions share a model, introducing within-map and within-model correlation.
  • domain assumption The 22 models have stable versions and APIs during the February-April 2026 evaluation window.
    Proprietary model behavior can change during API evaluation; no version snapshots are given for API-accessed models.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Do Maps Still Matter for Machines: Revisiting the Role of Choropleth Maps in Foundation Model Spatial Understanding." pith.science (2026). https://pith.science/paper/LB2KA2KZ

@misc{pith2026260717999,
  author       = {Pith},
  title        = {Pith review of: Do Maps Still Matter for Machines: Revisiting the Role of Choropleth Maps in Foundation Model Spatial Understanding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LB2KA2KZ}},
  note         = {Machine review of arXiv:2607.17999}
}
read the original abstract

Spatial understanding is crucial for foundation models (FMs), and maps have long helped humans organize and reason about geographic information. This study examines whether choropleth maps remain useful for machine spatial understanding when models can directly process structured geodata. We introduce ChoroplethMap-Bench, a controlled benchmark containing 2,400 synthetic choropleth maps, corresponding GeoJSON data, and 12,000 questions across five cognitive dimensions: Identify, Spatial Recognition, Compare, Rank, and Delineate. We evaluate 22 open-source and proprietary models under three input conditions: Data Only, Map Only, and Data + Map. The results show that maps substantially improve spatial reasoning, especially when combined with symbolic data and for tasks requiring higher-level understanding of spatial patterns. We further analyze the effects of map type, color hue, and spatial structure, as well as prompting strategies, language, geographic context, decoding settings, classification methods, and response stability. Overall, the Data + Map condition achieves the strongest performance, demonstrating that maps remain valuable external representations for foundation model spatial reasoning.

Figures

Figures reproduced from arXiv: 2607.17999 by the authors.

Figure 1
Figure 1. Overview of the ChoroplethMap-Bench construction and evaluation framework. The benchmark contains 2,400 synthetic choropleth maps, their corresponding GeoJSON data, and 12,000 multiple-choice questions covering five spatial-understanding dimensions. Foundation models are evaluated under three input conditions—Data + Map, Data Only, and Map Only—and the results are further analyzed with respect to model performance, … view at source ↗
Figure 2
Figure 2. • Cluster: For cluster maps, an adjacency dictionary is used to expand from randomly selected seed regions and form up to two compact high-value or low￾value clusters. Core regions are assigned extreme class values, while surrounding buffer regions receive intermediate values. • Trend: For trend maps, a direction is randomly selected from 0◦ , 45◦ , 90◦ , or 135◦ . Region centroids are projected onto this direction,… view at source ↗
Figure 2
Figure 2. Representative choropleth maps generated under four spatial-distribution patterns: (a) Cluster, in which adjacent regions form spatially concentrated high- or low-value groups; (b) Trend, characterized by a directional attribute gradient; (c) Structure, containing ring-shaped spatial configurations; and (d) Random, with no predefined global spatial organization. Each panel also presents an example benchmark question… view at source ↗
Figures from the paper (3 more)
Figure 3
Figure 3. Figure 3: Different design factors’ influence on models. (a) Accuracy by map type across models; (b) Accuracy by hue; (c) Accuracy distribution by spatial pattern. 5. Discussion 5.1. Robustness to Prompting Strategies Recent studies have shown that prompting strategies can subst…
Figure 4
Figure 4. Figure 4: Case 1: Local spatial relation reasoning (D2). D2 requires identifying the relative direction of one region with respect to another. Under Data + Map, the model directly uses the visual layout to infer east–west and north–south relations, and then uses coordinates only…
Figure 4
Figure 4. Figure 4: Two representative cases illustrating how map visualization supports spatial reasoning under the Data + Map condition compared with the Data Only condition. The left case concerns local directional￾relation reasoning, whereas the right case concerns recognition of a gl…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 5 linked inside Pith

  1. [1]

    Z. Wei, Y. Liu, H. Liao, and W. Xu,Thinking on Maps: How Foundation Model Agents Explore, Remember, and Reason Map Environments, arXiv preprint arXiv:2512.24504, 2025

  2. [2]

    Bommasani, D

    R. Bommasani, D. A. Hudson, E. Adeli, R. Altman, S. Arora, S. von Arx, M. S. Bernstein, J. Bohg, A. Bosselut, E. Brunskill, et al.,On the opportunities and risks of foundation models, arXiv preprint arXiv:2108.07258, 2021

  3. [3]

    Brown, B

    T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Nee- lakantan, P. Shyam, G. Sastry, A. Askell, et al.,Language models are few-shot learners, Advances in Neural Information Processing Systems, vol. 33, pp. 1877– 1901, 2020

  4. [4]

    Achiam, S

    J. Achiam, S. Adler, S. Agarwal, L. Ahmad, I. Akkaya, F. L. Aleman, D. Almeida, J. Altenschmidt, S. Altman, S. Anadkat, et al.,GPT-4 technical report, arXiv preprint arXiv:2303.08774, 2023

  5. [5]

    S. Reed, K. Zolna, E. Parisotto, S. G. Colmenarejo, A. Novikov, G. Barth-Maron, M. Gimenez, Y. Sulsky, J. Kay, J. T. Springenberg, et al.,A generalist agent, arXiv preprint arXiv:2205.06175, 2022

  6. [6]

    Awais, M

    M. Awais, M. Naseer, S. Khan, R. M. Anwer, H. Cholakkal, M. Shah, M.-H. Yang, F. S. Khan,Foundation models defining a new era in vision: a survey and outlook, 28 IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 47, no. 4, pp. 2245–2264, 2025

  7. [7]

    Z. Wei, N. Yang, Y. Chen, R. Guo,Generalized application of map generalization methods in the era of pan-map, Bulletin of Surveying and Mapping, vol. 0, no. 9, p. 118, 2025

  8. [8]

    Z. Wei, Y. Sun, Z. Liu, W. Xu, W. Dong, C. Liu, H. Liao,ChoroplethMap-Bench: A benchmark for evaluating the cognitive compression and spatial reasoning of MLLMs using 2,400 synthetic choropleth maps, GitHub repository, accessed May 8, 2026,https://github.com/Myantion/ChoroplethMap-Bench

Show all 49 references
  1. [9]

    V. R. Li, J. L. Sun, M. Wattenberg,Does visualization help AI understand data?, in 2025 IEEE Visualization and Visual Analytics (VIS), pp. 51–55, IEEE, 2025

  2. [10]

    Masry, X

    A. Masry, X. L. Do, J. Q. Tan, S. Joty, E. Hoque,Chartqa: A benchmark for question answering about charts with visual and logical reasoning, in Findings of the Association for Computational Linguistics: ACL 2022, pp. 2263–2279, 2022

  3. [11]

    Mangalam, R

    K. Mangalam, R. Akshulakov, J. Malik,Egoschema: A diagnostic benchmark for very long-form video language understanding, Advances in Neural Information Processing Systems, vol. 36, pp. 46212–46244, 2023

  4. [12]

    Majumdar, A

    A. Majumdar, A. Ajay, X. Zhang, P. Putta, S. Yenamandra, M. Henaff, S. Silwal, P. Mcvay, O. Maksymets, S. Arnaud, et al.,Openeqa: Embodied question answer- ing in the era of foundation models, Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, p...

  5. [13]

    F. Liu, G. Emerson, N. Collier,Visual spatial reasoning, Transactions of the As- sociation for Computational Linguistics, vol. 11, pp. 635–651, 2023

  6. [14]

    J. Yang, S. Yang, A. W. Gupta, R. Han, L. Fei-Fei, S. Xie,Thinking in space: How multimodal large language models see, remember, and recall spaces, Proceedings of the Computer Vision and Pattern Recognition Conference, pp. 10632–10643, 2025

  7. [15]

    A. M. MacEachren,How maps work: representation, visualization, and design, Guilford Press, 2004

  8. [16]

    D. R. Montello,Cognitive map-design research in the twentieth century: Theoret- ical and empirical approaches, Cartography and Geographic Information Science, vol. 29, no. 3, pp. 283–304, 2002

  9. [17]

    T. A. Slocum, R. B. McMaster, F. C. Kessler, H. H. Howard,Thematic cartogra- phy and geovisualization, CRC Press, 2022

  10. [18]

    Kraak, A

    M.-J. Kraak, A. MacEachren,Visualization for exploration of spatial data, Inter- national Journal of Geographical Information Science, vol. 13, no. 4, pp. 285–287, 1999

  11. [19]

    A. H. Robinson, A. Kimerling, et al.,Elements of cartography, New York, John Wiley and Sons, 1969

  12. [20]

    S. I. Fabrikant, A. Lobben,Introduction: Cognitive issues in geographic infor- mation visualization, Cartographica: The International Journal for Geographic Information and Geovisualization, vol. 44, no. 3, pp. 139–143, 2009

  13. [21]

    R. E. Roth,An empirically-derived taxonomy of interaction primitives for inter- active cartography and geovisualization, IEEE Transactions on Visualization and Computer Graphics, vol. 19, no. 12, pp. 2356–2365, 2013

  14. [22]

    W. Dong, H. Liao, Z. Zhan, B. Liu, S. Wang, T. Yang,New research progress of eye tracking-based map cognition in cartography since 2008, Acta Geographica Sinica, vol. 74, no. 3, pp. 599–614, 2019

  15. [23]

    W. Dong, T. Yang, H. Liao, L. Meng,How does map use differ in virtual reality and desktop-based environments?, International Journal of Digital Earth, vol. 13, 29 no. 12, pp. 1484–1503, 2020

  16. [24]

    N. Yang, Y. Wang, Z. Wei, F. Wu,MapColorAI: Designing contextually relevant choropleth map color schemes using a large language model, Cartography and Geographic Information Science, pp. 1–19, 2025

  17. [25]

    Z. Wei, N. Yang, Y. Wang, H. Liao, M. Zhou, W. Xu,MapColor-Agent: a large language model-integrated multi-agent framework for administrative map color design, Acta Geodaetica et Cartographica Sinica, vol. 55, no. 2, pp. 359–370, 2026

  18. [26]

    N. Yang, F. Wu, Z. Wei, Y Wang,Learning From Good Maps: A Dataset and Analysis of Layout Patterns, Transactions in GIS, vol. 30, no. 4, pp. e70329, 2026

  19. [27]

    Q. Luo, Q. Lin, L. Xu, S. Wu, R. Mao, C. Wang, H. Feng, B. Huang, Z. Du,Geo- JSON agents: a multi-agent LLM architecture for geospatial analysis—function calling vs. code generation, Big Earth Data, pp. 1–55, 2026

  20. [28]

    C. A. Brewer,Color use guidelines for mapping, Visualization in Modern Cartog- raphy, vol. 1994, no. 123-148, p. 7, 1994

  21. [29]

    C. A. Brewer, L. Pickle,Evaluation of methods for classifying epidemiological data on choropleth maps in series, Annals of the Association of American Geographers, vol. 92, no. 4, pp. 662–681, 2002

  22. [30]

    Anderson, Q

    P. Anderson, Q. Wu, D. Teney, J. Bruce, M. Johnson, N. S¨ underhauf, I. Reid, S. Gould, A. Van Den Hengel,Vision-and-language navigation: Interpreting visually- grounded navigation instructions in real environments, Proceedings of the IEEE Conference on Computer Vision and Pat...

  23. [31]

    Andrienko, G

    N. Andrienko, G. Andrienko,Exploratory analysis of spatial and temporal data: a systematic approach, Springer, 2006

  24. [32]

    Chiang, C

    Y.-Y. Chiang, C. A. Knoblock,Recognizing text in raster maps, GeoInformatica, vol. 19, no. 1, pp. 1–27, Springer, 2015

  25. [33]

    B. D. Dent, J. Torguson, T. Hodler,Cartography: Thematic Map Design, McGraw-Hill, 2008

  26. [34]

    G. F. Jenks,Generalization in statistical mapping, Annals of the Association of American Geographers, vol. 53, no. 1, pp. 15–26, Taylor & Francis, 1963

  27. [35]

    S. E. Kahou, V. Michalski, A. Atkinson, ´A. K´ ad´ ar, A. Trischler, Y. Ben- gio,Figureqa: An annotated figure dataset for visual reasoning, arXiv preprint arXiv:1710.07300, 2017

  28. [36]

    Krygier, D

    J. Krygier, D. Wood,Making maps, Guilford Publications, 2024

  29. [37]

    M. Wu, Y. Sun, X. Xia,Computational assessment of the aesthetic quality of map colors, Transactions in GIS, vol. 28, no. 6, pp. 1556–1576, Wiley Online Library, 2024

  30. [38]

    Martinez, A

    Th. Martinez, A. Hammoumi, G. Ducret, M. Moreaud, R. Deschamps, H. Pie- gay, J.-F. Berger,Deep learning ancient map segmentation to assess historical landscape changes, Journal of Maps, vol. 19, no. 1, p. 2225071, Taylor & Francis, 2023

  31. [39]

    Petitpierre, J

    R. Petitpierre, J. H. Uhl, I. Di Lenardo, F. Kaplan,A fragment-based approach for computing the long-term visual evolution of historical maps, Humanities and Social Sciences Communications, vol. 11, no. 1, pp. 1–18, Palgrave, 2024

  32. [40]

    X. Zhou, Y. Wen, Z. Shao, W. Li, K. Li, H. Li, X. Xie, Z. Yan,CartoMark: a benchmark dataset for map pattern recognition and map content retrieval with machine intelligence, Scientific Data, vol. 11, no. 1, p. 1205, Nature Publishing Group, 2024

  33. [41]

    G. Mai, C. Cundy, K. Choi, Y. Hu, N. Lao, S. Ermon,Towards a foundation 30 model for geospatial artificial intelligence (vision paper), Proceedings of the 30th International Conference on Advances in Geographic Information Systems, pp. 1–4, 2022

  34. [42]

    Y. Ji, S. Gao, Y. Nie, I. Maji´ c, K. Janowicz,Foundation models for geospatial reasoning: assessing the capabilities of large language models in understanding geometries and topological spatial relations, International Journal of Geographical Information Science, vol. 39, no....

  35. [43]

    Methani, P

    N. Methani, P. Ganguly, M. M. Khapra, P. Kumar,Plotqa: Reasoning over sci- entific plots, Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp. 1527–1536, 2020

  36. [44]

    C. Liu, Z. Wei, H. Liao, W. Dong,Can AI Observe Geographical Space? GVS- ABench For Evaluating Geo-Visuospatial Ability of Large Multimodal Models, Transactions in GIS, vol. 30, no. 1, p. e70189, Wiley Online Library, 2026

  37. [45]

    Zhang, W

    Y. Zhang, W. Zhang, Z. Zeng, K. Jiang, J. Li, W. Min, W. Luo, Q. Guan, J. Lin, W. Yu,MapReader: a framework for learning a visual language model for map analysis, International Journal of Geographical Information Science, vol. 39, no. 9, pp. 1973–2008, Taylor & Francis, 2025

  38. [46]

    D. Hong, B. Zhang, X. Li, Y. Li, C. Li, J. Yao, N. Yokoya, H. Li, P. Ghamisi, X. Jia, et al.,SpectralGPT: Spectral remote sensing foundation model, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 46, no. 8, pp. 5227–5244, IEEE, 2024

  39. [47]

    J. M. Olson,Noncontiguous area cartograms, The Professional Geographer, vol. 28, no. 4, pp. 371–380, 1976

  40. [48]

    R. G. Golledge,Wayfinding behavior: Cognitive mapping and other spatial pro- cesses, JHU Press, 1999

  41. [49]

    Chang, D

    S. Chang, D. Palzer, J. Li, E. Fosler-Lussier, N. Xiao,Mapqa: A dataset for question answering on choropleth maps, arXiv preprint arXiv:2211.08545, 2022. 31 Appendix A. Prompts Used in the Experiments The complete prompts used in all experimental settings are provided below. A...

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.