Pith. sign in

REVIEW 4 major objections 5 minor 44 references

Transport Network, Graph, and Air Pollution

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that the geometry and topology of a city's transport network carry measurable pollution signals visible from binary road maps alone, and that twelve graph indices can guide urban-planning changes to reduce air pollution.

desk verdict Impressive data assembly, weak inference: the width confound and missing statistics sink the causal claim, though the dataset could seed a real analysis. read the letter →

arxiv 2506.01164 v1 pith:G5ZS7GFT submitted 2025-06-01 physics.soc-ph cs.CV

Nan Xu This is my paper
classification physics.soc-phcs.CV
keywords complexnetworksgraphtheoryairpollutionpatternrecognitioncitiesmapstransportdeeplearning
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

The paper tries to establish that a city's air pollution is written into the geometry and topology of its transport network, independent of traffic, fuels, and other day-to-day factors. Using a conditional generative adversarial network (cGAN, a deep-learning colorization model) trained on roughly 300,000 street-map image tiles for about 1,700 cities, it claims to isolate pollution signals carried by network shape alone, then quantifies those signals with twelve graph-theoretic indices. If true, this gives planners a permanent-infrastructure lever: instead of only controlling vehicles, cities could redesign connectivity, compactness, walking paths, and orderliness to reduce pollution. The paper's stated conclusion is that improving almost all of the twelve indices, while avoiding extreme clustering, should improve air quality.

What carries the argument

The load-bearing machinery is the pairing of a cGAN image translator with twelve graph indices. The cGAN is trained on colour-coded maps in Lab space: the L channel carries only road geometry while the a&b channels carry pollution colours, so the trained network learns geometry-to-pollution associations; it is then applied to binary maps to prevent residual pollution information from leaking in. The twelve indices mix seven classic measures inherited from earlier transport-network research (mean degree, first Betti number, alpha and gamma indices, redundancy ratio, diameter, system dispersion) with five new ones (walk-to-drive path ratio, Ricci curvature, clustering coefficient, a scale-free index, and graph connectivity) that encode the geometric and small-world or scale-free patterns the author observed.

What would settle it

A matched-city comparison would settle it: if two cities with similar road density, size, land use, and weather but different street geometries produce cGAN predictions that do not track their actual NO2 and PM2.5 levels, the claimed standalone geometric effect fails.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that polluted cities tend to have transport networks that are poorly connected, low in redundancy, short on walking paths, structurally disordered, extreme in clustering coefficient, and scale-free in degree distribution, while cleaner cities show the opposite pattern on nearly every index. The image-interpretation step adds that walking paths, especially near abrupt shifts to drivable roads, and curvy or disorderly roads are the geometric elements the model most consistently marks as high-pollution. Because the inference stage feeds the cGAN binary black-and-white road maps rather than greyscale images that could leak pollution information, the paper argues the learned patterns represent a standalone geometric effect. As a graph-only study, it claims to separate the impact of permanent infrastructure from that of derived development, offering planners a more focused target for pollution reduction.

Load-bearing premise

The load-bearing premise is that the cGAN colourizations of binary road maps reveal a standalone effect of network geometry on pollution, not an artefact of road density, city size, land use, weather, or the road-type line widths.

Editorial extensions

If this is right

  • Raising connectivity (IDX 1–4 and IDX 12) is recommended as a pollution-reduction strategy, since poorly connected cities show a higher chance of high NO2.
  • Increasing the redundancy ratio is associated with cleaner air: in the sample, no city with redundancy ratio above 0.05 has NO2 above 20 ppb, and none above 0.09 has NO2 above 10 ppb.
  • Adding walking paths is associated with cleaner air: cities with NO2 above 20 ppb all have walk-to-drive path ratios below 15.7%.
  • Avoiding extreme clustering coefficients and scale-free degree distributions is associated with lower concentrations, so planners should steer away from clustering values near 1.
  • The twelve indices offer a monitorable, graph-only dashboard for tracking the network's contribution to air pollution without requiring detailed emission inventories.

Reading between the lines

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

  • Because the evidence is cross-sectional, a natural next step is to test the recommendations causally, such as comparing pollution before and after a city adds links, walking paths, or reorganisation that reduces clustering.
  • If the standalone geometric effect is real, it implies a ceiling on traffic-based measures: cleaning up vehicles alone may not clean the air while the permanent street skeleton stays the same.
  • The trained colorization model could be used as a counterfactual sketch tool: edit a city's binary road map, recolour it, and read off predicted pollution changes before any infrastructure is built.
  • The twelve indices could be packaged as an out-of-sample ranking instrument: score new cities from maps alone, then compare the ranking with monitor data to see whether the graph-only signal generalises beyond the 1,700 cities.
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

4 major / 5 minor

Summary. The manuscript proposes that static transport-network geometry and topology can be linked to urban air pollution. It trains a Pix2Pix cGAN to colorize binary road maps of about 1690 global cities using satellite-derived PM2.5 and NO2 concentrations, claims to identify geometric patterns (e.g., walking paths and curvy roads show higher concentrations), and then introduces 12 graph-based indices, plotting them against NO2 to derive recommendations for urban planning. The central claim is that these geometric and topological features have a standalone effect on pollution and can be improved to reduce air pollution.

Significance. If the central inference were established, the paper would contribute a novel, large-scale graph-based perspective to urban environmental planning and would be among the first to apply cGAN-based colorization to transport-network pollution analysis. The assembled dataset of 0.3 million image tiles from global cities, integrating raster and vector sources, is a valuable asset. However, the paper does not provide reproducible code, machine-checked proofs, parameter-free derivations, or falsifiable quantitative predictions, and the experimental design does not currently support the claimed separation of infrastructure geometry from other urban confounders. The significance of the contribution therefore depends on substantial additional validation.

major comments (4)
  1. [2.2, Table 1] The claimed isolation of a standalone effect of geometric patterns is confounded by road-type line widths. Table 1 specifies line widths for footways (0.5), service (2), residential (3), primary (5), and motorways (6) in the color training images, and the test inputs are described only as binary images with no statement about whether these widths are retained. If widths are retained, the cGAN can learn a width-class-to-pollution association, so the observation that walking paths are colored hotter than drivable roads (Section 2.3, Observation 1) does not identify a geometric effect. If widths are removed, the model has no visual information by which to identify walking paths, making that same observation inexplicable. The paper needs to specify the input construction and provide a control, such as uniform line width or width as a separate channel, before the geometric-pattern findings can be accepted.
  2. [2.3, Figures 3-11] The cGAN predictions are assessed only by visual inspection. No quantitative evaluation is reported: no per-pixel error, no correlation between predicted and ground-truth pollution colors, no spatial statistics, and no check that the colorized maps reproduce known city-specific pollution gradients. The claim that similar patterns appear for PM2.5 and NO2 (Section 2.3) is therefore subjective. Without numerical validation, the colorized outputs cannot be distinguished from systematic artifacts of the generator, which undermines Observations 1-3 and the indices IDX8 and IDX9 that are derived from them.
  3. [3, Figures 12-13 and Section 4] The 12-index analysis is descriptive only. The text reports threshold observations such as when the redundancy ratio exceeds 0.05 no city has NO2 above 20 ppb (IDX5) and that a high clustering coefficient equal to 1 is found to be correlated with high concentrations (IDX10), but no correlation coefficient, confidence interval, or hypothesis test accompanies these claims. There is also no adjustment for city size, population density, land use, weather, or traffic volume, all of which are plausible confounders. Consequently, the statement in Section 4 that we could improve air quality by increasing almost all IDXs is a causal and policy claim unsupported by the observational correlations presented.
  4. [3, Equations (9)-(11)] The construction of IDX8 and IDX9 is circular. The text states that these indices are designed to quantify two kinds of geometric patterns observed in image interpretation, that is, patterns inferred by a cGAN trained on the same pollution concentrations. Using these indices to confirm the existence of those patterns is not an independent test. No validation on held-out cities, comparison with a null model, or pre-registered pattern definition is provided. Similarly, IDX11 is a ratio of the least to most frequent degree rather than a fitted power-law exponent, so the statement that most cities could follow a scale-free pattern is not supported by the diagnostic that is actually plotted.
minor comments (5)
  1. [2.2] The statement that binary images are used to exclude any residual information about pollution that could be embedded in the greyscale images is unclear because binary images derived from the same rendering will still encode road-type line widths; please describe exactly how the binary test images are produced.
  2. [Equation (1)] Equation (1) writes RGB to L plus a&b, but Lab color space has three channels; please correct the notation and consistently use L, a, and b throughout.
  3. [3, IDX9, Equations (10)-(11)] The definition of kappa in Equations (10) and (11) is ambiguous: Equation (11) takes an argmax over i of P(kappa_E = kappa_i), but the notation does not specify the set of candidate values or how this empirical distribution is constructed; please clarify what is plotted in Figure 13C.
  4. [Caption of Figure 12] The caption reports 1691 city points for IDX 1-4 while the text says 1692 cities; please explain the missing point or correct the count.
  5. [Conclusion] The conclusion transitions from a correlational summary (could suggest) to a policy recommendation (we could improve air quality) without acknowledging the lack of causal identification; please add a caveat about the observational nature of the evidence.

Circularity Check

2 steps flagged · score 6.0 of 10

The cGAN 'geometry-only' inference reduces to the road-type width encoding, and IDX 8–9 restate the same cGAN pattern as confirmation.

  1. fitted input called prediction [Sec. 2.2 (Train & Test), Table 1, and Sec. 2.3 Observation 1]
    ""Binary images are used purposely in the test instead of the L channels to exclude any residual information about pollution that could be embedded in the greyscale images." Table 1: "Line Width 0.5: footway,steps,pedestrian,path,track; 2: service; 3: residential; 5: primary; 6: motorway." Sec. 2.3: "Roads for walking, for example, footpath, steps, pedestrians, etc., tend to show higher concentrations of pollution compared to drivable roads.""

    The cGAN is trained on images in which pollution is encoded as line color and road type is encoded as line width. The 'binary' test images are not stated to be skeletonized, so they can retain the very width code from Table 1. A network can then solve the task by learning width-to-color, so the finding that footways and pedestrian paths are colored 'hot' is a reproduction of the training association, not a standalone geometric effect. If widths are not retained, walking paths cannot be distinguished at all, making Observation 1 impossible. Either way, the claimed 'pollution induced solely by the effect of the geometry' reduces to a road-type width encoding or to an untestable assumption.

  2. renaming known result [Sec. 3, Table 2 (IDX 8–9) and Sec. 4 Conclusion]
    ""Equations (9) to (11) corresponding to IDX 8–9 are designed to quantify two kinds of geometric patterns observed in image interpretation" (Table 2 caption) and "Therefore, IDX 8 is introduced to examine the 1st type of patterns, and IDX 9 for the 3rd type." Conclusion: "inadequate paths for walking (IDX 8), disordered structure (IDX 9).""

    The indices are defined after, and explicitly to encode, the visual patterns read from the cGAN outputs. Presenting the later scatter plots of IDX 8 and IDX 9 against NO2 as confirmation of those same patterns is circular as evidence: the index is a numerical restatement of the visual observation, not an independent probe. The paper does not report an out-of-sample prediction or a test of whether an index chosen a priori, rather than post hoc from the same output, would show the association.

full rationale

The central derivation chain is: cGAN colorizes binary road maps, the colorized outputs are read as standalone geometric effects, and then indices IDX 8 and IDX 9 are built from those observed effects and used as supporting evidence. This chain is partially circular. The cGAN is trained on the same pollution data it is asked to explain, and the binary test inputs are never specified as width-normalized; if they retain the width code from Table 1, the 'walking paths are polluted' observation is an artifact of the width-to-color association already present in the training labels. If they do not retain widths, the observation is impossible. Moreover, IDX 8 and IDX 9 are explicitly designed to quantify the cGAN-observed patterns, so their correlation with NO2 is a re-description of the same pattern rather than independent confirmation. The graph-index scatter plots themselves are direct empirical associations and would not be circular if the indices were chosen a priori; no self-citation chain is load-bearing, and the data pipeline could support a properly controlled analysis. However, because the paper's headline claim about geometry-specific pollution effects rests on the cGAN outputs and post-hoc indices, the circularity score is elevated to 6.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central claim depends on domain assumptions about data completeness and representativeness, one ad hoc identification assumption that the cGAN isolates geometric effects, and two hand-chosen or fitted components: road line widths and the trained cGAN weights. No new physical entities are postulated.

free parameters (2)
  • Road line widths in cGAN input images = 0.5, 2, 3, 5, 6
    Chosen from the OSMnx tutorial, these widths encode road type into the binary images. The cGAN finding that walking roads show higher pollution depends on this encoding.
  • cGAN architecture and training schedule = Pix2Pix, 19 to 26 epochs, 10240x10240 then 1024x1024 tiles
    The model weights are fit to the training cities, and all subsequent geometric pattern findings are outputs of this fitted model.
assumptions (5)
  • domain assumption All city road networks are connected graphs with p=1
    Stated in Section 3 before Equation (2). Needed for the Betti number and connectivity formulas. Reasonable for cities but excludes edge cases.
  • domain assumption Annual satellite-derived NO2 and PM2.5 concentrations represent street-level pollution on road segments
    Used throughout as ground truth labels for roads, but the data are coarse gridded annual averages, not street-level measurements.
  • ad hoc to paper A cGAN trained on pollution-colored road images can isolate the standalone geometric effect on pollution from binary road maps
    Core identification assumption in Sections 2.2 and 2.3. No causal or statistical identification is provided, and line widths leak road type information.
  • domain assumption OpenStreetMap data are complete and comparable across 1700 global cities
    Graph indices are computed from OSM. Missing or heterogeneous mapping quality would bias the indices.
  • domain assumption The mode of Ricci curvature and clustering coefficient is a meaningful summary of network structure
    IDX 9 and IDX 10 use the most frequent value as a summary. This choice is not justified and discards the rest of the distribution.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Transport Network, Graph, and Air Pollution." pith.science (2026). https://pith.science/paper/G5ZS7GFT

@misc{pith2026250601164,
  author       = {Pith},
  title        = {Pith review of: Transport Network, Graph, and Air Pollution},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G5ZS7GFT}},
  note         = {Machine review of arXiv:2506.01164}
}
read the original abstract

Air pollution can be studied in the urban structure regulated by transport networks. Transport networks can be studied as geometric and topological graph characteristics through designed models. Current studies do not offer a comprehensive view as limited models with insufficient features are examined. Our study finds geometric patterns of pollution-indicated transport networks through 0.3 million image interpretations of global cities. These are then described as part of 12 indices to investigate the network-pollution correlation. Strategies such as improved connectivity, more balanced road types and the avoidance of extreme clustering coefficient are identified as beneficial for alleviated pollution. As a graph-only study, it informs superior urban planning by separating the impact of permanent infrastructure from that of derived development for a more focused and efficient effort toward pollution reduction.

Figures

Figures reproduced from arXiv: 2506.01164 by the authors.

Figure 1
Figure 1. Pollution-indicated transport network image interpretation. The training process is divided into two stages: one uses full-size images sampled as image A to learn the overall properties, and the other uses 1 10 -size images as in B, which are generated from A, to learn the details. The weights learnt from the training are used to test B&W images sampled as C in full and partial sizes. The resultant image will be com… view at source ↗
Figure 2
Figure 2. Loss curves from the training process of image interpretation. cGAN trains one generator and one discrim￾inator at the same time, the loss curves for the generator are aligned with the left coordinate; that for the discriminator are aligned with the right one. The generator loss curves for PM2.5 and NO2 are in green and blue respectively. The discriminator loss curves for PM2.5 and NO2 are in yellow and red respecti… view at source ↗
Figure 3
Figure 3. Results of inferences on 110 cities images for PM2.5 and NO2. Ground-truth images for 110 cities are listed in the 1 st column where A is for PM2.5 and D for NO2. Inferences on full-size images are listed in the 2nd column where B is for PM2.5 and E for NO2. Inferences on 1 10 -size images and then combined to form the full size are listed in the 3rd column where C is for PM2.5 and F for NO2. Similar patterns can be… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: PM2.5:Dortmund/DE [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 10
Figure 10. Figure 10: PM2.5:Hebi/CN [PITH_FULL_IMAGE:figures/full_fig_p004_10.png]
Figure 12
Figure 12. Figure 12: Transport networks in 12 indices: IDX 1–6. IDX 1–4 values of 1691 city points are plotted against NO2 as in A to D, in which a logarithmic scale is used in C and D to accommodate one extreme point. 1691 cities are the full range data for NO2 and 1689 cities are the fu…
Figure 13
Figure 13. Figure 13: Transport networks in 12 indices: IDX 7–12. IDX 7 and IDX 11 values of 952 city points are plotted against NO2 as in A and E. IDX 8, IDX 10, and IDX 12 values of 1691 city points are plotted against NO2 as in B, D, and F, in which a logarithmic scale is used in B to a…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

44 extracted references · 44 canonical work pages

  1. [1]

    Gregory, The Dictionary of Human Geography

    D. Gregory, The Dictionary of Human Geography . John Wiley & Sons, Incorporated, 2009

    work page 2009

  2. [2]

    W. L. Garrison and D. F. Marble, The Structure of Transportation Networks. Transportation Center at Northwestern University, 1961

    work page 1961

  3. [3]

    Avondo Bodino, Economic Applications of the Theory of Graphs

    G. Avondo Bodino, Economic Applications of the Theory of Graphs . Gordon and Breach, Science Publishers, 1962

    work page 1962

  4. [4]

    Kansky, Structure of Transportation Networks: Relationships Be- tween Network Geometry and Regional Characteristics

    K. Kansky, Structure of Transportation Networks: Relationships Be- tween Network Geometry and Regional Characteristics. University of Chicago, 1963

    work page 1963

  5. [5]

    Berge, The Theory of Graphs and Its Applications

    C. Berge, The Theory of Graphs and Its Applications . Greenwood Press, 1982

    work page 1982

  6. [6]

    J. D. Radke, Stochastic Models in Circuit Network Growth . Wilfrid Laurier University, 1977

    work page 1977

  7. [7]

    Giuliano and S

    G. Giuliano and S. Hanson, The Geography of Urban Transportation, Fourth Edition. Guilford Publications, 2017

    work page 2017

  8. [8]

    G. F. Newell, Traffic Flow on Transportation Networks. Cambridge, Mass.: MIT Press, 1980

    work page 1980

Show all 44 references
  1. [9]

    Vaughan, Urban Spatial Traffic Patterns

    R. Vaughan, Urban Spatial Traffic Patterns. Pion, London, 1987

    1987

  2. [10]

    Measuring the structure of road net- works,

    F. Xie and D. Levinson, “Measuring the structure of road net- works,” Geographical Analysis, vol. 39, p. 336–356, 2007

    2007

  3. [11]

    Information theory of complex networks: On evolution and architectural constraints,

    R. V . Sol ´e and S. Valverde, “Information theory of complex networks: On evolution and architectural constraints,” Complex Networks, vol. 650, pp. 189–207, 2004

    2004

  4. [12]

    Chung, Spectral Graph Theory

    F. Chung, Spectral Graph Theory. Conference Board of the Mathe- matical Sciences, 1997

    1997

  5. [13]

    Random graphs,

    B. Bollob ´as, “Random graphs,” Graph Theory. Graduate Texts in Mathematics, vol. 63, pp. 123–145, 1979

    1979

  6. [14]

    Collective dynamics of ‘small- world’ networks,

    D. J. Watts and S. H. Strogatz, “Collective dynamics of ‘small- world’ networks,” Nature, vol. 393, p. 440–442, 1998

    1998

  7. [15]

    Emergence of scaling in random networks,

    B. Albert-L ´aszl´o and A. R ´eka, “Emergence of scaling in random networks,” Science, vol. 286, p. 509–512, 1999

    1999

  8. [16]

    The structure and function of complex net- works,

    M. E. J. Newman, “The structure and function of complex net- works,” SIAM Review, vol. 45, pp. 167–256, 2003

    2003

  9. [17]

    Black, Transportation: A Geographical Analysis

    W. Black, Transportation: A Geographical Analysis. Guilford Publi- cations, 2003

    2003

  10. [18]

    Applications of graph theory and network science to transit network design,

    S. Derrible and C. Kennedy, “Applications of graph theory and network science to transit network design,” Transport Reviews , vol. 31, p. 495–519, 2011

    2011

  11. [19]

    A century of sprawl in the united states,

    C. Barrington-Leigh and A. Millard-Ball, “A century of sprawl in the united states,” Proceedings of the National Academy of Sciences , vol. 112, p. 8244–8249, 2015

    2015

  12. [20]

    Land use, transport, and population health: estimating the health benefits of compact cities,

    M. Stevenson, J. Thompson, T. H. de S ´a, R. Ewing, D. Mohan, R. McClure, I. Roberts, G. Tiwari, B. Giles-Corti, X. Sun, M. Wal- lace, and J. Woodcock, “Land use, transport, and population health: estimating the health benefits of compact cities,”The Lancet, vol. 388, p. 2925–...

    2016

  13. [21]

    Batty, The New Science of Cities

    M. Batty, The New Science of Cities. MIT Press, 2013

    2013

  14. [22]

    A chronology of global air quality,

    D. Fowler, P . Brimblecombe, J. Burrows, M. R. Heal, P . Grennfelt, D. S. Stevenson, A. Jowett, E. Nemitz, M. Coyle, X. Liu, Y. Chang, G. W. Fuller, M. A. Sutton, Z. Klimont, M. H. Unsworth, and M. Vieno, “A chronology of global air quality,” Philosophical Trans- actions of th...

    2020

  15. [23]

    The relationship between economic growth and air pollution—a regional comparison between china and south korea,

    M. Jiang, E. Kim, and Y. Woo, “The relationship between economic growth and air pollution—a regional comparison between china and south korea,” International Journal of Environmental Research and Public Health, vol. 17, p. 2761, 2020

    2020

  16. [24]

    The trans- port sector as a source of air pollution,

    R. Colvile, E. Hutchinson, J. Mindell, and R. Warren, “The trans- port sector as a source of air pollution,” Atmospheric Environment, vol. 35, p. 1537–1565, 2001. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 viii

    2001

  17. [25]

    Analysis of pollutants transport in heavy air pollution processes using a new complex-network-based model,

    M. Chen, Y. Chen, H. Zhu, Y. Wang, and Y. Xie, “Analysis of pollutants transport in heavy air pollution processes using a new complex-network-based model,” Atmospheric Environment, vol. 292, p. 119395, 2023

    2023

  18. [26]

    Doing more with less: How to design a good subgroup governance model for the air pollution transport network in “2+26

    W. Zeng, X. Chen, H. Dong, and Y. Liu, “Doing more with less: How to design a good subgroup governance model for the air pollution transport network in “2+26” cities of china?” Journal of Environmental Management, vol. 327, p. 116909, 2023

    2023

  19. [27]

    The impact of the covid-19 pandemic on air pol- lution: A global assessment using machine learning techniques,

    J. S. Wijnands, K. A. Nice, S. Seneviratne, J. Thompson, and M. Stevenson, “The impact of the covid-19 pandemic on air pol- lution: A global assessment using machine learning techniques,” Atmospheric Pollution Research, vol. 13, p. 101438, 2022

    2022

  20. [28]

    How road network trans- formation may be associated with reduced carbon emissions: An exploratory analysis of 19 major chinese cities,

    L. Zhang, H. Chen, S. Li, and Y. Liu, “How road network trans- formation may be associated with reduced carbon emissions: An exploratory analysis of 19 major chinese cities,” Sustainable Cities and Society, vol. 95, p. 104575, 2023

    2023

  21. [29]

    Air pollution prediction using machine learning techniques – an approach to replace existing monitoring stations with virtual monitoring stations,

    A. Samad, S. Garuda, U. Vogt, and B. Yang, “Air pollution prediction using machine learning techniques – an approach to replace existing monitoring stations with virtual monitoring stations,” Atmospheric Environment, vol. 310, p. 119987, 2023. [Online]. Available: https://www....

    2023

  22. [30]

    Image-to-image translation with conditional adversarial networks,

    P . Isola, J.-Y. Zhu, T. Zhou, and A. A. Efros, “Image-to-image translation with conditional adversarial networks,” in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017, pp. 5967–5976

    2017

  23. [31]

    How to colorize black & white photos with just 100 lines of neural network code,

    E. Wallner, “How to colorize black & white photos with just 100 lines of neural network code,” freeCodeCamp, Data Science. [Online]. Available: https://www.freecodecamp.org

  24. [32]

    Black and White Image Colorization with Deep Learning,

    A. Guilbert, “Black and White Image Colorization with Deep Learning,” Medium. [Online]. Available: https://anne- guilbert.medium.com

  25. [33]

    Satellite-derived pm2.5,

    A. van Donkelaar, “Satellite-derived pm2.5,” Washington Univer- sity in St. Louis, 2019

    2019

  26. [34]

    TROPOMI-derived ground level NO2 concentrations (2019 Annual mean),

    M. Cooper, “TROPOMI-derived ground level NO2 concentrations (2019 Annual mean),” Zenodo, 2022

    2019

  27. [35]

    Planet dump retrieved from https://planet.osm.org ,

    OpenStreetMap contributors, “Planet dump retrieved from https://planet.osm.org ,” OpenStreetMap. [Online]. Available: https://www.openstreetmap.org

  28. [36]

    General algorithm of assigning raster features to vector maps at any resolution, size or scale,

    N. Xu, M. Stevenson, and K. A. Nice, “General algorithm of assigning raster features to vector maps at any resolution, size or scale,” in 2024 IEEE International Conference on Future Machine Learning and Data Science (FMLDS), 2024, pp. 418–422

    2024

  29. [37]

    Modeling and Analyzing Urban Networks and Amenities with OSMnx,

    G. Boeing, “Modeling and Analyzing Urban Networks and Amenities with OSMnx,” University of Southern California, 2024. [Online]. Available: https://geoffboeing.com

    2024

  30. [38]

    Leveraging segment-anything model for automated zero-shot road width extraction from aerial imagery,

    N. Xu, K. Nice, S. Seneviratne, and M. Stevenson, “Leveraging segment-anything model for automated zero-shot road width extraction from aerial imagery,” in 2023 International Conference on Digital Image Computing: Techniques and Applications (DICTA), 2023, pp. 176–183

    2023

  31. [39]

    A rule-based tool for assisting colormap selection,

    L. Bergman, B. Rogowitz, and L. Treinish, “A rule-based tool for assisting colormap selection,” in Proceedings Visualization ’95, 1995, pp. 118–125

    1995

  32. [40]

    Tutorial: Graphriccicurvature

    C.-C. Ni, “Tutorial: Graphriccicurvature.” [Online]. Available: https://graphriccicurvature.readthedocs.io/

  33. [41]

    Clustering

    NetworkX, “Clustering.” [Online]. Available: https://networkx.org

  34. [42]

    Urban spatial order: street network orientation, con- figuration, and entropy,

    G. Boeing, “Urban spatial order: street network orientation, con- figuration, and entropy,” Appl Netw Sci, vol. 4, p. 67, 2019

    2019

  35. [43]

    Community detection on networks with ricci flow,

    C.-C. Ni, Y.-Y. Lin, F. Luo, and J. Gao, “Community detection on networks with ricci flow,” Scientific Reports, vol. 9, p. 9984, 2019

    2019

  36. [44]

    Measuring road network topology vulnerability by ricci curva- ture,

    L. Gao, X. Liu, Y. Liu, P . Wang, M. Deng, Q. Zhu, and H. Li, “Measuring road network topology vulnerability by ricci curva- ture,” Physica A: Statistical Mechanics and its Applications , vol. 527, p. 121071, 2019

    2019

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.