Pith. sign in

REVIEW 4 major objections 6 minor 25 references

Improving Local Air Quality Predictions Using Transfer Learning on Satellite Data and Graph Neural Networks

T0 review · 4 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A GraphSAGE model pre-trained on London and fine-tuned on Bristol predicts hourly NO2 at unseen Bristol locations 8.6% more accurately than the same model trained only on Bristol.

desk verdict A clearly written workshop paper with a plausible transfer-learning recipe and open code, but the 'virtual sensor' evaluation is undercut by seeding each held-out location with its true first NO2 value. read the letter →

arxiv 2505.05479 v1 pith:PLWYWWND submitted 2025-04-23 eess.SP cs.LG

Finn Gueterbock , Raul Santos-Rodriguez , Jeffrey N. Clark This is my paper · ORCID
classification eess.SPcs.LG
keywords airqualitypredictionnitrogendioxide(NO2)transferlearninggraphneuralnetworksSAGEsatellitedataautoregressionvirtualsensors
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 claims that a graph neural network pre-trained on London's denser air-quality monitoring network and fine-tuned on Bristol can predict hourly nitrogen dioxide (NO2) levels at Bristol locations the model never saw, using satellite and weather data plus the previous hour's reading. Compared with a GraphSAGE model trained only on Bristol, transfer learning reduces NRMSE by 8.6% and Gradient RMSE by 32.6%, and the transferred model also beats MLP, XGBoost, and CNN baselines. If correct, this makes 'virtual sensors' plausible for cities with few permanent monitors: satellite columns and a handful of ground readings could stand in for a dense monitoring network, which matters because NO2 varies sharply street to street and missing monitors hide local pollution hot spots.

What carries the argument

The engine of the method is GraphSAGE with autoregression. GraphSAGE is an inductive graph neural network that samples each sensor's local neighborhood (up to two hops, sampling 3 then 5 neighbors) and aggregates node features, so it can make predictions for nodes not seen in training. To capture the strong hour-to-hour persistence of NO2, the previous hour's NO2 value at each node is fed in as a feature; at test time the predicted value is fed back into the next step. Transfer learning is done by pre-training on London's 112 stations and fine-tuning on Bristol's 8, and the input features combine satellite NO2 and aerosol columns, a gridded meteorological reanalysis, time features, and distance to major roads. The key comparison is leave-one-location-out on the 8 Bristol stations, so every Bristol location is eventually treated as an unseen 'virtual sensor' site.

What would settle it

Run the transferred model at held-out Bristol locations seeded with deliberately wrong first-hour values (e.g., ±10 µg/m³ around the true reading) and measure whether error compounds; or deploy temporary reference monitors at unmonitored sites and compare hourly predictions to actual readings—if the transfer advantage disappears under either test, the central claim fails.

Watch

Extended reading notes

Core claim

The central discovery claim is that transferring a GraphSAGE model from London to Bristol improves local NO2 prediction at unseen locations across all error metrics, with the largest gains in temporal smoothness. The transferred GraphSAGE achieves RMSE 15.623 µg/m3, NRMSE 0.481, and Gradient RMSE 6.354 µg/m3, versus 17.016, 0.526, and 9.426 for the Bristol-only GraphSAGE—an 8.6% NRMSE improvement and a 32.6% improvement in Gradient RMSE—while also outperforming MLP, XGBoost, and CNN baselines. The authors argue that pre-training on a data-rich city and fine-tuning on a data-scarce city is a viable route to high-resolution, location-specific NO2 estimates, provided some initial seed value at the target site is available.

Load-bearing premise

The evaluation assumes that a trustworthy actual NO2 reading is available at the target location for the first hour, so the model never has to start from a guess; without that seed, every subsequent prediction inherits whatever error the initial estimate carries.

Editorial extensions

If this is right

  • Cities with sparse monitoring networks could get hourly, street-level NO2 estimates for unmonitored locations without installing permanent stations.
  • Smoother predictions (32.6% lower Gradient RMSE) reduce spurious hour-to-hour jumps that would distort exposure and health impact calculations.
  • Transfer learning also improved the CNN baseline, so the benefit of pre-training on a data-rich city is not exclusive to graph-based models.
  • Leave-one-location-out evaluation across all 8 Bristol sites suggests the approach generalises to locations not included in the fine-tuning set, at least within the same conurbation.
  • The method's dependence on satellite NO2 columns and meteorological fields makes it portable to other regions where those products exist, including countries without dense monitoring.

Reading between the lines

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

  • The reported transfer gain may partly reflect that London and Bristol share similar traffic-dominated NO2 regimes and that the seed value leaks the target location's true level; the margin could shrink in cities with different emission sources, such as biomass burning or industrial areas.
  • The 32.6% Gradient-RMSE drop suggests transfer mostly improves the hour-to-hour shape of predictions rather than the mean level; combined with the paper's admitted under-prediction of high NO2 values, exposure estimates for peak pollution hours could still be biased even where rank order improves.
  • A natural testable extension is zero-shot transfer (freeze the London weights, never fine-tune on the target city) or fine-tuning on a single month of data, measuring how much of the 8.6% comes from the target city's fine-tuning signal rather than the London pre-training.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes a GraphSAGE-based model with autoregressive inputs, satellite and meteorological features, and transfer learning from London to Bristol, with the goal of predicting hourly NO2 concentrations at unmonitored locations. The central empirical claim is that a model pre-trained on London and fine-tuned on Bristol outperforms a Bristol-only GraphSAGE baseline, achieving an 8.6% reduction in NRMSE and a 32.6% reduction in Gradient RMSE when evaluated on eight held-out Bristol monitoring stations under leave-one-location-out cross-validation. The authors also compare against MLP, XGBoost, and CNN baselines, and report that the transferred GraphSAGE model achieves the best RMSE, NRMSE, and Grad-RMSE. The paper's stated motivation is the creation of 'virtual sensors' for data-scarce regions.

Significance. If the central claim held, the paper would provide a useful and concrete demonstration that transfer learning from a data-rich city can improve graph-based air quality prediction in a data-poor city at held-out locations. The work is clearly framed, the code is made publicly available, and the evaluation protocol is falsifiable in principle. The comparison of several architectures is useful for the community, and the use of externally sourced London data as a transfer source gives the empirical comparison a meaningful out-of-city component. However, the significance is currently limited by the small number of Bristol test locations (n = 8), the absence of uncertainty quantification, and, most importantly, an evaluation protocol that uses information unavailable in the stated deployment scenario.

major comments (4)
  1. [Appendix A.2] The evaluation protocol seeds each held-out location with the true NO2 value at the first timestep: 'During development of the model, this was achieved by including the actual NO2 value for the first timestep.' Because the model is autoregressive and NO2 is strongly autocorrelated (Figure 3), this single true observation can anchor the entire predicted trajectory. The central claim in the abstract and Section 6 concerns prediction at unmonitored locations, but the reported metrics in Tables 1 and 2 measure a hybrid problem in which the target location receives one true observation. The proposed alternatives (estimates from portable monitors or a guessed initial value) are not tested. This is load-bearing for the 'virtual sensor' claim and needs to be addressed by either evaluating with no true initial value, or by explicitly re-scoping the claim to locations with a single initial sample.
  2. [Section 3.2 and Appendix A.2] The test-time graph inputs for a held-out node include the previous NO2 values of that node and its neighbors, since the model 'learn[s] how to aggregate the satellite data, meteorological data and previous NO2 values for each node and its neighbours.' In the leave-one-location-out evaluation, the neighbors of the held-out Bristol location are the other seven Bristol monitoring stations, whose true NO2 values are available. In the deployment scenario of predicting at a genuinely unmonitored location, such ground-based observations at neighboring locations may not exist. The graph context used at test time is therefore more informative than what would be available in the intended application. Please evaluate using a graph in which neighbor NO2 values are also withheld or treated as unobserved, or clearly state that the method assumes neighboring monitoring exists.
  3. [Table 1 and Section 4] The performance comparison is based on a single average over only eight Bristol locations, with no confidence intervals, per-location error distributions, or significance tests reported. The headline improvement of 8.6% in NRMSE (Table 2) is small relative to the likely variability across only eight spatially correlated sites; the authors themselves note that nearby locations may exhibit correlated air quality and could lead to optimistic performance estimates under leave-one-out evaluation. At minimum, report per-location errors and a paired significance test (e.g., bootstrap or Wilcoxon signed-rank over locations) to demonstrate that the improvement is not within noise.
  4. [Section 4] The paper does not compare against a persistence or simple autoregressive baseline, even though the model is seeded with the true initial value and NO2 shows strong autocorrelation (Figure 3). A persistence model that repeats the last observed value would be a natural and necessary control, especially at short horizons. The paper defers 'benchmarking against a time-series model' to future work, but without such a baseline the reported improvements over the chosen baselines are difficult to interpret in a way that supports the claim of accurate prediction at unmonitored locations.
minor comments (6)
  1. [Abstract] The abstract states 'a 8.6% reduction in Normalised Root Mean Squared Error (NRMSE) and a 32.6% reduction in Gradient RMSE compared to a baseline model,' but it does not specify that the baseline is the non-transferred GraphSAGE model as opposed to, say, the best non-GNN baseline. Please name the baseline in the abstract.
  2. [Section 3.1] The text says the Bristol dataset includes '246,572 data points across 8 locations,' while the opening of Section 3.1 mentions data from 19 different locations. Please clarify how the 19 locations are reduced to 8 and whether the filtering introduces any selection bias.
  3. [Section 3.2] The description of the model input at test time is incomplete: it states that the model outputs hourly NO2 predictions, but does not explicitly state that the autoregressive feature is the previous hour's prediction once the model is deployed. Clarify the difference between training-time teacher forcing (using the true previous NO2) and test-time autoregression.
  4. [Section 4] The statement that the transferred GraphSAGE errors 'can be considered acceptable within the context of urban NO2 forecasting' is subjective and unsupported by a quantitative threshold or a comparison to regulatory limits or prior work with comparable protocols. Please provide a more explicit justification.
  5. [Figure 2] The y-axis labels in Figure 2 are not fully legible; please label the units (µg/m³) explicitly and ensure the captions indicate whether the plot shows hourly values over the full several-week period or a sub-sampled period.
  6. [Section 3.1] The satellite data are described as 'treated as static between daily measurements to match its temporal resolution,' but the model operates hourly. Please clarify the exact mechanism by which a daily value is assigned to each hour (e.g., forward fill across the day).

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the transfer-learning comparison is an external empirical benchmark; the true-NO2 seeding assumption is a deployment mismatch, not a circular derivation.

full rationale

The paper's central claim is empirical: a GraphSAGE model pre-trained on London and fine-tuned on Bristol is compared, at held-out Bristol locations, against the same and other models trained on Bristol alone. The reported reductions (8.6% NRMSE, 32.6% Gradient RMSE) are measured on locations not used in training, with London data external to the Bristol test set. No parameter is fitted to the held-out target values, and no derived quantity is defined in terms of the quantity it is supposed to predict. GraphSAGE, transfer learning, and the baselines are all standard methods cited from external literature; there is no load-bearing self-citation chain and no uniqueness theorem imported from the authors' prior work. The only notable concern is in Appendix A.2: for an unseen node, the model is initialised with the actual NO2 value at the first timestep, so the evaluation is not a test of a genuinely unmonitored location. The paper explicitly acknowledges this and suggests portable sampling or an estimate as alternatives. This is an external-validity limitation about the deployment scenario, not a circularity in the derivation: the model's outputs are not constructed to equal its inputs, and the baseline comparison is equally affected by the same seeding protocol. The internal computation and the relative transfer-learning comparison are therefore self-contained and do not reduce to the fitted inputs or to any self-citation.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

This is an empirical ML paper with no mathematical derivation; the load-bearing assumptions are data-level (feature sufficiency, autoregressive seed availability, graph validity, cross-city transfer). The model hyperparameters are free parameters chosen by hand.

free parameters (6)
  • GraphSAGE aggregator = mean pooling
    Chosen from four candidate aggregators in StellarGraph by comparison; affects how neighborhood information is combined.
  • Number of hops = 2
    Set to the maximum hops in the Bristol graph, noted in A.2.
  • Max sampled nodes per hop = 3 and 5
    Chosen manually in A.2.
  • Dropout rate = 0.5
    Fixed for GraphSAGE and MLP in A.2.
  • Learning rate and layer sizes = not reported
    The paper states these were 'selected by trial and error' without reporting values, so the exact configuration is a free parameter.
  • Number of trees for XGBoost = 100
    Baseline hyperparameter listed in A.2.
assumptions (4)
  • domain assumption Sentinel-5P NO2 column and aerosol index, ERA5-Land meteorology, and distance-to-road are sufficient predictive features for ground-level NO2.
    Used as model inputs in Section 3.1; the paper does not test whether omitting satellite data changes accuracy.
  • domain assumption Autoregression can be initialized with an actual NO2 reading at the target location, and this reading is available in deployment via cheap sampling schemes.
    Stated in A.2; without this seed the model must guess the first value, and errors propagate.
  • domain assumption Spatial proximity between sensors encodes useful correlation for NO2, so a graph over sensor locations is a valid representation.
    GraphSAGE in Section 3.2 relies on neighborhood links among sensor nodes.
  • domain assumption London's learned relationships transfer to Bristol despite differences in geography, traffic, and monitoring networks.
    The transfer-learning claim in Sections 3.2 and 4 depends on this.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Improving Local Air Quality Predictions Using Transfer Learning on Satellite Data and Graph Neural Networks." pith.science (2026). https://pith.science/paper/PLWYWWND

@misc{pith2026250505479,
  author       = {Pith},
  title        = {Pith review of: Improving Local Air Quality Predictions Using Transfer Learning on Satellite Data and Graph Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PLWYWWND}},
  note         = {Machine review of arXiv:2505.05479}
}
read the original abstract

Air pollution is a significant global health risk, contributing to millions of premature deaths annually. Nitrogen dioxide (NO2), a harmful pollutant, disproportionately affects urban areas where monitoring networks are often sparse. We propose a novel method for predicting NO2 concentrations at unmonitored locations using transfer learning with satellite and meteorological data. Leveraging the GraphSAGE framework, our approach integrates autoregression and transfer learning to enhance predictive accuracy in data-scarce regions like Bristol. Pre-trained on data from London, UK, our model achieves a 8.6% reduction in Normalised Root Mean Squared Error (NRMSE) and a 32.6% reduction in Gradient RMSE compared to a baseline model. This work demonstrates the potential of virtual sensors for cost-effective air quality monitoring, contributing to actionable insights for climate and health interventions.

Figures

Figures reproduced from arXiv: 2505.05479 by the authors.

Figure 1
Figure 1. Maps showing the spatial distribution of NO [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Actual and predicted NO2 values from the transferred GraphSAGE model for two locations in Bristol: (a) Well’s Road - a location with typically low NO2 values, and (b) Colston Avenue - a location with typically high NO2 values. While the results are promising, challenges remain, particularly around computational efficiency. The need to reconstruct the graph for each timestep imposes significant costs, which future wo… view at source ↗
Figure 3
Figure 3. NO2 values at time t vs NO2 values at time t + 1. A.2 MODEL TRAINING GraphSAGE Since the GraphSAGE algorithm (Hamilton et al., 2017) works by sampling from a node’s local neighbourhood of connected nodes, feeding in the data all at once in this way does not allow the model to see other timesteps in the future or the past, as the graphs at each timestep are not connected to each other. In order to address this issue … view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 19 canonical work pages

  1. [1]

    Air quality data continuous, 2022

    Bristol City Council . Air quality data continuous, 2022. URL https://opendata.bristol.gov.uk/explore/dataset/air-quality-data-continuous/information/?disjunctive.location. [Last accessed 28/10/22]

    work page 2022

  2. [2]

    Characterising low-cost sensors in highly portable platforms to quantify personal exposure in diverse environments

    Lia Chatzidiakou, Anika Krause, Olalekan AM Popoola, Andrea Di Antonio, Mike Kellaway, Yiqun Han, Freya A Squires, Teng Wang, Hanbin Zhang, Qi Wang, et al. Characterising low-cost sensors in highly portable platforms to quantify personal exposure in diverse environments. Atmospheric measurement techniques, 12 0 (8): 0 4643--4657, 2019

    work page 2019

  3. [3]

    Xgboost: A scalable tree boosting system

    Tianqi Chen and Carlos Guestrin. Xgboost: A scalable tree boosting system. In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining, pp.\ 785--794, 2016

    2016

  4. [4]

    The origins of kriging

    Noel Cressie. The origins of kriging. Mathematical geology, 22 0 (3): 0 239--252, 1990

    work page 1990

  5. [5]

    Stellargraph machine learning library

    CSIRO's Data61. Stellargraph machine learning library. https://github.com/stellargraph/stellargraph, 2018

    work page 2018

  6. [6]

    Deep learning estimation of daily ground-level no2 concentrations from remote sensing data

    Masoud Ghahremanloo, Yannic Lops, Yunsoo Choi, and Bijan Yeganeh. Deep learning estimation of daily ground-level no2 concentrations from remote sensing data. Journal of Geophysical Research: Atmospheres, 126 0 (21): 0 e2021JD034925, 2021

    work page 2021

  7. [7]

    Hamilton, Rex Ying, and Jure Leskovec

    William L. Hamilton, Rex Ying, and Jure Leskovec. Inductive representation learning on large graphs. In Proceedings of the 31st International Conference on Neural Information Processing Systems, NIPS'17, pp.\ 1025–1035, 2017. ISBN 9781510860964

    work page 2017

  8. [8]

    Transforming our world: implementing the 2030 agenda through sustainable development goal indicators

    Bandy X Lee, Finn Kjaerulf, Shannon Turner, Larry Cohen, Peter D Donnelly, Robert Muggah, Rachel Davis, Anna Realini, Berit Kieselbach, Lori Snyder MacGregor, et al. Transforming our world: implementing the 2030 agenda through sustainable development goal indicators. Journal of public health policy, 37: 0 13--31, 2016

    work page 2016

Show all 25 references
  1. [9]

    u ller, Thomas M \

    Jos Lelieveld, Andy Haines, Richard Burnett, Cathryn Tonne, Klaus Klingm \"u ller, Thomas M \"u nzel, and Andrea Pozzer. Air pollution deaths attributable to fossil fuels: observational and modelling study. bmj, 383, 2023

    2023

  2. [10]

    Improving air quality prediction accuracy at larger temporal resolutions using deep learning and transfer learning techniques

    Jun Ma, Jack CP Cheng, Changqing Lin, Yi Tan, and Jingcheng Zhang. Improving air quality prediction accuracy at larger temporal resolutions using deep learning and transfer learning techniques. Atmospheric Environment, 214: 0 116885, 2019

    2019

  3. [11]

    Application of random forest algorithm to predict the atmospheric concentration of no2

    Adven Masih. Application of random forest algorithm to predict the atmospheric concentration of no2. In 2019 Ural Symposium on Biomedical Engineering, Radioelectronics and Information Technology (USBEREIT), pp.\ 252--255. IEEE, 2019

    2019

  4. [12]

    London air quality network summary report, 2020

    Louise Mittal. London air quality network summary report, 2020. URL https://londonair.org.uk/london/reports/2020_LAQN_Report.pdf. [Last accessed 28/11/22]

    2020

  5. [13]

    Era5-land: A state-of-the-art global reanalysis dataset for land applications

    Joaqu \' n Mu \ n oz-Sabater, Emanuel Dutra, Anna Agust \' -Panareda, Cl \'e ment Albergel, Gabriele Arduini, Gianpaolo Balsamo, Souhail Boussetta, Margarita Choulga, Shaun Harrigan, Hans Hersbach, et al. Era5-land: A state-of-the-art global reanalysis dataset for land applica...

    2021

  6. [14]

    Predicting pm2

    Pratyush Muthukumar, Emmanuel Cocom, Kabir Nagrecha, Dawn Comer, Irene Burga, Jeremy Taub, Chisato Fukuda Calvert, Jeanne Holm, and Mohammad Pourhomayoun. Predicting pm2. 5 atmospheric air pollution using deep learning with meteorological data and ground-based observations and...

    2022

  7. [15]

    Os open roads

    Ordnance Survey . Os open roads. https://www.ordnancesurvey.co.uk/business-government/products/open-roads.html, 2022

    2022

  8. [16]

    Graph neural network-based spatiotemporal air quality prediction model with satellite remote sensing and meteorological data

    Junyu Qi, Bin Liu, Lei Zhang, Jie Huang, and Honglei Zhao. Graph neural network-based spatiotemporal air quality prediction model with satellite remote sensing and meteorological data. Environmental Pollution, 316: 0 120596, 2023

    2023

  9. [17]

    Local air – mapping the local environment using e‑scooters, 2023

    James Thomas and Sam Gunner. Local air – mapping the local environment using e‑scooters, 2023. URL https://www.localair.uk/

    2023

  10. [18]

    A hybrid deep learning approach for no _2 concentration forecasting combining transformer and lstm

    Yujie Wang, Shuang Li, Ke Zhang, Hongyu Liu, and Wei Huang. A hybrid deep learning approach for no _2 concentration forecasting combining transformer and lstm. Atmospheric Pollution Research, 14 0 (2): 0 101615, 2023

    2023

  11. [19]

    Air pollution and child health: prescribing clean air: summary

    World Health Organization and others . Air pollution and child health: prescribing clean air: summary. Technical report, World Health Organization, 2018

    2018

  12. [20]

    How powerful are graph neural networks? arXiv preprint arXiv:1810.00826, 2018

    Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural networks? arXiv preprint arXiv:1810.00826, 2018

    2018 arXiv

  13. [21]

    Deep transfer learning on satellite imagery improves air quality estimates in developing nations

    Nishant Yadav, Meytar Sorek-Hamer, Michael Von Pohle, Ata Akbari Asanjan, Adwait Sahasrabhojanee, Esra Suel, Raphael Arku, Violet Lingenfelter, Michael Brauer, Majid Ezzati, et al. Deep transfer learning on satellite imagery improves air quality estimates in developing nations...

    2022 arXiv

  14. [22]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

  15. [23]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should ...

  16. [24]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@firs...

  17. [25]

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibset...

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.