Metropolis-Scale Road Network Datasets for Fine-Grained Urban Traffic Modeling
Pith reviewed 2026-05-21 20:35 UTC · model grok-4.3
The pith
Datasets for two major cities supply up to 100,000 road segments with real connectivity and 5-minute traffic speed and volume measurements.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
We introduce datasets representing fine-grained road networks of two major cities, which are unique in their scale (up to 100,000 road segments), use of real road connectivity, presence of time series measurements for both traffic speed and volume at a 5-minute resolution, and inclusion of rich static road attributes. These datasets enable in-depth analysis of spatiotemporal traffic patterns and can serve as benchmarks for various ML applications. As a practical demonstration, we use them for traffic forecasting and propose a simple and efficient baseline that scales to large road graphs while achieving forecasting performance competitive with other established spatiotemporal models.
What carries the argument
Metropolis-scale road network datasets that combine real connectivity, rich static attributes, and 5-minute resolution time series for speed and volume, together with a simple scalable baseline model for forecasting on large graphs.
If this is right
- Current traffic forecasting models face significant scalability problems when applied to real city-sized road networks.
- A simple baseline can scale to these large graphs and still reach accuracy levels comparable to established spatiotemporal approaches.
- The datasets support detailed study of how traffic speed and volume evolve across connected road segments.
- The resources can function as standard benchmarks for testing machine learning methods in urban traffic tasks.
Where Pith is reading between the lines
- Researchers could combine the static road attributes with the time series to test whether segment properties improve prediction accuracy in specific neighborhoods.
- The datasets open the door to comparing model performance across different city layouts rather than on abstract or highway-only graphs.
- Extensions might add external signals such as weather or events to see how they affect forecasting at this resolution and scale.
Load-bearing premise
The datasets accurately capture complete real-world road connectivity and precise time-series measurements without gaps or errors, and the baseline achieves competitive results on the full graphs without hidden implementation choices.
What would settle it
Public release of the datasets followed by independent teams reproducing the forecasting experiments and confirming whether the simple baseline matches the accuracy of other spatiotemporal models on the full 100,000-segment graphs.
read the original abstract
Modeling traffic dynamics is a critical challenge for urban computing, with applications from real-time traffic management to infrastructure planning. However, progress in this area is fundamentally constrained by a lack of large-scale public datasets that capture the subtle properties of real city road networks. Existing benchmarks are often limited by their small scale, reliance on sparse highway traffic sensors, absence of true road connectivity information, and lack of information about road properties. To address this issue, we introduce datasets representing fine-grained road networks of two major cities, which are unique in their scale (up to 100,000 road segments), use of real road connectivity, presence of time series measurements for both traffic speed and volume at a 5-minute resolution, and inclusion of rich static road attributes. These datasets enable in-depth analysis of spatiotemporal traffic patterns and can serve as benchmarks for various ML applications. As a practical demonstration of the utility of our datasets and the challenges they present, we use them for the task of traffic forecasting. The size of the real-world road networks in our datasets reveals significant scalability issues in current traffic forecasting models. To address them, we propose a simple and efficient baseline that not only scales to large road graphs but also achieves forecasting performance competitive with other established spatiotemporal models. We hope that the proposed datasets will serve as a foundational resource for a broad range of research in traffic modeling, urban computing, and smart city development.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript introduces two new large-scale road network datasets for major cities, claimed to be unique in combining up to 100,000 road segments, real road connectivity, 5-minute resolution time series for both traffic speed and volume, and rich static road attributes. These datasets are used to illustrate scalability challenges in traffic forecasting, for which the authors propose a simple and efficient baseline that scales to large graphs and achieves performance competitive with established spatiotemporal models.
Significance. If the uniqueness of the dataset properties is verified through exhaustive comparisons and the baseline competitiveness is demonstrated with reproducible details, this work would provide a valuable public benchmark resource for urban computing and spatiotemporal ML, addressing gaps in scale, connectivity, and attribute richness in prior traffic datasets. The empirical release of real-world metropolis-scale data with joint speed-volume measurements is a clear strength that could enable more realistic model evaluations.
major comments (3)
- [Abstract] Abstract: The claim that the datasets are 'unique in their scale (up to 100,000 road segments), use of real road connectivity, presence of time series measurements for both traffic speed and volume at a 5-minute resolution, and inclusion of rich static road attributes' is central to the contribution but lacks an exhaustive side-by-side table comparing exact properties (scale, connectivity type, sensor coverage, attribute richness) against all cited prior benchmarks.
- [Traffic Forecasting] Traffic Forecasting section: The assertion that the baseline 'achieves forecasting performance competitive with other established spatiotemporal models' is load-bearing for the practical demonstration but provides no details on hyper-parameter settings, exact implementation, data splits, or full results with error bars, undermining verification of competitiveness.
- [Dataset Description] Dataset Description: Quantitative verification of claimed properties such as graph connectivity density, completeness of 5-minute sensor coverage, and distribution of static attributes is absent, which is necessary to substantiate advantages over existing benchmarks for fine-grained modeling.
minor comments (2)
- [Abstract] Abstract: Specify the names of the two major cities and the precise segment counts for each dataset to give readers immediate scale context.
- [References] References: Include precise citations and links for all compared benchmarks to facilitate independent verification of the uniqueness claims.
Simulated Author's Rebuttal
We thank the referee for the constructive and detailed feedback. We agree that the suggested additions will strengthen the manuscript's claims regarding dataset uniqueness, baseline competitiveness, and quantitative properties. We will revise the paper to incorporate an exhaustive comparison table, expanded experimental details, and supporting statistics.
read point-by-point responses
-
Referee: [Abstract] Abstract: The claim that the datasets are 'unique in their scale (up to 100,000 road segments), use of real road connectivity, presence of time series measurements for both traffic speed and volume at a 5-minute resolution, and inclusion of rich static road attributes' is central to the contribution but lacks an exhaustive side-by-side table comparing exact properties (scale, connectivity type, sensor coverage, attribute richness) against all cited prior benchmarks.
Authors: We agree that a side-by-side comparison table is needed to fully substantiate the uniqueness claim. In the revised manuscript we will add a comprehensive table in the Dataset Description section that directly compares our datasets to all cited prior benchmarks on scale, connectivity type, sensor coverage, and attribute richness. revision: yes
-
Referee: [Traffic Forecasting] Traffic Forecasting section: The assertion that the baseline 'achieves forecasting performance competitive with other established spatiotemporal models' is load-bearing for the practical demonstration but provides no details on hyper-parameter settings, exact implementation, data splits, or full results with error bars, undermining verification of competitiveness.
Authors: We acknowledge that additional implementation and reproducibility details are required. The revised Traffic Forecasting section will include hyper-parameter settings, exact implementation choices, data splits, and full results with error bars from multiple runs. We will also release the code and configuration files to support verification. revision: yes
-
Referee: [Dataset Description] Dataset Description: Quantitative verification of claimed properties such as graph connectivity density, completeness of 5-minute sensor coverage, and distribution of static attributes is absent, which is necessary to substantiate advantages over existing benchmarks for fine-grained modeling.
Authors: We will add the requested quantitative verifications to the Dataset Description section. This will include statistics on graph connectivity density, the fraction of the network with complete 5-minute sensor coverage, and distributions of the static road attributes, thereby providing empirical support for the claimed advantages. revision: yes
Circularity Check
No circularity: empirical dataset release and baseline with no derivation reducing to inputs
full rationale
The manuscript introduces new road network datasets and a scalable baseline for traffic forecasting. Its claims rest on empirical properties (scale, connectivity, sensor resolution, attributes) and experimental performance comparisons rather than any mathematical derivation, first-principles prediction, or equation that reduces to fitted parameters or self-referential definitions. Uniqueness assertions compare against prior benchmarks via citation; competitiveness is shown through reported experiments. Neither step matches the enumerated circularity patterns, as there are no load-bearing self-citations, ansatzes smuggled via prior work, or predictions equivalent to inputs by construction. The contribution is self-contained as a data resource and practical method proposal.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.