Research map / Vision-language and multimodal / Human and embodied modeling
UrbanDataLayer: A Unified Data Pipeline for Urban Science
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes UrbanDataLayer, a suite of standardized urban data structures (grid, graph, point, linestring, polygon) and pipelines for scheme transformation, granularity alignment, and feature fusion, demonstrated on four classic urban tasks.
Problem
Urban research is fragmented: studies use disparate datasets and heterogeneous formats/granularities, making benchmarks non-reproducible and multi-modal data fusion laborious.
Approach
A unified data-layer abstraction for urban data with APIs for converting raw data into five standard layers, aligning granularity, and fusing features across domains, so the same raw data can feed multiple downstream tasks without repeated custom processing.
Key results
- In PM2.5 prediction in Shanghai, combining all features with XGBoost reaches R2 0.378 versus 0.089-0.287 for single or pairwise features.
- Built-up area classification in New York with GBDT reaches accuracy 0.873 with all features versus 0.808 with nightlight alone.
- For El Nino anomaly detection, the best AUC-ROC is 0.540 (CoLA with SP+ZW+MW), with reported differences among feature combinations described as modest.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add road intersection density, nightlight, and population together (ablation) | XGBoost and MLP · PM2.5 prediction, Shanghai | RMSE: 3.953 -> 3.355; 3.710 -> 3.103 | Helped |
| Add road intersection density, nightlight, and population together (ablation) | XGBoost · PM2.5 prediction, New York | RMSE: 0.608 -> 0.578 | Helped |
| Add road intersection density, nightlight, and population together (ablation) | MLP · PM2.5 prediction, New York | RMSE: 0.619 -> 0.644 | Hurt |
| Add nightlight, SMOD, and population together (ablation) | GBDT · Built-up areas classification, Shanghai | Accuracy: 0.786 -> 0.794 | Helped |
| Add nightlight, SMOD, and population together (ablation) | RF · Built-up areas classification, Shanghai | Accuracy: 0.732 -> 0.772 | Helped |
| Add nightlight, SMOD, and population together (ablation) | DT · Built-up areas classification, New York | Accuracy: 0.820 -> 0.819 | Hurt |
| Add road network to POI features (ablation) | KNN · Boundary identification, Shanghai | F1: 0.608 -> 0.542 | Hurt |
| Add road network to POI features (ablation) | KNN · Boundary identification, New York City | F1: 0.557 -> 0.867 | Helped |
| Add population to POI and road network features (ablation) | KNN · Boundary identification, New York City | F1: 0.867 -> 0.914 | Helped |
| Add nightlight and population to POI and road network features (ablation) | GMM · Boundary identification, New York City | IOU: 0.511 -> 0.613 | Helped |
| Use spatial information, zonal winds, and meridional winds instead of all listed features (ablation) | CoLA · El Nino anomaly detection | AUC-ROC: 0.425 -> 0.540 | Helped |
| Use spatial information, sea surface temperature, and air temperature instead of all listed features (ablation) | CoLA · El Nino anomaly detection | AUC-ROC: 0.425 -> 0.542 | Helped |
Surprising findings
- On administrative boundary identification in Shanghai, adding auxiliary data (POI + Roadnet, etc.) did not help; POI alone performed best, unlike New York City.
Limitations
- Alignment limited to geospatial information; other data aspects not covered.
- Experiments use classic methods and do not innovate models; the cases only demonstrate feasibility.
- Only spatial/grid-based and limited regions (Shanghai, New York) evaluated; no global coverage yet.
Experimental setup
Datasets: nightlight data population data road network data POI data SMOD NASA Socioeconomic Data and Applications Center PM2.5 El Nino dataset administrative district boundaries
Benchmarks: PM2.5 concentration prediction (Shanghai, New York State) built-up area classification (Shanghai, New York State) administrative boundary identification (Shanghai, New York City) El Nino anomaly detection
Metrics: RMSE MAE R2 accuracy F1 AUC-ROC IOU
Baselines: LOF CoLA ANOMALOUS GAE OCGNN ONE
Builds on: LibCity OGC API standards
Closest papers by meaning
- UrbanFeel：A Comprehensive Benchmark for Temporal and Perceptual Understanding of City Scenes through Human Perspective (ICLR 2026 · similarity 0.87)
- UrbanVLP: Multi-Granularity Vision-Language Pretraining for Urban Socioeconomic Indicator Prediction (AAAI 2025 · similarity 0.87)
- CityLens: Evaluating Large Vision-Language Models for Urban Socioeconomic Sensing (ICLR 2026 · similarity 0.87)
- DynamicVL: Benchmarking Multimodal Large Language Models for Dynamic City Understanding (NeurIPS 2025 · similarity 0.86)
- UrbanIng-V2X: A Large-Scale Multi-Vehicle, Multi-Infrastructure Dataset Across Multiple Intersections for Cooperative Perception (NeurIPS 2025 · similarity 0.85)
- CityPulse: Fine-Grained Assessment of Urban Change with Street View Time Series (AAAI 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.