Data portal · Method
Machine-learning and surrogate model datasets for urban flow
Datasets built to train and test machine-learning surrogates of urban wind, dispersion and microclimate, together with trained models and code. Most pair large ensembles of CFD or LES runs with the geometry that produced them, so that models can be fitted and compared on common data.
- 18 resources
- 1 field
- 17 simulation
- 1 benchmark
All 18 resources, largest first
-
High-resolution synthetic 3D concentration fields of an air pollutant in urban environments. Useful for ML training and dispersion model comparison.
Simulation Simulation output ML / surrogate
-
Benchmark package for source inversion of urban pollution dispersion using building geometry and sparse sensors, including PPMLES/MUST summaries. Useful for testing inversion and sensing methods.
Benchmark Simulation Case setupSimulation output LESML / surrogate
-
Data for diffusion-based sensor placement and reconstruction of turbulent flows in urban environments from DNS. Useful for ML reconstruction.
Simulation Simulation output DNSML / surrogate
-
Simulated plume concentration data in urban areas used to train the CityTransformer neural network. Useful for ML surrogates of urban dispersion.
Simulation Simulation output ML / surrogateLES
-
Example CFD simulation and code for distributed training of a U-Net predicting urban aerodynamic flow and pollutant dispersion. Useful for ML surrogates of urban flow.
Simulation Simulation outputSoftware & tools ML / surrogateRANS
-
RANS simulations of thermal street canyon pollutant dispersion with surrogate models.
Collection · 2 parts Simulation Simulation outputSoftware & tools RANSML / surrogate
-
Dataset of microscale atmospheric flow simulations around randomly generated urban geometries for several stability classes. Useful for training and testing ML surrogates of urban flow.
Simulation Simulation output RANSML / surrogate
-
Training and testing data and code for UCNN, a deep-learning urban canopy model for complex building morphologies.
Simulation Software & toolsSimulation output ML / surrogate
-
41 PALM NetCDF files of building-resolved wind fields for Camden, London, for scientific machine learning.
Simulation Simulation output LESML / surrogate
-
Open-source Python package for fast urban wind-field prediction from geometry.
Simulation Software & tools ML / surrogate
-
Field and CFD data for physics-informed surrogate modelling of urban wind under sparse measurements.
Field MeasurementsSimulation output RANSML / surrogate
-
Code for diffusion-based sensor placement and reconstruction of turbulent flows in urban environments. Useful for flow reconstruction and sensor design.
Simulation Software & tools ML / surrogate
-
Paper release of an agentic AI workflow that automates CFD wind simulations for buildings and cities. Useful for automated case setup.
Simulation Software & tools ML / surrogate
-
Deep learning model code predicting 3D wind speed fields in cities from building geometry.
Simulation Software & tools ML / surrogate
-
Python code for regression and machine-learning models predicting area-averaged external pressure coefficients on low-rise buildings. Useful for wind-load surrogates.
Simulation Software & tools ML / surrogate
-
Data and scripts for a fast surrogate wind model over heterogeneous urban surfaces and terrain. Useful for rapid urban wind simulation.
Simulation Software & toolsSimulation output ML / surrogate
-
UrbanFlowNet-MFSS, a physics-regularised multi-fidelity surrogate for urban flow reconstruction.
Simulation Software & tools ML / surrogate
-
LES velocity fields over urban canopies with inputs and U-net predicted fields, including perturbed cases. Useful for training and testing ML surrogates of urban flow.
Simulation Simulation output LESML / surrogate