High-Resolution Dataset of Time-Integrated 3D Urban Pollutant Concentration Fields

📄 Dataset Description

This dataset contains synthetic, time-integrated 3D concentration fields of an air pollutant in urban environments, generated using the PMSS (Parallel Micro Swift Spray) atmospheric dispersion modeling system. The data are provided as NetCDF4 files, representing pollutant concentration fields integrated over a two-hour period for various atmospheric and emission source scenarios. 

The dataset totals approximately 165 GB of compressed data and includes multiple simulation instances, each corresponding to a specific initialization defined by:

• A hypothetical point source emitting a unit mass of pollutant (gas or particulate matter) at 2 m height

• One of 108 stationary weather conditions, derived from combinations of wind directions (θ ∈ {0°, 10°, ..., 350°}) and wind speeds (v ∈ {1.5, 3.5, 6} m/s)

• Urban geometries based on two cities: Grenoble and Paris, each provided as a separate compressed folder.

📦 Folder Structure

The dataset is distributed in split archive format due to its large size, with the following compressed folders:

• raw_paris.tar.gz.part_aa ... raw_paris.tar.gz.part_af➡ produce raw_paris.tar.gz once reconstructed
• raw_grenoble.tar.gz.part_aa ... raw_grenoble.tar.gz.part_ak ➡ produce raw_grenoble.tar.gz once reconstructed

Each archive, once reconstructed and extracted, contains:

🔹 Integrated concentration fields (.nc)

Time-integrated 3D concentration fields simulated for each scenario. File naming convention:

• DXXX: Wind direction in degrees

• UYYY: Wind speed in m/s

• PTZZZ: Emission source reference index

🔹 Source coordinates

• PointsRejets_rel.csv: CSV file listing the relative X and Y coordinates of the emission sources. Each row corresponds to a unique emission source index.

• Reference coordinate systems:

  • Grenoble: (912000, 6456000)

  • Paris: (649000, 6862000)

• Emission height is constant across all experiments.

⚠️ Note for Grenoble:

• An additional file, PointsRejets_rel_new.csv, is provided and should be used only for simulations where the wind direction DXXX is such that DXXX / 10 is odd (i.e., DXXX ∈ {10, 30, ..., 270}).

• While the emission source locations remain unchanged, the indices in PointsRejets_rel_new.csv differ from those in PointsRejets_rel.csv.

🔹 Urban building maps (indice_building_grenoble.nc / indice_building_paris.nc)

NetCDF files encoding the 3D urban environment as a matrix:

• A value of 0 indicates no obstacle

• A positive value z represents the height of an obstacle

🖥️ Extraction Instructions

The archives are split for upload convenience. You must **recombine and extract them** before using the dataset.

🐧 On Linux/macOS:

cat raw_paris.tar.gz.part_* > raw_paris.tar.gz
tar -xvzf raw_paris.tar.gz

cat raw_grenoble.tar.gz.part_* > raw_grenoble.tar.gz
tar -xvzf raw_grenoble.tar.gz

🪟 On Windows (PowerShell):

Get-Content raw_paris.tar.gz.part_* -Encoding Byte -ReadCount 0 | Set-Content -Encoding Byte raw_paris.tar.gz
tar -xvzf raw_paris.tar.gz

Get-Content raw_grenoble.tar.gz.part_* -Encoding Byte -ReadCount 0 | Set-Content -Encoding Byte raw_grenoble.tar.gz
tar -xvzf raw_grenoble.tar.gz

🧠 Technical Details

The simulations were conducted using high-performance computing (HPC) resources with the PMSS modeling system, developed by CEA and ARIA Technologies. The model computes:

• Steady-state 3D wind fields

• Transient 3D pollutant dispersion fields

The computational domain is defined with a high-resolution mesh:

• 2 m horizontal and vertical resolution up to building height, over a ~1 km² surface area

• Progressively coarsening vertical resolution above rooftops, while maintaining 2 m horizontal resolution

📚 Applications

This dataset is suitable for:

• Training and benchmarking scientific machine learning models

• Studying urban pollutant dispersion

• Evaluating model accuracy in atmospheric physics and air quality modeling