| _version_ | 1866902295163174912 |
|---|---|
| author | Moreira Beltrami, Gabriel Calafell, Joan Gonçalves dos Santos, Rogério Mateu Armengol, Jan |
| author_facet | Moreira Beltrami, Gabriel Calafell, Joan Gonçalves dos Santos, Rogério Mateu Armengol, Jan |
| contents | <h2>Dataset Description</h2> <p>This dataset was generated using Reynolds-Averaged Navier–Stokes (RANS) simulations to model pollutant dispersion in a 2D idealized street canyon. It contains <strong>49,590 samples</strong>, divided into training+validation and test subsets. Each sample represents a <strong>64×64 pollutant concentration field</strong>, produced as part of a <strong>data augmentation study for surrogate modeling</strong>.</p> <h3>Data Structure</h3> <ul> <li> <p><strong>Concentration fields</strong> are stored in NumPy <code>.npy</code> format and can be loaded using <code>numpy.load(file)</code>.<br>The arrays have shape:</p> <ul> <li> <p><code>(n_samples, 64, 64)</code> or</p> </li> <li> <p><code>(n_samples, 4096)</code>, which are interconvertible via <code>.reshape()</code>.</p> </li> </ul> </li> <li> <p><strong>Input parameters</strong> associated with each sample are stored in a separate <code>.npy</code> file and correspond positionally (i.e., the i-th row matches the i-th sample in the concentration dataset).</p> </li> <li> <p><strong>The mask file</strong> (in <code>DataCoordinates/Mask.npy</code>) identifies grid points inside building areas, which should be excluded from statistical analyses or visualizations of the street canyon.</p> </li> </ul> <h2>Input Parameters</h2> <p>Each sample is defined by 18 input parameters representing physical, turbulence, and emission characteristics. The values are <strong>normalized</strong>, and can be restored using the <code>inverse_norm_parameters</code> function provided in the code base. The parameters and their physical meaning are:</p> <ol> <li> <p><strong>Sct</strong> – Turbulent Schmidt number</p> </li> <li> <p><strong>Ce1</strong> – k−ε model parameter</p> </li> <li> <p><strong>Ce2</strong> – k−ε model parameter</p> </li> <li> <p><strong>Cμ</strong> – k−εk model parameter</p> </li> <li> <p><strong>σk</strong> – k−εk model parameter</p> </li> <li> <p><strong>B</strong> – k−ε model parameter</p> </li> <li> <p><strong>κ</strong> – von Kármán constant</p> </li> <li> <p><strong>Uτ</strong> – Friction velocity</p> </li> <li> <p><strong>y0</strong> – Aerodynamic roughness length</p> </li> <li> <p><strong>pk</strong> – TKE source scaling factor</p> </li> <li> <p><strong>pϵ</strong> – Dissipation rate source scaling factor</p> </li> <li> <p><strong>Bk</strong> – Background concentration offset</p> </li> <li> <p><strong>pB</strong> – Background concentration slope</p> </li> <li> <p><strong>Q</strong> – Emission rate</p> </li> <li> <p><strong>Qh</strong> – Emission source height</p> </li> <li> <p><strong>Qp</strong> – Heat release rate</p> </li> <li> <p><strong>θ</strong> – Solar incidence angle</p> </li> <li> <p><strong>ΔT</strong> – Background temperature increase</p> </li> </ol> <h2>Surrogate Models</h2> <p>Two surrogate modeling approaches based on deep learning are trained and evaluated using this dataset:</p> <ul> <li> <p><strong>PCA + MLP</strong>: A two-step model combining Principal Component Analysis (PCA) for dimensionality reduction and a Multi-Layer Perceptron (MLP) to predict the modal coefficients.</p> </li> <li> <p><strong>U-Net</strong>: A convolutional encoder–decoder architecture designed for structured image prediction.</p> </li> </ul> <p>Both models are optimized using the NSGA-II algorithm and included in the <code>TrainedModels/</code> directory.</p> <h2>Usage</h2> <p>To demonstrate how to load and evaluate the models using the dataset, we provide two example scripts:</p> <ul> <li> <p><code>MLPExample.py</code> – loads the MLP surrogate, reconstructs PCA-based fields, and plot a sample.</p> </li> <li> <p><code>UnetExample.py</code> – loads the Unet surrogate and plot a sample.</p> </li> </ul> <p>Each model requires loading associated parameter and coordinate data, as well as the building mask.</p> <p>For more details on the simulation setup, optimization, and evaluation strategy, please refer to the associated preprint <strong>Assessment of deep-learning strategies for surrogate modeling of pollution dispersion in a thermal street canyon</strong>, <strong>Beltrami, G., et al. 2025</strong>, submitted to computer & fluids.</p> <p><span>The authors acknowledge financial support from the AIR-URBAN project (TED2021-130210A-I00/ AEI/10.13039/501100011033/ European Union NextGenerationEU/PRTR).</span></p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_17055531 |
| institution | Zenodo |
| language | eng |
| publishDate | 2025 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | Dataset and Surrogate models based on 2D RANS thermal street canyon pollutant dispersion. Moreira Beltrami, Gabriel Calafell, Joan Gonçalves dos Santos, Rogério Mateu Armengol, Jan <h2>Dataset Description</h2> <p>This dataset was generated using Reynolds-Averaged Navier–Stokes (RANS) simulations to model pollutant dispersion in a 2D idealized street canyon. It contains <strong>49,590 samples</strong>, divided into training+validation and test subsets. Each sample represents a <strong>64×64 pollutant concentration field</strong>, produced as part of a <strong>data augmentation study for surrogate modeling</strong>.</p> <h3>Data Structure</h3> <ul> <li> <p><strong>Concentration fields</strong> are stored in NumPy <code>.npy</code> format and can be loaded using <code>numpy.load(file)</code>.<br>The arrays have shape:</p> <ul> <li> <p><code>(n_samples, 64, 64)</code> or</p> </li> <li> <p><code>(n_samples, 4096)</code>, which are interconvertible via <code>.reshape()</code>.</p> </li> </ul> </li> <li> <p><strong>Input parameters</strong> associated with each sample are stored in a separate <code>.npy</code> file and correspond positionally (i.e., the i-th row matches the i-th sample in the concentration dataset).</p> </li> <li> <p><strong>The mask file</strong> (in <code>DataCoordinates/Mask.npy</code>) identifies grid points inside building areas, which should be excluded from statistical analyses or visualizations of the street canyon.</p> </li> </ul> <h2>Input Parameters</h2> <p>Each sample is defined by 18 input parameters representing physical, turbulence, and emission characteristics. The values are <strong>normalized</strong>, and can be restored using the <code>inverse_norm_parameters</code> function provided in the code base. The parameters and their physical meaning are:</p> <ol> <li> <p><strong>Sct</strong> – Turbulent Schmidt number</p> </li> <li> <p><strong>Ce1</strong> – k−ε model parameter</p> </li> <li> <p><strong>Ce2</strong> – k−ε model parameter</p> </li> <li> <p><strong>Cμ</strong> – k−εk model parameter</p> </li> <li> <p><strong>σk</strong> – k−εk model parameter</p> </li> <li> <p><strong>B</strong> – k−ε model parameter</p> </li> <li> <p><strong>κ</strong> – von Kármán constant</p> </li> <li> <p><strong>Uτ</strong> – Friction velocity</p> </li> <li> <p><strong>y0</strong> – Aerodynamic roughness length</p> </li> <li> <p><strong>pk</strong> – TKE source scaling factor</p> </li> <li> <p><strong>pϵ</strong> – Dissipation rate source scaling factor</p> </li> <li> <p><strong>Bk</strong> – Background concentration offset</p> </li> <li> <p><strong>pB</strong> – Background concentration slope</p> </li> <li> <p><strong>Q</strong> – Emission rate</p> </li> <li> <p><strong>Qh</strong> – Emission source height</p> </li> <li> <p><strong>Qp</strong> – Heat release rate</p> </li> <li> <p><strong>θ</strong> – Solar incidence angle</p> </li> <li> <p><strong>ΔT</strong> – Background temperature increase</p> </li> </ol> <h2>Surrogate Models</h2> <p>Two surrogate modeling approaches based on deep learning are trained and evaluated using this dataset:</p> <ul> <li> <p><strong>PCA + MLP</strong>: A two-step model combining Principal Component Analysis (PCA) for dimensionality reduction and a Multi-Layer Perceptron (MLP) to predict the modal coefficients.</p> </li> <li> <p><strong>U-Net</strong>: A convolutional encoder–decoder architecture designed for structured image prediction.</p> </li> </ul> <p>Both models are optimized using the NSGA-II algorithm and included in the <code>TrainedModels/</code> directory.</p> <h2>Usage</h2> <p>To demonstrate how to load and evaluate the models using the dataset, we provide two example scripts:</p> <ul> <li> <p><code>MLPExample.py</code> – loads the MLP surrogate, reconstructs PCA-based fields, and plot a sample.</p> </li> <li> <p><code>UnetExample.py</code> – loads the Unet surrogate and plot a sample.</p> </li> </ul> <p>Each model requires loading associated parameter and coordinate data, as well as the building mask.</p> <p>For more details on the simulation setup, optimization, and evaluation strategy, please refer to the associated preprint <strong>Assessment of deep-learning strategies for surrogate modeling of pollution dispersion in a thermal street canyon</strong>, <strong>Beltrami, G., et al. 2025</strong>, submitted to computer & fluids.</p> <p><span>The authors acknowledge financial support from the AIR-URBAN project (TED2021-130210A-I00/ AEI/10.13039/501100011033/ European Union NextGenerationEU/PRTR).</span></p> |
| title | Dataset and Surrogate models based on 2D RANS thermal street canyon pollutant dispersion. |
| url | https://doi.org/10.5281/zenodo.17055531 |