[Paper Review] End-to-end data-driven prediction of urban airflow and pollutant dispersion

Nishant Kumar, Franck Kerhervé+2 more|arXiv (Cornell University)|Mar 18, 2026

Wind and Air Flow Studies0 citations

TL;DR

The paper presents a modular end-to-end data-driven reduced-order model to predict instantaneous and long-term urban canyon airflow and pollutant dispersion using SPOD, autoencoders, LSTM forecasting, and CNN-based velocity-to-scalar mapping, trained on LES data.

ABSTRACT

Climate change and the rapid growth of urban populations are intensifying environmental stresses within cities, making the behavior of urban atmospheric flows a critical factor in public health, energy use, and overall livability. This study targets to develop fast and accurate models of urban pollutant dispersion to support decision-makers, enabling them to implement mitigation measures in a timely and cost-effective manner. To reach this goal, an end-to-end data-driven approach is proposed to model and predict the airflow and pollutant dispersion in a street canyon in skimming flow regime. A series of time-resolved snapshots obtained from large eddy simulation (LES) serves as the database. The proposed framework is based on four fundamental steps. Firstly, a reduced basis is obtained by spectral proper orthogonal decomposition (SPOD) of the database. The projection of the time series snapshot data onto the SPOD modes (time-domain approach) provides the temporal coefficients of the dynamics. Secondly, a nonlinear compression of the temporal coefficients is performed by autoencoder to reduce further the dimensionality of the problem. Thirdly, a reduced-order model (ROM) is learned in the latent space using Long Short-Term Memory (LSTM) netowrks. Finally, the pollutant dispersion is estimated from the predicted velocity field through convolutional neural network that maps both fields. The results demonstrate the efficacy of the model in predicting the instantaneous as well as statistically stationary fields over long time horizon.

Motivation & Objective

• Motivate fast and accurate prediction of urban airflow and pollutant dispersion for decision-making under urban growth and climate change pressures.
• Develop a modular end-to-end framework that preserves key physics while enabling real-time or rapid parametric studies.
• Combine spectral POD with nonlinear dimensionality reduction and nonlinear time forecasting to handle multiscale urban flow dynamics.
• Demonstrate the capability to predict both instantaneous flow fields and long-term statistics in skimming-flow street canyons.

Proposed method

• Construct a reduced basis via spectral proper orthogonal decomposition (SPOD) of LES data to extract spectrally coherent flow structures.
• Compress SPOD coefficients with a dense autoencoder to obtain a compact latent representation.
• Forecast the latent space with a Long Short-Term Memory (LSTM) network to predict temporal evolution.
• Reconstruct the flow field and estimate pollutant concentration by mapping predicted velocity fields to scalar fields using a convolutional neural network (CNN).
• Integrate the four components into an end-to-end data-driven ROM capable of reproducing instantaneous and statistically stationary fields over long horizons.

Experimental results

Research questions

• RQ1 Can SPOD effectively isolate coherent structures in urban canyon flows and provide a parsimonious set of features for reduced-order modeling?
• RQ2 Does an autoencoder sufficiently compress SPOD coefficients to a low-dimensional latent space while preserving predictive fidelity?
• RQ3 Can LSTM-based forecasting accurately predict the nonlinear temporal evolution of the latent space for long-horizon predictions?
• RQ4 Can a CNN reliably map velocity fields to pollutant concentration fields to reproduce dispersion patterns from the predicted flow?

Key findings

• The framework successfully predicts both instantaneous and long-horizon statistical fields of urban canyon flows.
• Dimensionality reduction via SPOD plus nonlinear autoencoding yields a compact latent space suitable for time forecasting.
• LSTM-based forecasting in the latent space captures nonlinear temporal dynamics and enables accurate reconstruction of the velocity field.
• CNN mapping from predicted velocity to concentration effectively reconstructs pollutant dispersion patterns from the ROM.
• The modular architecture demonstrates efficacy in reproducing high-fidelity LES results while reducing computational costs.