Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Multimodal Physics-Informed Neural Network Approach for Mean Radiant Temperature Modeling

Pouya Shaeri, Saud AlKhaled, Ariane Middel

TL;DR

This work tackles outdoor thermal comfort assessment by estimating Mean Radiant Temperature $T_{mrt}$ in urban environments using a multimodal Physics-Informed Neural Network (PINN). The approach fuses meteorological data, built-environment descriptors, and fisheye-image shading to enforce the six-directional radiative transfer physics within a deep learning model. Key results show the PINN achieving RMSE of $3.50$ and $R^2$ of $0.88$, with shade prediction accuracy of $94\%$, and the ability to deliver competitive performance even with reduced input metadata by leveraging image-derived shading. The framework advances scalable, physically consistent urban microclimate assessment, offering a practical pathway for heat-health planning in hot desert cities and beyond.

Abstract

Outdoor thermal comfort is a critical determinant of urban livability, particularly in hot desert climates where extreme heat poses challenges to public health, energy consumption, and urban planning. Mean Radiant Temperature ($T_{mrt}$) is a key parameter for evaluating outdoor thermal comfort, especially in urban environments where radiation dynamics significantly impact human thermal exposure. Traditional methods of estimating $T_{mrt}$ rely on field measurements and computational simulations, both of which are resource intensive. This study introduces a Physics-Informed Neural Network (PINN) approach that integrates shortwave and longwave radiation modeling with deep learning techniques. By leveraging a multimodal dataset that includes meteorological data, built environment characteristics, and fisheye image-derived shading information, our model enhances predictive accuracy while maintaining physical consistency. Our experimental results demonstrate that the proposed PINN framework outperforms conventional deep learning models, with the best-performing configurations achieving an RMSE of 3.50 and an $R^2$ of 0.88. This approach highlights the potential of physics-informed machine learning in bridging the gap between computational modeling and real-world applications, offering a scalable and interpretable solution for urban thermal comfort assessments.

A Multimodal Physics-Informed Neural Network Approach for Mean Radiant Temperature Modeling

TL;DR

This work tackles outdoor thermal comfort assessment by estimating Mean Radiant Temperature in urban environments using a multimodal Physics-Informed Neural Network (PINN). The approach fuses meteorological data, built-environment descriptors, and fisheye-image shading to enforce the six-directional radiative transfer physics within a deep learning model. Key results show the PINN achieving RMSE of and of , with shade prediction accuracy of , and the ability to deliver competitive performance even with reduced input metadata by leveraging image-derived shading. The framework advances scalable, physically consistent urban microclimate assessment, offering a practical pathway for heat-health planning in hot desert cities and beyond.

Abstract

Outdoor thermal comfort is a critical determinant of urban livability, particularly in hot desert climates where extreme heat poses challenges to public health, energy consumption, and urban planning. Mean Radiant Temperature () is a key parameter for evaluating outdoor thermal comfort, especially in urban environments where radiation dynamics significantly impact human thermal exposure. Traditional methods of estimating rely on field measurements and computational simulations, both of which are resource intensive. This study introduces a Physics-Informed Neural Network (PINN) approach that integrates shortwave and longwave radiation modeling with deep learning techniques. By leveraging a multimodal dataset that includes meteorological data, built environment characteristics, and fisheye image-derived shading information, our model enhances predictive accuracy while maintaining physical consistency. Our experimental results demonstrate that the proposed PINN framework outperforms conventional deep learning models, with the best-performing configurations achieving an RMSE of 3.50 and an of 0.88. This approach highlights the potential of physics-informed machine learning in bridging the gap between computational modeling and real-world applications, offering a scalable and interpretable solution for urban thermal comfort assessments.

Paper Structure

This paper contains 19 sections, 9 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Urban Climate and Heat Assessment
  4. Physics-Informed Neural Networks (PINNs)
  5. Computer Vision for Radiation Estimation
  6. MaRTy Dataset
  7. Methodology
  8. Preprocessing of the MaRTy Data
  9. Architecture of the Deep Learning Model
  10. Baseline Neural Network (NN)
  11. Physics-Informed Neural Network (PINN)
  12. Multimodal PINN Architecture
  13. Evaluation Metrics of the Prediction Model
  14. Results
  15. Shade Prediction Accuracy
  16. ...and 4 more sections

Figures (4)

  • Figure 1: Overview of the proposed Physics-Informed Neural Network (PINN) workflow for $T_{mrt}$ estimation.
  • Figure 2: Sky mask generated using SegFormer on the fisheye image, distinguishing sun-exposed and shaded areas based on sun position
  • Figure 3: Physics-Informed Neural network architecture for $T_{mrt}$ estimation.
  • Figure 4: Comparison of RMSE for different machine learning and deep learning models.