Construction workers are highly vulnerable to summer heat stress, yet thermal comfort assessment is often constrained by limited site-specific meteorological data. This study aims to develop a calibration framework that adjusts publicly available meteorological data to approximate on-site conditions and to evaluate thermal comfort indices for predicting workers’ thermal sensation.
Field data were collected from 341 construction workers at a residential site in South Korea during summer. Personal characteristics, subjective thermal responses, and environmental parameters were obtained through surveys and measurements. Public meteorological data were calibrated using linear regression, and the Physiological Equivalent Temperature (PET) and Universal Thermal Climate Index (UTCI) were calculated using public, calibrated, and field datasets. Machine learning regression models were applied to predict Mean Thermal Sensation Vote.
Calibration significantly improved agreement with field measurements, reducing RMSE from 1.61 °C to 0.84 °C for air temperature, from 8.05% to 3.18% for relative humidity, and from 1.28 m/s to 0.50 m/s for wind speed, with the largest improvement observed for humidity. Decision Tree Regression showed the highest explanatory power (R2 = 0.82 for PET; R2 = 0.86 for UTCI). UTCI outperformed PET, reflecting greater sensitivity to humidity and wind effects.
The study presents a low-cost, scalable framework enabling reliable, site-specific thermal comfort assessment using calibrated public data, supporting improved heat risk management in construction under climate change.
