A monthly-adaptive sky temperature model for the Mediterranean area based on symbolic regression: development, validation and impact on building energy simulations

Accurate estimation of sky temperature is essential for building energy simulations, especially in climates characterized by significant seasonal variability. Most existing empirical models have been developed using limited-site datasets and do not adapt to monthly or regional atmospheric dynamics, reducing their reliability when applied outside the original calibration conditions. This study proposes a novel monthly-adaptive sky temperature model based on symbolic regression, trained using climatic data collected in 2023 from nine micrometeorological stations in the Lazio region (Italy). A key feature is the inclusion of the month of the year as a predictor, enabling the model to capture seasonal behavior typical of Mediterranean climates. The model’s accuracy was assessed by comparison with experimental sky temperatures derived from longwave radiation measurements. Additional tests using data from previous years (2019, 2021, 2022) confirmed its temporal robustness. The proposed formulation consistently outperformed widely used empirical models. When integrated into a TRNSYS-based dynamic simulation of two representative building archetypes, it led to significant variations in heating and cooling energy demand compared to reference formulations such as ISO 13790 and Berdahl & Martin. Using alternative models instead of the proposed one resulted in heating energy demands up to 43 % lower and cooling demand overestimations exceeding 100%, depending on the correlation adopted. These findings demonstrate that data-driven, seasonally adaptive models tailored to specific climate conditions can significantly enhance the reliability of building energy simulations. Future work will focus on extending the model’s applicability across climatic zones and evaluating integration into standardized simulation frameworks.