As climate change (CC) continues to reshape environmental conditions, combining projected climate data into building performance analysis is crucial for resilient design.
This study investigates the influence of CC on energy consumption, economic outcomes, and simulation-derived thermal comfort indicators in a modern office building in Riyadh, Saudi Arabia, characterized by hot-arid conditions and high cooling demand.
Therefore, the XGBoost and LSTM results should be interpreted as measurement-informed predictions for historical energy consumption and simulation-informed predictions for future energy use and thermal comfort indicators; no field-measured thermal comfort data or occupant surveys were used.
Results indicate that the selected scenario weather inputs produced scenario-dependent and non-monotonic changes in total operational energy while shifting the simulation-derived PMV and PPD indices toward less favorable modeled thermal comfort conditions.
However, the Holm-adjusted corrected Diebold-Mariano tests did not identify statistically significant differences between the models for the three evaluated targets.
As climate change (CC) continues to reshape environmental conditions, combining projected climate data into building performance analysis is crucial for resilient design. This study investigates the influence of CC on energy consumption, economic outcomes, and simulation-derived thermal comfort indicators in a modern office building in Riyadh, Saudi Arabia, characterized by hot-arid conditions and high cooling demand. The single building case study design was adopted because this building provided a continuous 2009–2019 record of measured monthly electricity consumption together with sufficiently detailed information on its geometry, envelope, occupancy, operational schedules, and HVAC system for model calibration. Consequently, the findings are specific to this building and should not be generalized to the broader Riyadh office building stock without validation using additional buildings and independent datasets. A hybrid modeling approach was employed in which extreme gradient boosting (XGBoost) and long short-term memory (LSTM) models were combined with a calibrated EnergyPlus simulation workflow. Historical climate variables for 2009–2019 and scenario-based climate variables for 2020–2024, including temperature, solar radiation, and humidity, together with building-envelope parameters, were defined as inputs for the RCP 4.5 and RCP 6.0 analyses. For the historical modeling period, measured monthly electricity consumption was used as the energy target, whereas the PMV and PPD targets and the 2020–2024 scenario period energy outputs were generated by the calibrated EnergyPlus model. Therefore, the XGBoost and LSTM results should be interpreted as measurement-informed predictions for historical energy consumption and simulation-informed predictions for future energy use and thermal comfort indicators; no field-measured thermal comfort data or occupant surveys were used. ML models were assessed using a chronological holdout split and expanding-window rolling-origin validation, preserving the temporal order of the monthly observations while acknowledging the limited sample size and case-specific nature of the analysis. Results indicate that the selected scenario weather inputs produced scenario-dependent and non-monotonic changes in total operational energy while shifting the simulation-derived PMV and PPD indices toward less favorable modeled thermal comfort conditions. Within the internal chronological evaluation of this single-building dataset, LSTM yielded slightly higher R2 values and lower point estimates of prediction error than XGBoost for energy consumption and the EnergyPlus-generated PMV and PPD indices. However, the Holm-adjusted corrected Diebold-Mariano tests did not identify statistically significant differences between the models for the three evaluated targets. The findings are therefore interpreted as descriptive and case-specific rather than as evidence of general model superiority.
