Carbon-Aware Digital-Twin Decision Support for Earthen Heritage: A Proof-of-Concept Case Study in Ushaiger, Saudi Arabia

Source abstract

Saudi Arabia’s expanding rehabilitation of historic places requires decision tools that connect monitored performance with conservation priorities and carbon consequences. This study develops a carbon-aware heritage decision twin for a traditional mud-brick dwelling in Ushaiger using indoor–outdoor monitoring from September 2025 to September 2026. Quality control yielded 10,757 unique paired records, including 10,755 observations on a 17,109-step half-hour grid. A hybrid piecewise cubic Hermite interpolating polynomial (PCHIP)–Extreme Gradient Boosting (XGBoost 3.4.1) reconstruction was tested with artificial gaps of 7–61 days; the mean indoor-temperature mean absolute error (MAE) increased from 0.93 to 1.35 °C across these gap lengths. XGBoost with SHapley Additive exPlanations (SHAP 0.52.0) associated indoor temperature with accumulated outdoor forcing, with 3 and 24 h means receiving the largest contributions in the revised model. Non-dominated Sorting Genetic Algorithm II (NSGA-II) optimization at a fixed effective ventilation coupling reduced degree-hours above 35 °C by 2.62%; a joint coupling–persistence sensitivity produced 0.62–6.91% under the tested assumptions. Of 124 hot days, 45 entered the operational-ventilation class, 35 the combined or alternative reversible-measure classes, and 44 escalation review. Threshold and reconstruction-error tests showed that these counts remain scenario dependent. A screening life-cycle assessment (LCA) resulted in 9.7–29.1 kg carbon dioxide equivalent (CO2e) for assisted ventilation versus 242.6–727.7 kg CO2e for localized cooling within a fixed 588 h comparison window. The framework combines evidence confidence, explainable artificial intelligence (AI), passive-cooling opportunity, life-cycle carbon, and heritage compatibility to prioritize the least intrusive adequate action. The framework is presented as a single-building proof of concept; its decision architecture may support future multi-building applications, but its numerical thresholds and intervention rules require validation across additional Saudi heritage buildings, climatic contexts, and monitoring locations before portfolio-scale deployment.