Robust decision-making for minimizing building life-cycle GHG emissions and costs under uncertainties: Application to office buildings in China's cold regions

Uncertainties can substantially affect building life-cycle assessment (LCA) results. Existing optimization methods commonly adopt certainty assumptions and limited variables without sufficient robustness interpretation. This study develops a probabilistic decision-making framework for minimizing life-cycle greenhouse gas emissions (LCGHG) and life-cycle costs (LCC) under deterministic, mean, and 90th percentile (P90) risk metrics, which is applied to multistorey office building design in China's cold regions. The life-cycle performances of nine stock-derived form prototypes, combined with envelopes, HVAC, and rooftop photovoltaics, are explored under uncertain future climate, grid mix, discount rates, and component lifespans. Pareto-diagrams show that solutions obtained from different robustness criteria fall into two performance clusters mainly distinguished by HVAC energy sources. Yet deterministic and mean-based Pareto-fronts demonstrate materially higher upper-tail values of LCGHG and/or LCC. P90-based optimal ground-source heat pump solutions achieve LCGHG (90%) of 22.79–23.57 kg CO 2 -eq./(m 2 ·a) and LCC (90%) of 4558–5518 CNY/m 2 , whereas air-source heat pump solutions reduce costs by approximately 50% but increase emissions by about 30%. Within each cluster, building forms further differentiate performances, while envelopes and photovoltaics fine-tune outcomes. Design ranking consistency and performance distributional analyses indicate that solutions with lower P90 values generally correspond to lower medians and narrower variations. Moreover, P90-optimal solutions show similar operational emissions with trade-offs between equipment-related embodied emissions and costs. Sensitivity analysis identifies grid mix and HVAC lifespans as key drivers of LCGHG uncertainty, whereas discount rates and HVAC lifespans govern LCC variability. Although numerical findings are specific to the case, the proposed workflow and design guidance are transferable to other contexts.