Climate-reliable power systems via stochastic programming with scenarios generated with a posteriori time aggregation

The decarbonization of power systems requires managing intermittent renewable generation and climate variability while ensuring social welfare and reliability. However, incorporating extensive climate datasets into optimization models to capture variability compromises tractability, uncertainty traceability, or reliability. We propose a novel framework for climate-reliable power system design that integrates a bi-criteria a posteriori aggregation technique with a stochastic program that incorporates operational scenarios with a non-uniform number of representative days. This scenario construction enables efficient representation of extreme weather events while maintaining computational feasibility. To benchmark the framework, capacity portfolios are optimized separately for each climate dataset, distinguishing between climate scenario and climate model uncertainty. The approach is demonstrated using 47 climate projections for Saudi Arabia by 2050 under two policy settings: a 100 million tonnes of CO 2 emissions target and a fully renewable system. Results indicate substantial cost dispersion across climate models within the same emissions scenario, reaching up to 40% in the fully renewable system under the RCP4.5 climate scenario. Strong economic incentives for climate change mitigation are identified, primarily driven by the high sensitivity of electricity demand to rising temperatures. Uncertainty-aware designs with the proposed approach yield 3–6% higher design costs than single-climate designs. However, cross-scenario assessment reveals substantial supply shortfalls and excess emissions for single-climate designs, with potential additional costs that are at least 2.6 times the additional investment required by the proposed framework. Compared with stochastic programs having a uniform number of representative days, the proposed approach achieves lower unserved energy and lower worst-case evaluation costs with comparable total computational effort, demonstrating its value for climate-aware power system planning.