Does downscaling method matter? Assessing compound hot-dry event impacts on renewable power in Southern China

Recent summer power outages in China highlight the vulnerability of energy systems to compound hot-dry events (CHDE). These events strain power grids by increasing cooling demand while reducing renewable generation, particularly from hydropower. As China transitions toward a weather-dependent renewable energy system, understanding CHDE impacts becomes crucial for ensuring system resilience. However, current research on climate-energy interactions relies primarily on coarse global climate models and simplified empirical equations, potentially obscuring critical local-scale dynamics and technoeconomic factors relevant to adaptation planning. While high-resolution climate models better represent regional climate patterns, their advantages for energy system planning remain largely unexplored.

This study investigates how climate change-induced CHDE affects power generation and demand in Southern China, with particular focus on evaluating high-resolution downscaling approaches for energy planning. We validate two downscaling frameworks: a Generative Adversarial Network (GAN)-based statistical model and a dynamical regional climate model, by reproducing recent blackout events in 2022. We then apply both methods to downscale projections from MPI-ESM1-2-HR under SSP2-4.5 and SSP3-7.0 scenarios, generating 4-km, hourly resolution climate data for the near-term (2041–2060) and long-term (2081–2100) periods. Using these downscaled outputs, we estimate renewable power generation potential and temperature-driven electricity demand. We systematically quantify uncertainties arising from downscaling method choice and emission scenarios. Our findings demonstrate that climate data resolution significantly influences energy system planning outcomes and that rigorous uncertainty characterization across modeling chains is essential for robust climate impact assessments.

[Acknowledgement]

This research was supported by Research Grants Council of Hong Kong through Theme-based Research Scheme (T31-603/21-N) and General Research Fund (GRF16308722).