Quantifying Climate Change Impacts on Renewable Energy Generation: A Super-Resolution Recurrent Diffusion Model

Driven by global climate change and the ongoing energy transition, the coupling between power supply capabilities and meteorological factors has become increasingly significant. Over the long term, accurately quantifying the power generation of renewable energy under the influence of climate change is essential for the development of sustainable power systems. However, due to interdisciplinary differences in data requirements, climate data often lacks the necessary hourly resolution to capture the short-term variability and uncertainties of renewable energy resources. To address this limitation, a super-resolution recurrent diffusion model (SRDM) has been developed to enhance the temporal resolution of climate data and model the short-term uncertainty. The SRDM incorporates a pre-trained decoder and a denoising network, that generates long-term, high-resolution climate data through a recurrent coupling mechanism. The high-resolution climate data is then converted into power value using the mechanism model, enabling the simulation of wind and photovoltaic (PV) power generation on future long-term scales. Case studies were conducted in the Ejina region of Inner Mongolia, China, using fifth-generation reanalysis (ERA5) and coupled model intercomparison project (CMIP6) data under two climate pathways: SSP126 and SSP585. The results demonstrate that the SRDM outperforms existing generative models in generating super-resolution climate data. Furthermore, the research highlights the estimation biases introduced when low-resolution climate data is used for power conversion.

💡 Research Summary

The paper addresses a critical gap in renewable energy modelling: the lack of hourly‑resolution climate data needed to capture short‑term variability and uncertainty of wind and solar resources under future climate change. To bridge this gap, the authors develop a Super‑Resolution Recurrent Diffusion Model (SRDM) that up‑scales daily climate outputs from global climate models (GCMs) to hourly resolution while preserving stochastic short‑term fluctuations.

SRDM builds on latent diffusion models (LDMs). A variational auto‑encoder (VAE) is first trained on high‑frequency climate time series to learn a low‑dimensional latent space where dimensions are decorrelated. The decoder part of this VAE is later reused as a pre‑trained component that maps latent vectors back to high‑resolution climate fields. The diffusion (denoising) network operates in this latent space, progressively removing Gaussian noise according to a learned schedule. Crucially, the model is “recurrent”: the high‑resolution output of day t‑1 is fed as the initial condition for day t, while the low‑resolution daily data serve as boundary conditions. This design guarantees temporal continuity and enables the generation of long, multi‑year sequences without breaking the Markovian structure of diffusion.

Mathematically, the model learns a conditional distribution p(x_hr | x_lr, x_hr^{t‑1}) where x_hr denotes hourly climate variables and x_lr denotes daily variables. The denoising network predicts the mean and covariance of the latent posterior at each diffusion step, while the decoder reconstructs the physical variables. Loss functions combine reconstruction error (including an L1‑based perceptual term) with a weighted KL‑divergence to balance fidelity and latent regularization.

Once hourly climate series are generated, they are fed into deterministic mechanism models for wind turbines and photovoltaic (PV) modules. These models implement physics‑based power curves h(·) that map wind speed, solar irradiance, temperature, etc., to instantaneous power output p_t = h(x_t). By applying the same conversion to both SRDM‑generated high‑resolution data and to the original low‑resolution data, the authors quantify the bias introduced when coarse climate data are used directly for power estimation.

The methodology is evaluated on the Ejina region of Inner Mongolia, China, using ERA5 reanalysis as a reference and CMIP6 projections under two Shared Socio‑Economic Pathways: SSP126 (low‑emission) and SSP585 (high‑emission). Results show that SRDM outperforms state‑of‑the‑art GAN‑based super‑resolution models and simple linear interpolation in terms of Mean Absolute Error (MAE) and Structural Similarity Index (SSIM). The model excels at reproducing extreme events (high wind gusts, peak solar irradiance), which are critical for grid reliability studies.

A key finding is that using daily‑averaged climate data for power conversion leads to systematic biases: annual wind power estimates can be off by 5–9 % and PV output by 7–12 % depending on the scenario. These discrepancies arise because low‑resolution data smooth out diurnal cycles and stochastic fluctuations that strongly affect capacity factor calculations. Consequently, long‑term planning studies that rely on coarse climate inputs may misjudge the need for storage, backup generation, or transmission expansion.

In summary, SRDM provides a computationally efficient, physically consistent framework to generate hourly climate fields from GCM outputs, preserving both long‑term trends and short‑term stochasticity. By coupling these fields with mechanistic power conversion models, the approach delivers more accurate forecasts of renewable generation under climate change, supporting better-informed policy and grid‑integration decisions. Future work suggested includes extending the model to spatial super‑resolution, incorporating additional climate variables (precipitation, temperature), and validating against high‑frequency observational datasets to further improve reliability.