EnergAIze: Nesting step replacement experiment using the AI models

As part of the EnergAIze project, a nesting step replacement in the conventional downscaling chain was carried out. Climate models typically operate at coarse spatial resolution (100–250 km), which is insufficient for applications in complex terrains such as the Alps or for high-resolution renewable energy modeling. Traditional dynamical downscaling methods using nested Regional Climate Models (RCMs) are computationally demanding and time-intensive. This feasibility task investigated whether AI-based global weather prediction models could replace one or more nesting steps in a conventional downscaling chain. This approach is motivated by recent advances in AI-based weather prediction (AIWP) models, such as Pangu-Weather, Aurora, GraphCast, and FourCastNetv2, which offer fast, autoregressive forecast capabilities trained on high-resolution datasets (e.g., ERA5). These models can produce global weather forecasts at 0.25° resolution in a fraction of the runtime of NWP models, with promising skill across multiple atmospheric variables. The main goals were to: Evaluate the AI model ability to downscale CMIP6 data to ~0.25° ERA5-like resolution. Compare forecast and downscale modes for spatial/temporal enhancement. Assess whether output fields are physically consistent and suitable for RCM boundary conditions. Benchmark multiple AI architectures and operational setups.