In recent years traditional numerical methods for accurate weather prediction have been increasingly challenged by deep learning methods. Numerous historical datasets used for short and medium-range weather forecasts are typically organized into a regular spatial grid structure. This arrangement closely resembles images: each weather variable can be visualized as a map or, when considering the temporal axis, as a video. Several classes of generative models, comprising Generative Adversarial Networks, Variational Autoencoders, or the recent Denoising Diffusion Models have largely proved their applicability to the next-frame prediction problem, and is thus natural to test their performance on the weather prediction benchmarks. Diffusion models are particularly appealing in this context, due to the intrinsically probabilistic nature of weather forecasting: what we are really interested to model is the probability distribution of weather indicators, whose expected value is the most likely prediction. In our study, we focus on a specific subset of the ERA-5 dataset, which includes hourly data pertaining to Central Europe from the years 2016 to 2021. Within this context, we examine the efficacy of diffusion models in handling the task of precipitation nowcasting. Our work is conducted in comparison to the performance of well-established U-Net models, as documented in the existing literature. Our proposed approach of Generative Ensemble Diffusion (GED) utilizes a diffusion model to generate a set of possible weather scenarios which are then amalgamated into a probable prediction via the use of a post-processing network. This approach, in comparison to recent deep learning models, substantially outperformed them in terms of overall performance.

Urban building energy modeling (UBEM) is crucial for assessing energy consumption patterns at the city-scale and for supporting data driven planning and decarbonization strategies. However, its practical deployment is often hindered by the need to balance detailed physics-based simulations with acceptable computation times when thousands of buildings are involved. This work presents a large-scale real world UBEM case study and proposes a workflow that combines EnergyPlus simulations, high-performance computing (HPC), and open urban datasets to model the energy consumption of the building stock in the Municipality of Bologna, Italy. Geometric data such as building footprints and heights were acquired from the Bologna Open Data portal and complemented by aerial light detection and ranging (LiDAR) measurements to refine elevations and roof geometries. Non-geometrical building characteristics, including wall materials, insulation levels, and window properties, were derived from local building regulations and the European TABULA project, enabling the assignment of archetypes in contexts where granular information about building materials is not available. The pipeline’s modular design allows us to analyze different combinations of retrofitting scenarios, making it possible to identify the groups of buildings that would benefit the most. A key feature of the workflow is the use of Leonardo, the supercomputer hosted and managed by Cineca, which made it possible to simulate the energy consumption of approximately 25,000 buildings in less than 30 min. In contrast to approaches that mainly reduce computation time by simplifying the physical model or aggregating representative buildings, the HPC-based workflow allows the entire building stock to be individually simulated (within the intrinsic simplifications of UBEM) without introducing further compromises in model detail. Overall, this case study demonstrates that the combination of open data and HPC-accelerated UBEM can deliver city-scale energy simulations that are both computationally tractable and sufficiently detailed to inform municipal decision-making and future digital twin applications.