Francesco Immorlano

Gabriele Accarino, Davide Donno, Francesco Immorlano et al.

Tropical Cyclones (TCs) are counted among the most destructive phenomena that can be found in nature. Every year, globally an average of 90 TCs occur over tropical waters, and global warming is making them stronger, larger and more destructive. The accurate detection and tracking of such phenomena have become a relevant and interesting area of research in weather and climate science. Traditionally, TCs have been identified in large climate datasets through the use of deterministic tracking schemes that rely on subjective thresholds. Machine Learning (ML) models can complement deterministic approaches due to their ability to capture the mapping between the input climatic drivers and the geographical position of the TC center from the available data. This study presents a ML ensemble approach for locating TC center coordinates, embedding both TC classification and localization in a single end-to-end learning task. The ensemble combines TC center estimates of different ML models that agree about the presence of a TC in input data. ERA5 reanalysis were used for model training and testing jointly with the International Best Track Archive for Climate Stewardship records. Results showed that the ML approach is well-suited for TC detection providing good generalization capabilities on out of sample data. In particular, it was able to accurately detect lower TC categories than those used for training the models. On top of this, the ensemble approach was able to further improve TC localization performance with respect to single model TC center estimates, demonstrating the good capabilities of the proposed approach.

Francesco Immorlano, Veronika Eyring, Thomas le Monnier de Gouville et al.

Precise and reliable climate projections are required for climate adaptation and mitigation, but Earth system models still exhibit great uncertainties. Several approaches have been developed to reduce the spread of climate projections and feedbacks, yet those methods cannot capture the nonlinear complexity inherent in the climate system. Using a Transfer Learning approach, we show that Machine Learning can be used to optimally leverage and merge the knowledge gained from global temperature maps simulated by Earth system models and observed in the historical period to reduce the spread of global surface air temperature fields projected in the 21st century. We reach an uncertainty reduction of more than 50% with respect to state-of-the-art approaches while giving evidence that our method provides improved regional temperature patterns together with narrower projections uncertainty, urgently required for climate adaptation.

Prakhar Srivastava, Farrin Marouf Sofian, Francesco Immorlano et al.

Despite recent advances in test-time scaling and finetuning of diffusion models, guidance in Auto-Regressive Diffusion Models (ARDMs) remains underexplored. We introduce an amortized framework that augments pretrained ARDMs with a lightweight controller network, trained offline by previewing future ARDM rollouts and learning stepwise controls that anticipate upcoming observations under a terminal cost objective. We evaluate this framework in the context of data assimilation (DA) for chaotic spatiotemporal partial differential equations (PDEs), a setting where existing methods are often computationally prohibitive and prone to forecast drift under sparse observations. Our approach reduces DA inference to a single forward rollout with on-the-fly corrections, avoiding expensive adjoint computations and/or optimizations during inference. We demonstrate that our method consistently outperforms four state-of-the-art baselines in stability, accuracy, and physical fidelity across two canonical PDEs and six observation regimes. We will release code and checkpoints publicly.