Leader
Italo Epicoco

The objectives of this research unit regard the analysis and optimization on “multi-cores” hybrid architectures of the main computational kernels featured in the models used at CMCC, by considering the performance impacts of optimized compilers, numerical libraries and new parallel paradigms. In addition, the activities also focus on the study of the impact of exascale computing architectures on the numerical algorithms used in the main climate models studied at CMCC. In particular, the following aspects are analyzed in exascale terms: (i) the use of advanced parallel algorithmic structures to reduce the current “dynamical cores” communication overhead; (ii) the optimal management of “multi-model” and “multi-emission” ensemble experiments; (iii) the optimal management of the memory system and its hierarchies; (iv) the rationalization of I/O operations; (v) the adoption of new communication parallel paradigms and parallel tasks synchronization mechanisms.

Leader
Donatello Elia

The main goal of this Research Unit concerns the design and implementation of Data Science open source solutions addressing efficient access, analysis and mining of scientific data in the climate change domain. In particular, the activities focus on (i) the management of scientific data in major international contexts/initiatives like the ENES Climate Data Infrastructure, the Earth System Grid Federation, and the European Open Science Cloud, (ii) the definition of new storage models to enable efficient access to climate data (including parallel I/O approaches), and (iii) the development of advanced Data Science environments for climate scientists leveraging High Performance Data Analytics solutions as well as machine/deep learning frameworks to accelerate scientific discovery, (iv) the development of fault tolerant workflow automation tools and applications for the optimized scheduling of large number of tasks on HPC infrastructures designed to support weather and climate use cases.

Leader
Gabriele Accarino

The objective of the EMLC2 Research Unit concerns the exploration and development of cutting-edge Machine/Deep Learning techniques and related applications in the context of climate science. The activities of this research unit are strictly connected to the availability of both huge amounts of data produced by the simulations and the increasing computational power of the forthcoming Exascale architectures. Specifically, the activities of this research unit focus on (i) the investigation of hybrid approaches, where critical computing Kernels of climate models are replaced with Neural Network algorithms without affecting results accuracy, (ii) the application of Machine/Deep Learning techniques for downscaling activities, by also exploring and comparing the application of different Neural Networks approaches (e.g. CNN or GAN), (iii) the analysis of the predictive capabilities of the Neural Networks (e.g. LSTM) for time series predictions in different weather and climate use cases, (iv) exploration of the application of ML/DL techniques in various scientific use cases: Tropical Cyclones, Conflicts Prediction, Monitoring & Processing of agroforestry parameters, study of Melting Glaciers and Analysis of Mars radargrams.