Final Report Summary - PANTARHEI (Interdisciplinary Integrated Forecasting System for Fluid Flows)
The European Centre for Medium-Range Weather Forecasts (ECMWF) hosted the European Research Council funded project PantaRhei, which explored ways to augment highly optimised numerical weather prediction (NWP) models with novel capabilities, both technical and scientific, and in so doing alleviate intrinsic limits imposed by the existing NWP framework. This aspect is of fundamental importance in view of the flexibility required to adapt to emerging energy-efficient, heterogeneous hardware for high performance computing (HPC). The PantaRhei project pursued an entirely flexible, hierarchical and evolutionary approach capable of satisfying the needs of current and future NWP. The idea is to mitigate the limitations of the existing hydrostatic simulation model by providing an influx of information from complementary dynamical modules, with individual capabilities of cloud-resolving models, concurrently driven by an influx of large-scale hydrostatic information. The first step in the pursuit of this paradigm is the development of an all-scale finite-volume module (FVM) supplementing ECMWF’s Integrated Forecasting System (IFS). The FVM on its own can provide solutions representative of elements of real weather over a large range of scales using computation and communication patterns that are entirely different from current state-of-the-art global spectral transform-based NWP models. The FVM provides relevant algorithmic building blocks (weather and climate dwarfs) for co-design of hardware technology for extreme scale applications, where global communication must be minimised. Moreover, the proposed concept offers exciting new opportunities for forming synergies between operational NWP and cloud-resolving research while advancing hierarchical modelling concepts and multi-level algorithms. IFS augmented by FVM provides the required advanced environment to explore alternative discretisation techniques, applied in the context of time-critical operational weather forecasting but relevant well beyond, while providing strategic answers towards the big data and energy usage challenges of current and future HPC applications.