With our understanding of weather phenomena and their interaction with sea currents, pollutants etc, we have been able to create very complex and elaborate simulator models for numerical weather predictions (NWP). These models rely on specialised “classical” solvers, which are handcrafted to simulate a particular physical process. While accurate and reliable, these solvers produce deterministic solutions, can be quite slow and can only be simulated on a rather restrictive coarse grid for global or regional simulations. Machine learning and computational statistics, broadly data science, has been fused with these classical simulations to assimilate observed data (data-assimilation), to produce probabilistic simulations (stochastic parameterisation or ensemble prediction), to fill the gap between these classical simulations to km-scale weather predictions (statistical downscaling).
Data driven approaches have also been used to create neural PDE solvers for weather forecasting which are competitive with, and in some cases exceed the performance of, traditional NWP models but at a fraction of the computational cost.
The first edition of this workshop on “Fusing simulation with data science”, jointly organised by Dept. of statistics in University of Warwick and Met Office, aims to provide an up-to-date snapshot of this fusion between the paradigm of classical simulations and data science and to facilitate discussion among data scientists (probabilist, applied mathematicians, statisticians and machine learners) and meteorologists about the current opportunities and challenges.