Abstract
Society has benefited enormously from the continuous advancement in numerical weather prediction that has occurred over many decades driven by a combination of outstanding scientific, computational and technological breakthroughs. Here, we demonstrate that data-driven methods are now positioned to contribute to the next wave of major advances in atmospheric science. We show that data-driven models can predict important meteorological quantities of interest to society such as global high resolution precipitation fields (0.25°) and can deliver accurate forecasts of the future state of the atmosphere without prior knowledge of the laws of physics and chemistry. We also show how these data-driven methods can be scaled to run on supercomputers with up to 1024 modern graphics processing units and beyond resulting in rapid training of data-driven models, thus supporting a cycle of rapid research and innovation. Taken together, these two results illustrate the significant potential of data-driven methods to advance atmospheric science and operational weather forecasting.
Highlights
Not all significant physical processes, those operating at small scales, can be explicitly resolved by weather and climate models and instead are parameterised (Bauer, Thorpe, and Brunet 2015)
(2) As a first step, we demonstrate that a deep neural networks (DNN) is capable of learning the complex physical relationship between geopotential height and precipitation at high spatial and temporal resolutions comparable to modern global weather forecasting systems
(3) We investigate scaling of the DNN model by training the model using thousands of graphics processing units (GPU) devices, a necessary requirement when working with terabytes of training data
Summary
Not all significant physical processes, those operating at small scales, can be explicitly resolved by weather and climate models and instead are parameterised (Bauer, Thorpe, and Brunet 2015). Deep learning constructs a representation of input data through a series of connected non-linear layers These networks are able to produce the desired output by optimising a set of adjustable parameters, or weights, that minimise a desired loss function. To avoid over-fitting, practitioners commonly match the increasing capacity of models with larger datasets, providing a regularisation effect that leads to better generalisation This requirement for large quantities of data leads to the characterisation of many machine learning approaches as big-data driven methods. The combination of CNNs and with an LSTM allows us to model complex multi-dimensional time series data and is likely to enjoy widespread application to scientific problems, including weather forecasting and climate modelling (Shi, Chen, Wang, Yeung, Wong, and Woo 2015)
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callMore From: The International Journal of High Performance Computing Applications
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
