Grant: Discovery Grants Program - Individual

Scientific Machine Learning is concerned with using methods from machine learning to tackle problems that have traditionally been investigated using classical scientific computing. Until very recently, science, and in particular scientific computing, has followed the classical formula From rules to data, meaning one first defines a mathematical theory or a computational algorithm which generates predictions (data), that is then compared to some benchmarks, such as real-world observations. Machine learning reverses the direction to From data to rules. Data is being analysed and rules are being derived from it. In other words, machine learning is devoted to learning from data. Given the ever-increasing amount of data that has been generated over the past 50 years or so in virtually all areas of the mathematical sciences, including astronomy, biology, chemistry, geophysics, meteorology and oceanography, the increased interest in applying machine learning to all these areas is quite natural. This project is devoted to the advancement of the field of scientific machine learning as applied to the atmospheric sciences. Traditionally, weather and climate models are built around numerical approaches for solving the governing equations of the atmosphere. With purely data-driven approaches emerging, this project will be devoted to studying the advantages and disadvantages of this new paradigm in comparison to the classical approaches used so far.