MACHINE LEARNING-DRIVEN OPTIMIZATION OF NUMERICAL WEATHER PREDICTION MODELS

In order to anticipate future weather, weather forecasting mostly employs numerical weather prediction models that incorporate weather observation data, such as temperature and humidity. For weather forecasting, the Korea Meteorological Administration (KMA) has embraced the UK's GloSea6 numerical weather prediction model. Supercomputers are necessary to run these models for research reasons in addition to using them for real-time weather predictions. However, several researchers have encountered challenges while attempting to run the models because to the restricted supercomputer resources. Low GloSea6, a low-resolution model created by the KMA to solve this problem, can be operated on small and medium-sized servers at research institutes; nonetheless, it still consumes a lot of computer resources, particularly in the I/O load. Model I/O optimisation is crucial because I/O load may degrade performance for models with heavy data I/O, yet user trial-and-error optimisation is ineffective. In order to optimise the Low GloSea6 research environment's hardware and software characteristics, this study offers a machine learning-based method. There were two phases in the suggested approach. In order to determine hardware platform parameters and Low GloSea6 internal parameters under different settings, performance data was first gathered using profiling tools. Second, to identify the ideal hardware platform characteristics and Low GloSea6 internal parameters for novel research contexts, a