An Example of Global Forecasting with FourCastNetv2 Made by a University Research Lab Using GPU

View a PDF file from the paper entitled democracy for Numerical Weather Models of AI: An example of global prediction with Fourcastnetv2 made by a university research laboratory using GPU, by Iman Khadir and 7 other authors

PDF HTML (experimental) view

a summary:This paper shows the feasibility of adding a democratic character to give the democratic character to the global weather forecast models that AI drives between university research groups by taking advantage of the GPU processing units and artificial intelligence models available freely, such as NVIDIA FourCastnetv2. Fourcastnetv2 is an advanced nervous network in NVIDIA to predict the weather and is trained in a group of data 73 channels from the European Medium -range weather forecast (ECMWF) Reanalyss V5 (ERA5) at one levels and different levels of pressure. Although the training specifications of Fourcastnetv2 are not released to the public, the training documents for the first generation of the form, Fourcastnet, are available to all users. Training was 64 A100 graphics processing units and took 16 hours to complete. Although NVIDIA models provide significant discounts in both time and cost compared to traditional predictions of numerical weather (NWP), reproduction of the published prediction results represents continuous challenges for resource -bound university research groups with the availability of GPU. We explain (I) to take advantage of Fourcastnetv2 to create predictions through the applicant programming interface (API) and (2) using NVIDIA devices to train the original Fourcastnet model. Moreover, this paper shows the capabilities and restrictions of NVIDIA A100 for limited research groups in universities. We also explore data management, training efficiency, and verify the form of the form, with highlighting the advantages and challenges of using high -performance computing resources. Consequently, this paper and the corresponding GitHub materials may serve as a preliminary guide for other university research groups and courses related to automated learning, climate science, and data science to develop research and education programs on predicting weather of artificial intelligence, and thus helps in adding a democratic character to AI NWP in the digital economy.