Oct 01, 2021 09:00:00

'Fast Eddy', a system that realizes high-resolution and high-speed weather forecasts using GPU, will be developed

AI development company

DeepMind has developed AI that can forecast the weather as accurately as a weather forecaster, and the technology related to weather forecasting is advancing day by day. Meanwhile, the National Center for Atmospheric Research (NCAR) has developed a system called 'FastEddy ' that can forecast the weather in areas that are more subdivided than conventional weather forecasts using the GPU.

GPUs open the potential to forecast urban weather for drones and air taxis | NCAR & UCAR News
https://news.ucar.edu/132811/gpus-open-potential-forecast-urban-weather-drones-and-air-taxis

In recent years, the development of drone delivery technology has progressed, and medical supplies delivery drones are

playing an active role in regions such as Africa. According to the NCAR, these drones are strongly affected by the wind, so in order for the drone to play an active role, it is necessary to make high-resolution predictions of the weather such as wind that changes depending on the building and terrain.

In the conventional weather prediction technology, even with a relatively high resolution method,

numerical weather prediction is performed using squares of about 3 to 4 km square, so the location of buildings and fine undulations of land are not taken into consideration. The resolution was insufficient for weather prediction for drones. NCAR has also put into practical use a weather usage model called 'WRF-LES ' that enables extremely high-resolution weather forecasts, but for the use of WRF-LES, large-scale computer equipment and forecasts are calculated. It takes a long time. Therefore, NCAR worked on the development of a model that realizes higher resolution and faster forecasts than the conventional weather forecast model.

As a result, NCAR has succeeded in developing a weather forecast model 'Fast Eddy' that uses squares of 5 meters square. In the conventional weather forecast model, the CPU is used for the calculation process, but FastEddy uses the GPU used for the calculation process such as games to execute the calculation. Therefore, FastEddy can predict the weather 8 times faster with the same power consumption as the weather forecast model using the CPU.

NACR says, 'Depending on the wind flow in the city, the drone's battery consumption speed may be tripled.' It helps you decide if you can get to your destination. '

In addition, NCAR talks about the future prospects that he wants to continue improving the functions of FastEddy so that he can simulate how the sound spreads in the future.