Fig. 3: Example of Aardvark’s gridded forecasts for the U10 wind component. | Nature

Fig. 3: Example of Aardvark’s gridded forecasts for the U10 wind component.

From: End-to-end data-driven weather prediction

Fig. 3

al, Plots of the initial condition (ac) and subsequent forecasts (dl) for U10, showing Aardvark’s prediction (a,d,g,j), the ERA5 ground truth34 (b,e,h,k) and the difference between the two (c,f,i,l). Lead time t = 0 corresponds to 00:00 on 11 January 2018. Aardvark correctly predicted the large-scale features for this variable and correctly predicted the formation and positioning of the tropical cyclone Berguitta (highlighted in the magent a boxes), which reached peak intensity on 15 January 2018 off the coast of Madagascar. We emphasize that the model made these predictions entirely from raw observations40,41,42,43,44,45,46,47,48, without any NWP products as input.

Back to article page