Ensemble prediction

[shoyer]

Thanks for sharing your models!

It would be interesting for the community to be able to run the ensemble version of Pangu weather (i.e., the version from Figure 5 of the Nature paper). Is this something you could share?

Thanks!

[198808xc]

Hi Stephen,

Yes. Given any input data (IFS initial or reanalysis), you can make use of the last section of our pseudocode to add Perlin noise to the input data. With these perturbed inputs, you can get a large number (e.g., 99+1 as in our paper) of forecast results, on which you can make an ensemble forecast.

Best, Lingxi

[HanPPYi]

尊敬的作者：

您好，所提供伪代码的PerlinNoise()部分： perlin_noise = noise_scale*GenerateFractalNoise((H, W), (period_number, period_number), octaves, persistence)

但根据https://github.com/pvigier/perlin-numpy/中的参数设置： generate_fractal_noise_3d(shape, res, octaves=1, persistence=0.5, lacunarity=2,tileable=(False, False, False)） 其中shape, res需要是(tuple of 3 ints)，和伪代码中传入的(H, W), (period_number, period_number),维度不符。

请问在实际运行过程中，GenerateFractalNoise中的几个参数具体如何设置？

感谢您的分享与指导~