Skip to main content
. 2023 Jul 5;619(7970):526–532. doi: 10.1038/s41586-023-06184-4

Extended Data Fig. 1. Architecture details of NowcastNet.

Extended Data Fig. 1

a, Evolution network. b, Generative network. c, Temporal discriminator. d, Noise projector. e, Basic blocks. The input fields are of height H and width W. The convolutional layer uses (N,N)-kernel. Leaky ReLU is the leaky rectifier linear unit with negative slope of 0.2. BN is the batch normalization. Up and Down are bilinear interpolations to expand or reduce spatial size. Avg Pool is the spatial average pooling. Spatial Norm and Instance Norm are the normalizations applied within the spatially adaptive normalization to implement the physics-conditioning mechanism between the generative network and the evolution network.

HHS Vulnerability Disclosure