Creating synthetic night-time visible-light meteorological satellite images using the GAN method

Meteorology satellite visible-light images are critical for meteorologists. However, there no channels data at night, so we propose a method based on deep learning to create synthetic during night. Specifically, produce realistic-looking products, trained generative adversarial network (GAN) model. The model can generate from corresponding infrared (IR) and numerical weather prediction (NWP) products. Considering explicitly evaluating the contributions of different IR NWP products elements, suggest using channel-wise attention mechanic, e.g. ‘Squeeze Extraction Block’ (SEBlock) quantitatively weigh importance input channels. experiments meteorology show that proposed is effective realistic