A spatial weather generator based on conditional deep convolution generative adversarial nets (cDCGAN).

High-resolution weather data is crucial for assessing future climate change impacts on local environments, yet downscaling low-resolution Global Climate Models (GCMs) outputs and addressing associated uncertainty remain significant challenges. In this study, we propose a novel spatial weather generator using generative networks, specifically a numerical conditional deep convolutional generative adversarial network (cDCGAN), as a promising solution. The cDCGAN generates high-resolution weather data from low-resolution GCM outputs and was applied to four case areas in China under four Shared Socio-economic Pathway (SSP) scenarios. The results demonstrate the cDCGAN's accuracy, consistency, and stability, with low uncertainties. The model performs optimally in low-elevation plains and tropical regions. The cDCGAN offers advantages in uncertainty analysis over traditional downscaling methods, serving as a valuable tool for climate change analysis, response estimation, and environmental management decision-making within the spatial statistics domain. [ABSTRACT FROM AUTHOR]