Statistical downscaling of precipitation with deep neural networks

Accurate weather predictions are essential for many aspects of social society. Providing a reliable high-resolution precipitation field is essential to capture the finer scale of heavy precipitation events,  which is normally poorly represented in the numerical models. Statistical downscaling is an appealing tool since it is computationally inexpensive. Thus, it has been widely used over the last three decades. In recent years, super-resolution with deep learning has been successfully applied to generate high-resolution from low-resolution images in the computer vision domain. This task is somewhat analogous to downscaling in the meteorological domain.

Inspired by this, we explore the use of deep neural networks with a super-resolution approach for statistical precipitation downscaling. We apply the Swin transformer architecture (SwinIR) as well as convolutional neural network (U-Net) with a Generative Adversarial Network (GAN) and a diffusion component for probabilistic downscaling. We use short-range forecasts from the Integrated Forecast System (IFS) on a regular spherical grid with Δx_IFS=0.1° and map to the high-resolution observation radar data RADKLIM (Δx_RK=0.01°). The neural networks are fed with nine static and dynamic predictors, similar to the study by Harris et al., 2022. All the models are comprehensively evaluated by grid point-level errors as well as error metrics for spatial variability and the generated probability distribution. Our results demonstrate that the Swin Transformer model can improve accuracy with lower computation cost compared to the U-Net architecture.  The GAN and diffusion models both further help the model to capture the strong spatial variability from the observed data.   Our results encourage further development of DNNs that can be potentially leveraged to downscale other challenging Earth system data, such as cloud cover or wind.