Noelia Otero, Pascal Horton

In recent years, the use of deep learning methods has rapidly increased in many research fields. Similarly, they have become a powerful tool within the climate scientific community. Deep learning methods have been successfully applied for different tasks, such as identification of atmospheric patterns, weather extreme classification, or weather forecasting. However, due to the inherent complexity of the atmospheric processes, the ability of deep learning models to simulate natural processes, such as precipitation, is still challenging. Therefore, a thorough evaluation of their performance and robustness in predicting precipitation fields is still needed, especially for extreme precipitation events, which can be devastating in terms of infrastructure damage, economic losses, and even loss of life. In this study, we present a comprehensive evaluation of a set of deep learning architectures to realistically simulate precipitation, including heavy precipitation events (>95th percentile) and extreme events (>99th percentile) over the European domain. Moreover, we examine the optimal number of inputs based on the importance of the predictors derived from a layer-wise relevance propagation procedure. Among the architectures analyzed here, the U-Net network was found to be superior and outperformed the other networks to simulate precipitation events. Moreover, we found that a simplified version of the original U-Net with a single encoder-decoder level achieves similar skill scores as deeper versions for predicting precipitation extremes, significantly reducing overall complexity and computing resources.