Monthly precipitation prediction at regional scale using deep convolutional neural networks

Abstract

Variations in monthly precipitation are associated with climate extremes having significant socio‐economic and eco‐environmental impacts. Knowledge of monthly precipitation information is therefore valuable for policy making. Extensive research has been conducted on dynamic prediction using state‐of‐the‐art coupled climate models. However, the skilful prediction of monthly precipitation with dynamical models remains a challenge. With the development of machine learning tools, statistical predictions show comparable performance with dynamic models, but they are limited to at‐site monthly prediction, due to the lack of ability for processing spatially connected geophysical data. To improve monthly precipitation forecasting and provide regional forecasts, we propose a model termed UNet‐RegPre based on convolutional neural network and U‐net architecture. The model shows comparable prediction skills with recurrent state‐of‐the‐art dynamic forecasts (CFSv2), and has the capability to capture spatiotemporal patterns of precipitation and reproduce the main process of representative droughts. The key precursors for rainfall development identified by UNet‐RegPre shows that the constructed model can detect the climatic connection between rainfall in eastern China and summer monsoon, showing the potential to advance hydrometeorological understanding with a deep learning‐based model. These results suggest that the proposed model can be a potential tool for precipitation prediction.

Deep learning based sub-seasonal precipitation and streamflow forecasting over the source region of the Yangtze River

“…Despite general improvements of forecasts, they tend to smooth the extreme precipitation at sub-seasonal scales (Baño-Medina et al, 2021;Kim et al, 2022), likely due to insufficient heavy precipitation samples (Chen et al, 2022). Many studies have since introduced more recent variants of CNNs including U-Net (Ni et al, 2023) and SmaAt-UNet (Li et al, 2024), or coupled standard CNNs with different structures, such as Auto-Encoder (Ling et al., 2022), Transformer (Ling et al, 2024), and in particular ResNet, which shows the potential of mitigating the vanishing gradient issue by introducing the residual paths (Nie et al, 2024). Others have attempted to introduce specialized loss functions to balance heavy and light rains, such as the exponentially weighted mean squared error (Ebert-Uphoff et al, 2020) and Dice loss (You et al, 2022).…”