Monthly-scale hydro-climatic forecasting and climate change impact evaluation based on a novel DCNN-Transformer network

Climate change has emerged as one of the foremost global challenges confronting humanity today, leading to a heightened frequency and intensity of extreme weather phenomena, including droughts, floods, and erratic rainfall patterns. Accurately predicting changes in runoff patterns under future climate conditions holds significant importance for effective regional water resource planning and management. Recent research on runoff forecast has centered on optimizing hyperparameters of ELM, RNN, LSTM models using PSO, GWO, SSA, and other algorithms. Additionally, key features are extracted through input variable decomposition and preprocessing methods like EMD, EEMD, and VMD. However, these approaches have difficulties in extracting the long-term dependencies information of sequence units, parallel computing, and hyperparameter sensitivity. To address these shortcomings, this study proposes a novel end-to-end deep runoff prediction model based on deep convolutional neural network and Transformer (DCTN). The deep convolutional modules of DCTN employs the deep convolutional operation to extract local features of climate data while the Transformer of DCTN makes full use of self-attention to capture the long-term dependencies, which can achieve more accurate runoff predictions. Experiments on historical runoff forecasting at the Shanjiaodi hydrology station in the Dagu River Basin show that the proposed DCTN obtains a notable improvement of approximately 30.9% compared to traditional models. Based on the prediction results of three shared socioeconomic pathways, the potential impacts of climate change on runoff in Dagu River Basin were evaluated using the DCTN model. The results reveal that the likelihood of spring floods is substantially amplified in the mid-century and late-century, while the probability of extreme summer runoff diminishes. This study advances the understanding of runoff prediction and its implications under changing climate scenarios, paving the way for more informed decision-making and effective water resource management strategies.