Multi-factor impacts identification and runoff forecasting based on ensemble model: A case study of the source region of Yellow River

Study region The Source Region of Yellow River (SRYR), China. Study focus To further enhance the performance and reliability of runoff prediction models base on signal processing, this study incorporated Frequency Analysis (FA) techniques into the ensemble model. By combining FA with the advantages of Variational Mode Decomposition (VMD) and Convolutional Long Short-Term Memory (ConvLSTM) network in feature extraction and multi-factor prediction, the study constructed an ensemble model capable of uncovering the spatiotemporal contribution of various factors to the runoff process and combining the joint effects of multiple hydrological factors. New hydrological insights for the region Based on the observational data from local hydrological and meteorological stations, along with reanalysis dataset, the VMD-FA-ConvLSTM model reached the best prediction result with NSE and KGE of 0.92 and R 2 of 0.93, displaying better performance than many traditional models. This affirms that the proposed ensemble model yields higher precision and greater applicability with hydrological signals. Additionally, the study forecasts the future runoff trends in the study area for the next 20 years based on four Shared Socioeconomic Pathway (SSP) scenarios. The results indicate that the average runoff in SRYR from 2021 to 2040 decreases by about 1.55% under the SSP1–2.6 scenario compared with that from 2001 to 2020, whereas the forecasted runoff increases by about 5.22%, 10.77%, and 6.42% under the SSP2–4.5, SSP3–7.0, and SSP5–8.5 scenarios, respectively.