Improving Daily Streamflow Forecasting Using Deep Belief Net-Work Based on Flow Regime Recognition

Streamflow forecasting is of great significance for water resources planning and management. In recent years, numerous data-driven models have been widely used streamflow forecasting. However, the traditional single model ignores utilization different regimes. This study proposed an integrated framework daily based on regime recognition flow sequences. The integrates self-organizing maps (SOM) identifying sub-sequences, random forests (RF) algorithm to select input variables a deep belief network (DBN) establishing complex relationships between selected streamflows sub-sequences. Specifically, was applied forecast at Xiantao hydrological station in Hanjiang River Basin, China. results show that developed has higher prediction accuracy than (i.e., DBN this study), with Nash efficiency coefficient (NSE) 0.91/0.81 determination (R2) 0.93/0.89 framework/DBN during validation period, respectively. Additionally, peak flood also improved. relative error derived from reduced by 4.6%, compared model. Overall, constructed integration framework, considering characteristic regimes, could improve