Technology has greatly promoted ecohydrological model development, but runoff generation and confluence simulations have fallen behind in ecohydrological model development due to limited innovations. To fully understand ecohydrological processes and accurately describe the coupling between ecological and hydrological processes, a distributed ecohydrological model was constructed by integrating multisource information into distributed dynamic process model (DDPM). We mainly describe runoff generation and convergence modules. Based on the improved HBV model and degree-3 h factor method, runoff generation and snow routines were constructed for semiarid grassland basins. In view of meandering and variable steppe river channels and steep hydrological relief characteristics, a confluence module was constructed; the 1-km bend radius equivalent concept was innovatively proposed to unify river channel bend degrees. The daily runoff simulation validation results obtained using two datasets were R2 = 0.947 and 0.932, NSE = 0.945 and 0.905, KGE = 0.971 and 0.680. In the 3-hour flood simulations, the DDPM could better restore small long-distance water flows than the confluence method that did not consider actual river lengths or bend energy losses; the DDPM more accurately simulated the flood peak arrival time than the confluence method that did not consider overflow. The simulated mainstream overflow frequency increased by 0.84/10 years, and significant interaction periods of 10 to 13 years occurred with local precipitation, ecological status, and global climate change. An approximately 2-year lag occurred in the global climate change response. This study helps us further understand and reveal the ecohydrological processes of steppe rivers in semiarid regions.