<p>Due to the flat terrain, insufficient drainage during heavy rainfall frequently results in waterlogging disasters in floodplains. Developing distributed hydrological models to simulate and predict flood characteristics such as inundation extent, water depth, and duration is essential for mitigating their impacts. Existing distributed hydrological models typically employ unidirectional coupling, focusing solely on the influence of runoff production on runoff confluence. This approach involves first calculating the runoff production (commonly referred to as net rainfall) and then using this result as a prerequisite for runoff confluence calculations. However, in actual hydrological processes, runoff production and confluence occur almost simultaneously, characterized by complex water exchanges between hydrological response units and their adjacent units. Consequently, runoff production and confluence processes are intricately coupled in a bidirectional manner within hydrological models. The aforementioned unidirectional coupling cannot fully and accurately represent the dynamic behaviour of real-world hydrological systems, leading to discrepancies in simulating these processes. To address this issue, this paper presents a raster-based waterlogging simulation model with bidirectional coupling of runoff production and confluence in plain areas. Firstly, the response units for raster are divided based on digital elevation model (DEM), and a method for segmentation and calculation of depression fill volume is established, which regards flood below the depression fill volume as static flood. Then, an adaptive runoff production model is established based on the SCS-CN runoff production method, and two empirical parameters are introduced to correct the errors of DEM and depression fill volume respectively. On this basis, by solving the water balance equation and Manning’s equation simultaneously, the slope confluence model with bidirectional coupling of production and confluence is constructed. Finally, a river confluence model is established by using the Muskingum method. These modules are integrated to form the proposed waterlogging simulation model. The model results of study area show that the peak flow errors were within ± 15%, and the NSEs were over 0.84. It effectively addresses the limitations of unidirectional coupling and improves the accuracy of plain waterlogging simulation, thus providing novel theoretical support for hydrological simulation, as well as a scientific tool for flood warning and water resource management, with broad application prospects and practical value.</p>