Abstract: Deep learning techniques have a promising application in rainfall-runoff simulations, but they are limited by the availability of training samples and need coupling with a traditional hydrological model that can provide training data. Selection of coupled data and hyperparameters has a significant impact on the simulation performance of a coupled model, but it lacks deep study. In this paper, we present a rainfall-runoff simulation model by coupling different module data of the Xin’anjiang model with a bidirectional long short-term memory network and optimizing the hyperparameters using the Grey Wolf optimization algorithm, along with an application to the Dongwan watershed. The results show the model improves the simulations of daily runoffs and flood events when coupled with different data, especially runoff data and simulated flow data. The hyperparameter scheme needs to be adjusted to different coupled data, and the Grey Wolf optimization algorithm can meet such a demand. This study provides new ideas and methods for enhancing the runoff simulation capability of the coupled models.

Key words: bidirectional long and short-term memory network model, Xin’anjiang model, coupling model, grey wolf optimization algorithm, runoff simulation