Data-driven models are one of the effective tools for medium and long term runoff forecasting. This paper classifies the main development directions of medium and long term runoff forecasting methods based on data-driven models from two approaches: runoff time series analysis and runoff causality analysis. It summarizes the research progress of key technologies and looks ahead to the future development of data-driven models in medium and long term runoff forecasting. The results indicate that: (1) the main research directions of runoff time series analysis can be divided into runoff time series decomposition methods, forecast model combination methods, and model hyperparameter optimization algorithms; (2) the main research directions of runoff causality analysis can be divided into forecast factor selection methods, forecast model combination methods, and forecast result correction methods; (3) although data-driven models have shown promising computational performance and development potential in medium and long term runoff forecasting, they still face challenges such as low interpretability, poor generalization ability, and the lack of a universal business platform. Future research could focus on embedding physical mechanisms, introducing interpretable machine learning algorithms, applying transfer learning methods, and building a universal business platform with multiple operational functionalities.