The Application of Artificial Intelligence in Streamflow Forecasting: A Review

Accurate streamflow modelling and forecasting are crucial for proper water resources planning, engineering, and management. Since the last two decades, Artificial Intelligence (AI) methods have been increasingly developed and used in streamflow forecasting due to their robustness in dealing with non-linear, uncertain, and complex hydrological datasets. The current paper is a retrospective study on the different AI methods used in streamflow forecasting. Based on the review of research papers published in different journals from 1999 to 2023, this paper presents an analysis of the various AI techniques and algorithms in use. It identified feed-forward neural network (FFNN) as one of the most prominently used AI methods in streamflow forecasting. Also, the review examined papers from the perspective of forecast window time frame used, only to find that majority of the research considered a short forecast window of 1-7 day(s) lead time as compared to longer time frames. Additionally, the paper also provides information on the different input parameters used by researchers over time in defining their AI models. The analysis shows that primarily used parameters include streamflow, rainfall, and temperature data, and negligible use of snow or other cyclic factors as inputs. Based on all these statistical information, this review paper contributes to the possibility of exploring other inputs parameters and extending the existing forecasting window time frame for better forecasting of streamflow, and hence, efficient management of water resources. It has also forwarded directions for future research in streamflow modelling, including the exploration of time series neural network models for forecasting streamflow over larger windows. This review thus, identifies the challenges associated with streamflow prediction and the possible ways to overcome them.