Scale Effects of the Monthly Streamflow Prediction Using a State-of-the-art Deep Learning Model

The accurate prediction of monthly streamflow is important in sustainable water resources planning and management. There is a growing interest in the development of deep learning models for monthly streamflow prediction with the advances in computer sciences. This study aims at investigating the spatial and temporal scale effects on predictive performance when using the deep learning model for monthly streamflow prediction. To achieve this goal, a hybrid deep learning prediction model combining Convolutional Neural Network and Gated Recurrent Unit (i.e., CNN-GRU) was first proposed and applied to many watersheds with varying hydroclimatic characteristics around globe. The Nash–Sutcliffe efficiency coefficient (NSE) and mean relative error (MRE) are used as criteria to evaluate the predictive performance. The results show that the deep learning model is more suitable for monthly streamflow predictions on watersheds with large drainage areas. The drainage area of 3,000 km2 can be considered as a threshold for the predictive performance. The median NSE increases from 0.31 to 0.40, while the median MRE decreases from 53.2% to 46.2% for watersheds with areas larger than 3,000 km2 compared with those with areas smaller than 3,000 km2. In addition, the predictive performance tends to get better with the extension of a training period for the model. When the length of the training period increases stepwise from 10 to 50 years, there is a large increase in NSE (from 0.28 to 0.40) and a moderate decrease in MRE (from 50.3% to 46.2%) for watersheds with areas larger than 3,000 km2. Similar changes can also be found for watersheds smaller than 3,000 km2. The 25- to 35-year training period is the minimum length to obtain a stable predictive performance for most watersheds.