Influence of topographic features and stream network structure on the spatial distribution of hydrological response

Abstract

Topographic data and geomorphological characteristics are recognized as significant factors that affect the hydrological response of a river basin. There have been a variety of lumped or semi-distributed hydrological models developed based on the use of geomorphological parameters. However, the possibility of applying geomorphological factors to describe the temporal and spatial distribution of hydraulic variables in a watershed is still questioned. This study focuses on thoroughly exploring the relationship between watershed topography and two variables, the water depth and discharge. Machine learning (ML) techniques are utilized to predict the spatial distribution of the two hydraulic variables based on geomorphological factors. To enhance the overall predicted accuracy, the structure and unique distribution of the stream network exiting a river basin are investigated for use in modifying the training procedure of the ML model. A novel method of data classification according to the stream network structure and cluster analysis is established to reinforce the capability of predicting hydraulic variables. Application results of two river basins in Taiwan island show that the performance of water depth and discharge predictions are satisfactory as long as the data classification is conducted for model training. Therefore, the ML-based model developed in this study is a promising way to replace the fully distributed numerical model in order to improve executing efficiency and bypass any problem of instability.