ABSTRACT
Machine learning methods are increasingly used in the automatic generalization of river networks, but previous research lacks a comparative analysis of different methods using the same data set. This innovative study considers eight river network indicators, such as river length, river grade, river spacing, seasonality, connectivity, catchment area, tributaries at the next grade, and total number of tributaries, which can precisely describe the characteristics of the river network. The experiments were carried out and automated selection of river network was established based on back-propagation neural network (BPNN), support vector machine (SVM) and decision tree (DT) methods. We established that BPNN and SVM have high selection accuracy, but the parameters are complex. SVM is more suitable for small samples. In addition, DT has unique advantages due to its visualized tree structure and the characteristic of derivable rules. We hope that this study will provide a reference for the selection of river generalization methods in the future.
Notes on the contributor
Chaode Yan is a professor and doctoral supervisor at the School of Water Conservancy Science and Engineering, Zhengzhou University. He received his PhD degree from China University of Mining and Technology (Beijing) in 2007. His main research interests include cartography, geospatial analysis, watershed calculation and analysis.
Authors' contributions
Conceptualization: C. Yan, X. Liu; methodology, C. Yan, X. Liu; software: C. Yan, X. Liu; validation: C. Yan, X. Liu; formal analysis: C. Yan, X. Liu; investigation: C. Yan, X. Liu; resources: C. Yan, X. Liu; data curation: C. Yan, X. Liu; writing-original draft preparation: C. Yan, X. Liu, MW. Boota and Z. Pan; writing-review and editing: C. Yan, X. Liu, MW. Boota and Z. Pan; visualization: X. Liu; supervision: C. Yan; project administration: C. Yan; funding acquisition: C. Yan.
Disclosure statement
No potential conflict of interest was reported by the author(s).