Drainage Pattern Recognition Method Using Graph Convolutional Networks Combined With Three‐Dimensional Elevation Features

ABSTRACT Drainage pattern recognition plays a crucial role in flood management, hydraulic engineering site selection, and biodiversity maintenance. Although deep learning methods have been widely employed in drainage pattern recognition in recent years, most research has focused on two‐dimensional f...

Full description

Saved in:
Bibliographic Details
Published in:Transactions in GIS Vol. 29; no. 1
Main Authors: Qiang, Bo, Liu, Tao, Du, Ping, Li, Pengpeng, Wang, Wenning, Xu, Shenglu
Format: Journal Article
Language:English
Published: Oxford Blackwell Publishing Ltd 01.02.2025
Subjects:
Accuracy
Artificial neural networks
Biodiversity
Chebyshev approximation
Deep learning
Drainage
Drainage patterns
Flood control
Flood management
Geographic information systems
graph convolutional network
Hydraulic engineering
Machine learning
Mathematical morphology
Pattern recognition
Polynomials
river
Site selection
Support vector machines
three‐dimensional elevation
ISSN:1361-1682, 1467-9671
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:ABSTRACT Drainage pattern recognition plays a crucial role in flood management, hydraulic engineering site selection, and biodiversity maintenance. Although deep learning methods have been widely employed in drainage pattern recognition in recent years, most research has focused on two‐dimensional features, with limited attention given to the impact of three‐dimensional elevation features. This study, therefore, introduces a graph convolutional network (GCN) approach that incorporates three‐dimensional elevation features. This method integrates three‐dimensional elevation features into the feature system of deep learning‐based drainage pattern recognition for the first time, combining two‐dimensional geometric and topological features to comprehensively capture variations in both vertical and planar morphology of drainage patterns. Specifically, this study extracts representative drainage pattern samples (e.g., dendritic, skeleton, parallel, and fan) from OSM and HydroSHEDS river vector data. Six features are designed based on the micro‐to‐macro processes involved in drainage pattern development, and a novel drainage pattern feature description index is proposed. A GCN model architecture based on the first‐order Chebyshev polynomial is then developed to extract both global and local information of drainage patterns under three‐dimensional elevation features. Experimental results demonstrate that the recognition accuracy of this method achieves 90% on the test dataset, with precision, recall, and F1 scores improving by at least 3% compared to methods such as support vector machines (SVM) and GraphSAGE. This suggests that the feature description index incorporating three‐dimensional elevation features more comprehensively reflects the development process of drainage patterns, thereby improving recognition accuracy and offering new methods and perspectives for drainage pattern research.
Bibliography:Funding
This work was supported by National Natural Science Foundation of China: 42261076, Key Research and Development Project of Lanzhou JiaoTong University: LZJTU‐ZDYF2301, Major Technology Project of Gansu Province: 22ZD6GA010 and Gansu Youth Science and Technology Fund: 24JRRA275.
ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1361-1682
1467-9671
DOI:10.1111/tgis.70002