ORIGINAL RESEARCH
Water Quality Inversion for Tidal Surge on
Qiantang River Based on UAV Multispectral
Remote Sensing and Machine Learning
More details
Hide details
1
Zhejiang Tongji Vocational College of Science and Technology, Hangzhou, 311231, China
2
Department of Environmental Engineering, China Jiliang University, Hangzhou, Zhejiang, 310018, China
3
Zhejiang Institute of Hydraulics & Estuary, Hangzhou, Zhejiang, 310020, China
Submission date: 2025-01-23
Final revision date: 2025-03-28
Acceptance date: 2025-04-19
Online publication date: 2025-06-05
Corresponding author
Tao Ding
Department of Environmental Engineering, China Jiliang University, No. 258, Xueyuan Street, 310018, Hangzhou, China
KEYWORDS
TOPICS
ABSTRACT
By leveraging drone remote sensing technology, this study addresses the limitations of traditional
monitoring methods in expansive water bodies, such as restricted sampling points, data acquisition
challenges, and insufficient spatiotemporal resolution. It overcomes challenges, including synchronous
data acquisition in dynamic tidal bore environments and nonlinear relationship modeling. Focusing on
Qibao and Yanguan stations along the Qiantang River, the research integrates in-situ water quality data
and multispectral remote sensing images during tidal bores. Machine learning models—Support Vector
Machine Regression (SVR), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost)—were
developed to invert suspended sediment concentration (SSC) and turbidity. Bayesian Optimization was
applied to enhance model performance. Results demonstrate that the XGBoost model optimized by the
Bayesian algorithm achieved superior accuracy, with determination coefficients (R²) of 0.89 (SSC) and
0.93 (turbidity), and reduced root mean square errors to 310.54 mg/L and 33.36 NTU, confirming model
stability and predictive capability. Inversion results revealed abrupt SSC and turbidity surges near bridge
piers during flood tides (peaking at 6000 mg/L and 820 NTU), indicating intense bed scouring by tidal
bore dynamics. The model effectively captures spatiotemporal patterns of water quality parameters and
provides an efficient solution for dynamic tidal bore monitoring, highlighting the potential of integrating
multispectral imagery with machine learning. This approach offers a novel framework for highfrequency
water quality assessment in turbulent hydrodynamic environments.