.png){kind=icon}
.png){kind=icon}
Return
Reservoir Water Quality Inversion During Dredging Based on Sentinel-2 Imagery
.png){kind=icon}
PRE
AI
Q
Z
L
J
Y
F
DOI：10.1007/s11269-026-04854-2.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Accurate and dynamic water quality monitoring is vital for water resource security and ecological governance. However, traditional monitoring methods suffer from poor timeliness, limited spatial coverage, and high cost, and machine learning models’ inherent “black box” nature limits their credibility. To address these issues, this study takes Linqi Reservoir—a medium-sized reservoir undergoing dredging and expansion with elevated suspended sediment concentrations—as the study area. Based on field-measured water quality data and Sentinel-2 satellite imagery collected during March–April 2025, a comprehensive inversion scheme integrating TPE-optimized XGBoost and SHAP interpretability analysis is proposed for four core parameters: Chemical Oxygen Demand (COD), Total Phosphorus (TP), Total Nitrogen (TN), and Ammonia Nitrogen (NH₃-N). Unlike existing studies that focus on large lakes and low-turbidity waters, this work fills the under-explored gap of water quality inversion in dredging-disturbed, high-turbidity water bodies. Results show that the TPE-XGBoost model achieves test set R² values of 0.78, 0.80, 0.81, and 0.78, representing an improvement of 8.33%–11.11% over the unoptimized XGBoost and outperforming Random Forest and SVM models. SHAP analysis reveals that Band 9 (water vapor band) dominates COD inversion (20.8%), Band 2 (blue band) dominates NH₃-N inversion (15%), and Band 10 (shortwave infrared-cirrus band) and Band 9 serve as the primary influencing bands for TP and TN inversions, respectively. This study provides a technical reference for dynamic water quality monitoring of similar small and medium-sized reservoirs, and future work will incorporate multi-season datasets across diverse water bodies to further enhance the model’s generalizability.
Keywords:
Water quality inversion
Sentinel-2
XGBoost
Tree-structured Parzen Estimator
SHAP interpretability
Journal
IF:
4.7
Papers:
8.1K
Citations:
1.6W
Organization
Cited Papers