Machine learning classification and driver analysis of diel variability in dissolved oxygen in Taihu Lake
Dissolved oxygen (DO) plays a crucial role in aquatic ecosystems, yet its diel variations are influenced by complex environmental interactions. This study analyzed high-frequency DO data from Sanshandao Island in Taihu Lake (2020–2022) to classify diel DO variation patterns and identify key drivers....
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-06-01
|
| Series: | Ecological Indicators |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1470160X25005230 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Dissolved oxygen (DO) plays a crucial role in aquatic ecosystems, yet its diel variations are influenced by complex environmental interactions. This study analyzed high-frequency DO data from Sanshandao Island in Taihu Lake (2020–2022) to classify diel DO variation patterns and identify key drivers. Using K-means clustering, we identified three distinct types: Type I (warm, humid, rainy, moderate DO fluctuations, late DO peaks, influenced by photosynthesis and precipitation), Type II (warm, dry, high radiation, largest diel DO amplitude, early peaks, photosynthesis-dominated), and Type III (cold-season conditions, high DO levels, minimal diel fluctuations, temperature-driven). Photosynthetically active radiation (PAR) and precipitation were major regulators of diel DO dynamics. PAR strongly influenced DO variations in Type II, while precipitation played a key role in distinguishing Type I from Type II by affecting vertical mixing. To enhance interpretability and predictive accuracy, XGBoost regression models were trained separately for each type, with SHAP analysis quantifying the contributions of individual drivers. The classification-based modeling approach improved performance significantly (R2 increased from 0.73 to > 0.8 in Type I and III). This study presents an integrated framework combining unsupervised clustering and interpretable machine learning to uncover the mechanisms of diel DO variation. The results underscore the need to account for DO pattern heterogeneity in prediction and management and offer new tools for developing targeted water quality strategies in eutrophic lake systems. |
|---|---|
| ISSN: | 1470-160X |