Advanced classification of optical water types and ensemble learning models for Chl-a inversion in Dongting and Poyang lakes using Sentinel-2 remote sensing: assessing the impact of extreme drought events
Accurate and robust retrieval of chlorophyll-a (Chl-a) concentrations in optically complex lakes is critical for effective water quality monitoring, especially under frequent extreme climatic conditions. This study proposes an integrated framework combining optical water type (OWTs) classification a...
Salvato in:
| Autori principali: | , , , , , , , , , |
|---|---|
| Natura: | Articolo |
| Lingua: | inglese |
| Pubblicazione: |
Elsevier
2025-08-01
|
| Serie: | Ecological Indicators |
| Soggetti: | |
| Accesso online: | http://www.sciencedirect.com/science/article/pii/S1470160X25006685 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| Riassunto: | Accurate and robust retrieval of chlorophyll-a (Chl-a) concentrations in optically complex lakes is critical for effective water quality monitoring, especially under frequent extreme climatic conditions. This study proposes an integrated framework combining optical water type (OWTs) classification and ensemble learning algorithms to enhance the re- mote sensing retrieval accuracy of Chl-a in Dongting Lake and Poyang Lake during 2020–2023, particularly during the extreme drought event of 2022. Firstly, K-means clustering combined with spectral feature analysis was used to classify both lakes into four distinct OWTs. Dongting Lake classifications effectively separated phytoplankton-dominated waters (OWT1, OWT2, OWT4) from those dominated by non-algal suspended matter (OWT3). For Poyang Lake, classifications identified mixed water types (OWT1), frequent algal bloom areas (OWT2), high biomass phytoplankton areas (OWT3), and areas significantly masked by non-algal particles (OWT4). Further analysis indicated that OWT2 in Dongting Lake is prone to algal blooms, while OWT3 and OWT4 are dominated by non-algal particles. Secondly, the performance of ensemble learning methods (Bagging, Boosting, Stacking, and Voting) was evaluated against traditional single models (SVR and BPNN). The results demonstrated the superior stability and predictive accuracy of the Voting strategy under low Chl-a conditions in Dongting Lake, achieving a maximum MAPE reduction of 84.76 %. Meanwhile, the Stacking method exhibited outstanding robustness in Poyang Lake’s complex optical conditions, with RMSE reduced by up to 93.12 %. Additionally, hierarchical modeling based on OWT-specific ensemble methods significantly reduced errors compared to traditional global models, with decreases of 36 % in Dongting Lake and 42.19 % in Poyang Lake. Finally, this study revealed that the extreme drought event in the Yangtze River basin in 2022 significantly altered seasonal and interannual variations in the OWTs of Dongting Lake and Poyang Lake. The drought led to lake shrinkage and increased dominance of non-algal suspended particles, highlighting the vulnerability of these connected lake ecosystems under extreme climatic conditions. The research emphasizes the importance of combining a lake-specific OWTs classification and advanced ensemble learning techniques for precise water quality assessment under dynamic environmental conditions. |
|---|---|
| ISSN: | 1470-160X |