Retrieval of non-optical active water quality parameters in complex Lake environments using a novel zoning-based ensemble modeling strategy
Remote sensing technology has provided significant support for the spatial and quantitative limitations of traditional water quality monitoring methods. However, accurate retrieval of non-optically active water quality parameters (NAWQPs) remains challenging due to their weak spectral responses and...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-07-01
|
| Series: | Ecological Indicators |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1470160X25006533 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Remote sensing technology has provided significant support for the spatial and quantitative limitations of traditional water quality monitoring methods. However, accurate retrieval of non-optically active water quality parameters (NAWQPs) remains challenging due to their weak spectral responses and interference from diverse aquatic vegetation. In this study, we proposed a novel zoning-based ensemble modeling strategy (ZBEMS) by integrating aquatic vegetation classification with hyperspectral features derived from ZY-1 02D images, and tested it in Nansi Lake to retrieve NAWQPs (ammonia nitrogen (NH3-N), chemical oxygen demand (COD), and dissolved oxygen (DO)). Firstly, diverse aquatic vegetation was identified using aquatic vegetation index (AVI), floating algae index (FAI), and normalized difference vegetation index (NDVI), and the dominant type within each 3 × 3 Km grid determined vegetation zones (floating emergent vegetation (FEVA), submerged aquatic vegetation (SAV), and algae bloom (AB)). Secondly, multi-spectral scale morphological combination features (MSMCF) were extracted from the ZY-1 02D images. Finally, the ZBEMS integrating four machine learning models (Random Forest Regression (RFR), Gradient Boosting Regressor (GBR), eXtreme Gradient Boosting (XGBoost), and Support Vector Regression (SVR)) was applied across different zones for NAWQPs retrieval. Within the sampling area, the model achieved R2 values of 0.56, 0.54, and 0.57 and root mean square errors (RMSE) of 0.04 mg/L, 4.56 mg/L, and 1.87 mg/L for retrieval of NH3-N, COD, and DO, respectively. Compared with traditional ensemble learning models, ZBEMS model improved the R2 by approximately 0.13 for three parameters. These results indicate that the ZBEMS offers a promising approach for NAWQPs retrieval in complex aquatic environments. |
|---|---|
| ISSN: | 1470-160X |