基于集成学习方法的鄱阳湖水质反演模型研究OA

To address the limited accuracy of water quality remote sensing inversion using traditional statistical models and sin-gle machine learning algorithms,this study proposes an ensemble learning optimization method based on multi-source remote sensing imagery,aiming to improve the inversion accuracy of water quality parameters,specifically water temperature,dissolved ox-ygen(DO),and pH.Taking Poyang Lake as the study area,multiple linear regression and three individual machine learning mod-els,K-Nearest Neighbors(KNN),Support Vector Regression(SVR),and Multilayer Perceptron(MLP),were established as baseline models.In addition,three ensemble learning strategies were further constructed:Bagging(including Random Forest,RF),Boosting(including AdaBoost,GBDT,and XGBoost),and Stacking.These approaches were applied to invert the above-mentioned water quality parameters,and feature engineering was conducted to optimize spectral band combinations.The results show that for water temperature inversion,the Stacking strategy based on Ridge regression achieved the best performance,with an R2 of 0.829-8.2%higher than that of the best single model(KNN).For pH inversion,the Random Forest model under the Bagging strategy performed best,yielding an R2 of 0.868,representing a 12.9%improvement over KNN.Similarly,for dissolved oxygen inversion,the Random Forest model again excelled,achieving an R2 of 0.868-11.7%higher than KNN.Furthermore,feature combinations with a limited number of spectral bands demonstrated good effectiveness within the ensemble learning frame-work.In summary,ensemble learning methods significantly enhance the capability of satellite remote sensing data for water quality inversion compared to single-model approaches,and are particularly suitable for monitoring weakly optically active water quality parameters such as pH and dissolved oxygen.This study provides an efficient and stable technical solution for dynamic water qual-ity monitoring in lakes.