Research on Stacking Ensemble Learning-Based Remote Sensing Retrieval of Total Phosphorus Concentration in Poyang Lake and Its Multidimensional Driving Mechanisms

As a critical ecological security barrier in the Yangtze River Basin, Poyang Lake's water quality status has a decisive impact on regional ecological and environmental stability. Remote sensing retrieval of total phosphorus (TP) concentrations serves as an effective technical approach for water quality monitoring and eutrophication assessment in Poyang Lake. However, constrained by the scarcity of monitoring data and inherent limitations of conventional methods, existing TP concentration inversion techniques still face significant accuracy constraints in complex inland aquatic environments. To address this challenges, this study develops a novel inversion framework for TP concentration estimation. The framework innovatively incorporates a two-stage parameterized perturbation mechanism based on aquatic optical characteristics, simultaneously enhancing multispectral reflectance data (B02-B12 bands) and TP concentration distribution patterns through spectral feature optimization strategies. This approach successfully expands the sparse TP monitoring dataset to 14 times its original scale. Furthermore, grounded in Stacking ensemble learning theory, the framework employs LightGBM and Extra Trees algorithms—distinguished by their excellent TP inversion performance—as base learners, while integrating a Random Forest meta-learner to achieve nonlinear fusion reconstruction of TP concentrations. The experimental results showed that the constructed inversion framework achieved optimal predictive accuracy (RMSE = 0.0148 mg/L, R² = 0.6121), representing a 6.02% performance improvement compared to the best single model Extra Trees (RMSE = 0.0159 mg/L, R² = 0.5519). Spatial analysis revealed distinct regional differences in TP concentrations. Higher values (0.15–0.213 mg/L) were found in the northern and southern marginal regions, which are affected by agricultural runoff and urban wastewater discharge. Lower concentrations (0.062–0.12 mg/L) occurred in the central waters. These results are consistent with field measurements and historical pollution patterns.Conducting in-depth analysis of driving factors is not only a key link in deciphering the spatial distribution patterns of total phosphorus, but also a scientific prerequisite for achieving precise phosphorus control and targeted water quality management. By integrating Pearson correlation and PCA analyses, the study found that TP spatial heterogeneity resulted from combined exogenous inputs and endogenous releases. Meteorological factors such as evapotranspiration, precipitation, and temperature regulated these processes through hydro-meteorological coupling pathways. This research established a reliable “spectral enhancement–integrated inversion” technical framework for remote sensing–based TP monitoring in Poyang Lake.Meanwhile, through systematic analysis of driving factors, it has consolidated the scientific foundation for water quality management. It provides important technical support for water quality management in this ecologically sensitive region.