Water quality parameters inversion based on multispectral remote sensing

To address the challenges of non-real-time monitoring and high manpower consumption in lake pollution management, this paper proposes an innovative framework integrating multispectral remote sensing technology with intelligent water quality prediction. Focusing on Hongze Lake, we establish inversion models for total phosphorus (TP) and total nitrogen (TN) through systematic spectral data acquisition coupled with outlier correction and standardized preprocessing. The adaptive boosting (AdaBoost) algorithm, Kepler optimization algorithm (KOA), and genetic algorithm (GA) are used to optimize the predictive ability of the random forest (RF) algorithm for total phosphorus and total nitrogen content. Experimental results demonstrate that these three improved models outperform conventional random forest models in predicting water quality. Notably, the KOA-RF model exhibits superior predictive performance (R 2 = 0.94 ± 0.02), followed by the A-RF model and GA-RF model. The proposed improved algorithms prove feasible for water quality prediction with promising predictive accuracy. These advancements provide critical algorithmic support for developing integrated space-air-ground lake monitoring systems. • The water quality data is preprocessed using outlier processing. • Three algorithms are used to optimize the predictive ability of the RF algorithm. • These three improved models outperform conventional RF models.