TP Concentration Inversion and Pollution Sources in Nanyi Lake Based on Landsat 8 Data and InVEST Model

Abstract

Phosphorus is a limiting nutrient in freshwater ecosystems. Therefore, it is of great significance to use remote sensing technology to estimate the Total phosphorus (TP) concentration in the lake body and identify the contribution of TP inflow load in the surrounding area of the lake body. In this study, two main frameworks (empirical method and machine learning algorithm) for TP estimation are proposed and applied to the development of the Nanyi Lake algorithm. Based on the remote sensing data and ground monitoring data, the results obtained by the two main algorithms are compared to explore whether the machine learning algorithm has better performance than the empirical method in the TP inversion prediction of Nanyi Lake. The Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) model was used to simulate the TP inflow load in the Nanyi Lake Basin and determine the key pollution source areas. The results show that the accuracy of the machine learning algorithm is higher than that of the empirical method and has better performance. Among the four machine learning algorithms—support vector machines (SVR), artificial neural network (BP), extreme gradient boosting algorithm (XGBoost) and random forest regression (RF)—the TP concentration inversion model established by the XGBoost algorithm is more accurate and has strong spatiotemporal heterogeneity. The simulation results in the southern and northeastern parts of the Nanyi Lake Basin contribute the most to the pollution load of the lake area, and the simulation results can provide direction for the effective prevention and control of Nanyi Lake, help to further effectively identify the key source areas of TP pollution in the water body of Nanyi Lake, and provide a meaningful scientific reference for water quality monitoring and management, to comprehensively improve the water quality of Nanyi Lake.

Estimation of Non-Optically Active Water Quality Parameters in Zhejiang Province Based on Machine Learning

“…This might be because the KNN and the SVR were constructed based on the distance between variables [43,48]. Although they have performed well in some previous studies, their training sets were almost two or three hundred or even dozens [32,38,49]. When the amount of the training set or the number of input variables increases, the complexity of the model significantly increases, and the training efficiency and estimation accuracy decrease.…”

“…For example, a back-propagation (BP) neural network was successfully leveraged to estimate chemical oxygen demand (COD), the permanganate index (COD Mn ), the total nitrogen (TN), and TP [28][29][30]. In addition, decision trees, support vector machine regression (SVR), random forest (RF), and other methods are also widely used [31][32][33]. Some studies compared the performance of empirical methods and machine learning methods on the same data, and the results often showed that machine learning had higher accuracy [5,28].…”

Estimation of Total Phosphorus Concentration in Lakes in the Yangtze-Huaihe Region Based on Sentinel-3/OLCI Images

“…Therefore, cross-validation is employed to divide the dataset into different training and validation sets to minimize the impact of sample size on model performance and enhance its generalization ability [44,45]. Li et al [46] utilized 67 samples to construct a machine learning model for estimating the concentration of TP in water bodies, while Ding et al [47] constructed an XGBoost model for estimating TP concentration in Nanyi Lake using 78 samples, both achieving good results. This suggests that, by selecting suitable algorithms and employing appropriate processing methods, machine learning can still yield good validation results even with a limited number of samples.…”

Investigating Urban Flooding and Nutrient Export under Different Urban Development Scenarios in the Rouge River Watershed in Michigan, USA

“…Due to the difficulty in obtaining measured runoff depth/volume data in the UFRM model and nutrient retention data in the NDR model, past studies often do not perform model calibration [6,9,122,123]. Although we have conducted calibration based on the correlation between nutrient export and the SQI dataset, there is still room for improvement in terms of observed data in the future [124][125][126][127][128]. However, in general, the results of the model calibration meet the research needs.…”