| The watershed of Poyang Lake is vast and contains rich wetland resources,flora and fauna resources and forestry resources.The lake can effectively stabilize the regional water balance by its strong freshwater saving capacity,and plays an important role in ensuring food security while guaranteeing water demand for agriculture and farming.The continuous growth of industrial and agricultural water use and discharges of industrial effluents and waste water make the water quality of Poyang Lake deteriorate day by day,which directly affects the ecological function of Poyang Lake.This paper takes Poyang Lake as the study area,selects the suspended solids concentration as the index of water quality parameter inversion,uses the multi-timeseries Landsat-8 OLI,Sentinel-3 OLCI images and Landsat-8+Sentinel-3 image,combines the actual measured data of suspended solids concentration,uses RFE(recursive feature elimination)algorithm and XGBoost machine learning framework constructs the quantitative inversion model of suspended solids concentration in Poyang Lake with different image data sources,and the best model is used to perform remote sensing inversion of suspended solids concentration in Poyang Lake for three periods of April,July and October 2018,and the influence factors of model performance limitation are analyzed.And based on the inversion results of the three periods,the spatial and temporal variation patterns of suspended sediment concentration in Poyang Lake and its influence factors were explored.This study mainly draws the following conclusions:(1)The remote sensing reflectance extracted from Landsat-8 OLI image,Sentinel-3 OLCI image and Landsat-8+Sentinel-3 image are combined with the actual measured data of suspended sediment concentration,and the RFE selection method is used to filter the sensitive wavebands of the Images from three periods data.The sensitive bands for Landsat-8 OLI images are 1,2,5,and 6 bands,and for Sentinel-3 OLCI images are 1,2,3,4,5,6,7,9,10,11,15,16,17,18,and 19 bands,and for Landsat-8+Sentinel-3 image are 1,4,5,7,8,9,13,14,15,16,20,21,and 24 bands.(2)Based on the sensitive feature bands selected by the RFE algorithm,the XGBoost machine learning algorithm is used to construct quantitative inversion models of suspended solids concentration for each of the three data sources.Among them,the quantitative inversion model of suspended solids concentration based on Landsat-8 OLI image has R~2 of 0.57,RMSE of 2.98 mg/L and MAE of 2.62;the quantitative inversion model of suspended solids concentration based on Sentinel-3 OLCI image has R~2 of 0.89,RMSE of 3.06 mg/L and MAE of 2.54;the quantitative inversion model of suspended solids concentration based on fused image The quantitative inversion model of suspended solids concentration based on Landsat-8+Sentinel-3 image had R~2 of 0.93,RMSE of 2.38 mg/L,and MAE of 1.90,and this model performed the best.(3)The concentration of suspended solids in Poyang Lake in April,July and October 2018is estimated by using three models,according to the inversion results,the concentration of suspended solids in Poyang Lake in April and October is higher,the variation range of suspended solids concentration in the northern channel and the five Rivers entering the lake is larger,and the overall distribution of suspended solids concentration shows the trend of higher around the lake surface and lower in the center of the lake.The magnitude of suspended sediment concentration change in July is small,and the overall distribution is relatively average.(4)Combined with the quantitative inversion final results of suspended solids concentration,it is found that the natural influences of runoff and sediment increase the suspended solids concentration in Poyang Lake during the flood season from April to July,especially in the north of the lake and the entrance of five rivers.The man-made influences of sand mining,boating,and water conservancy hub will change the hydrological characteristics and sediment movement pattern of Poyang Lake. |
PDF Full Text Request