Research On Remote Sensing Retrieval Of Water Quality Parameters For Typical Inland Waters Based On Multivariate Optimal Modeling Strategy

Inland waters are an integral part of surface water.As the ecological and environmental issues of inland waters become increasingly prominent,the multiplatform remote-sensing technique with huge advantage in the acquisition of temporalspatial information has become one of the important means for the dynamic monitoring of aquatic environment of inland waters.However,since there are complicated interactions and influencing mechanisms between the components of inland waters,it is difficult to quantitatively recognize and decompose the responses of optical features of water to its components.As a result,how to establish a universal high-precision water quality parameter remote sensing retrieval model for inland waters has become a hot topic.Based on the considerations above,with three representative inland waters(typical middle and lower reaches of Hanjiang River,Chaohu Lake and Donghu Lake)as the study area,focusing on the research objects of chlorophyll a,total suspended solids and turbidity,by following the multivariate optimal ideas of water quality parameter retrieval models of inland waters,measured hyper-spectral data and multisensor satellite remote sensing data are respectively employed to construct a water quality parameter remote sensing retrieval model for inland waters,and remote sensing retrieval experiments are conducted on water quality parameters of the water in the study area.The following work is mainly done:(1)To solve the existing problems of low accuracy and of poor expandability of water quality parameter remote sensing retrieval model for inland waters,multivariate optimal modeling strategy is proposed,and the optimal unites of retrieval models based on measured hyper-spectral data and satellite multispectral data are correspondingly extracted and analyzed.The means of establishing water quality parameter remotesensing retrieval model for inland waters are optimized,and the accuracy,stability and expandability of the model proposed are enhanced,laying a solid foundation for the response and simulation to remote sensing technique-based water contamination events and the prediction of the trends of pollutants.(2)The water quality parameter hyper-spectral retrieval model VIP-BP for inland waters is built based on multivariate optimal modeling strategy.With the spectral information of water bodies as the optimal unit of water quality parameter hyperspectral retrieval model of inland waters,by extracting and integrating the spectral information of water bodies,the spectral characteristics and their influencing factors,sensitive bands of water quality parameter and optimal spectral transformation form are obtained for Hanjiang River,Chaohu Lake and Donghu Lake in the study area.Taking advantages of VIP method and BP neural network,VIP-BP retrieval model is established.Before modeling,the effective information of water bodies in the study area has been extracted.During the modeling process,the nonlinear fitting relations between spectral reflectance and water quality concentration are fully considered.Due to the unique advantages of neural network such as high fault-tolerance and strong learning ability,the influence of non-target water quality parameters on the retrieval effect is greatly weakened.Compared with other mainstream retrieval models,VIP-BP model is more accurate,applicability and stable.(3)A water quality parameter multi-spectral retrieval model CM-BP is built for inland waters based on multivariate optimal modeling strategy in which the idea of combined modeling is introduced.With the usability of single models and the combination of single models as the optimal units for water quality parameter multispectral retrieval model of inland waters,correlation analysis and neural network algorithm are adopted for the selection and integration of single models.Compared with regular band combination models,CM-BP reduces the uncertainty through effective integration of multiple single models.(4)The influence of spatial-temporal resolution of remote-sensing satellite data on the retrieval results of river water quality parameters is discussed.Spatial resolution has an insignificant contribution to the increase in the accuracy of retrieval model of water quality parameter,but it is still a major factor to be considered for the selection of data source when the target water region has a small area or the river is narrow in width because the accurate acquisition of target water boundary will affect the overall credibility of retrieval results;only the remote sensing data with high temporal resolution can satisfy the remote sensing monitoring needs of inland waters,especially for rivers.Taking into account the retrieval results,satellite data acquisition means and fees,band settings,spatial resolution,temporal resolution,image width and other factors,domestic HJ-1A/B CCD and GF-1 WFV data can be used to satisfy the water quality parameter remote sensing monitoring needs of inland waters,and GF-1 WFV data can also satisfy the water quality parameter remote sensing monitoring needs of rivers.