An Algorithm For Extimating Concentration Of TSM Based On Classification Of Remote Sensing Reflectance

Our study area includes three lakes and one reservoir(i.e. Lake Taihu, Lake Chaohu, Lake Dianchi and Three Gorges Reservoir). Based on datasets which were acquired through field investigation, we analyzed apparent optical properties(AOPs) and inherent optical properties(IOPs) of these waters, and then identified the common sensitive bands for total suspended matter(TSM). Thereafter, we developed models for retrieval of TSM concentration using remote sensing reflectance(Rrs):firstly, the waters were classfied into various types by means of sensitive band combination; secondly, different retrieval models were developed relying on different water types. This algorithm was validated by in situ measured data and HJ-1 hyperspectral data, and the result showed higher accuracy than algorithms without classification previously. The conclusions of this study are listed as below:(1) AOPs and IOPs of three lakes and one reservoirLake Taihu, Lake Chaohu, Lake Dianchi and Three Gorges Reservoir all have typical optical properties of inland waters. Meanwhile, because of different sampling location and time, there also existed some discrepancy.For example, the shapes of reflectance spectra were dominated by phytoplankton in Lake Chaohu and Lake Dianchi, but by non-algal particles in Three Gorges Reservoir. Seasonal variations were observed in Lake Taihu, e.g. in spring, it was mainly influenced by non-algae particles, while in autumn, it was mainly influenced by phytoplankton.Some discrepancies also appeared in the shape and value of absorption coefficient owing to different water constituents and concentration. Though the absorption coefficients of CDOM and scattering spectral curve both exponentially decresed, the values and slopes were dissimilar.(2) Common sensitive bands for TSM concentration of three lakes and one reservoirAccording to optical mechanism and the relationship between RrsN(normalized Rrs by its integral calculated over 400—700nm) and TSM concentration,550nm, 650nm and 800nm were determined as common sensitive bands for TSM concentration of three lakes and one reservoir.(3) Classification algorithmThe bands combination [RrsN(650)*RrsN(800)]/RrsN(550), which was highly correlated with TSM concentration, was chosen as the factor to classify water types. Class 1 with low TSM concentration was defined by [RrsN(650)*RrsN(800)]/RrsN(550) <0.5, and class 2 with high TSM concentration was defined by [RrSN(650)* RrsN(800)]/RrsN(550)>0.5. The algorithm proposed in this study out-performed other classification algorithms by distinguishing low-high TSM concentration.(4) TSM concentration retrieval modelsAfter analyzed the relationship of [RrsN(650)*RrsN(800)]/RrsN(550) and TSM concentration in Class 1 and Class 2 water, a simple and easy liner model Y=117.4*X-8.114, which also has physical significance,was built for Class 1, and the two near-infrared method,which can only be applied to waters with high concentration of TSM,was built for Class 2.Because of the complexity of optical properties in inland waters, we have to develop specific retrieval model for specific waters. The algorithm introduced in this study, which suggest we should classify waters prior to estimate concentration, to some extent, can avoid the drawbacks of traditional methods.