Modelling Phycocyanin And CDOM Concentration From Hyperspectral Reflectance Data In Lake Taihu

This paper took Lake Taihu as the case study area, by using massive field samples and laboratory analysis test, mathematical simulation dataset, the seasonal and space distribution of phytoplankton spectral absorption and CDOM in different regions were analyzed,upstream contributions to the CDOM in the Lake Taihu during summer rainy seasons have been researched, and we have established phytoplankton spectral absorption and CDOM retrieval models basing on bio-optical model and inherent optical properties data to carry out accuracy verification and then choose the better monitor sensing estimation model, moreover, optimized the cyanophycin model of Simis et al.(2007) by using the three-band model of phytoplankton spectral absorption which efficiently improved the accuracy of retrieval, thus enable it to be more applicable to large, turbid and shallow lake. provide scientific evidence for the remote sensing monitoring of Taihu CDOM and cyanophycin.In the context of the whole lake,the distribution of phytoplankton spectral absorption in terms of time are highest in spring and lowest in winter. From spring to summer, autumn and winter, it show a decreasing trend. Judging from space, the maximum absorption of the four seasons are of in MeiLiang Bay of Lake Taihu and gradually decrease from north to south; The minimum is often seen in the central and southern and East of Lake Taihu and the central lake area is less than coastal water. The three-band remote sensing model 2.131[Rrs-1(673)-Rrs-1(698)]×Rrs(731)+0.095 of aph(665) was calibrated and validated, and its performance was compared and assessed with the published band-ratio method. With the three-band model, the root mean square error (RMSE) and mean relative error (MRE) were 0.150 m-1 (50.5% accounting for the mean value) and 45.7% respectively; with the published band-ratio method, the values were 0.290 m-1(97.3% accounting for the mean value) and 213.0% respectively, based on an independent validation dataset. Furthermore, the three-band and band-ratio models worked well in estimating phytoplankton spectral absorption with simulated MERIS bands data with higher precision for the three-band model in Lake Taihu. The result showed that the three-band model was superior to the published band-ratio method, and thus the former can be used to improve the estimation precision of remote sensing of phycocyanin.The distribution of CDOM from the time perspective is highest in winter and lowest in spring,and which is slightly larger than the fall and is significant larger than the summer and spring in winter,and the summer is slightly larger than the spring. The time distribution and space distribution of phytoplankton are opposite; from the perspective of spatial distribution spring and winter showed the same law,that is to say, show the highest in Meiliang Bay and then Significantly decrease to the Southern region.That indicates that CDOM Mainly distributes in MeiLiang Bay in spring and winter and show the same rule in summer and autumn. The central of lake as dividing line,the north is larger than the southern,and the distribution of CDOM show no significant difference within the two lake.The mean CDOM spectral absorption at 355 nm for the sites in Zhihugang, Dapukou and Changdougang rivers was 4.76±0.79 m-1, which was significantly larger than the value of 3.62±0.84m-1 in the open water in Lake Taihu (ANOVA, p<0.005). From the upstream of river to river inflowing mouth and further to the open water in Lake Taihu, CDOM spectral absorption decreased gradually reflecting that alltochthonous CDOM input from rivers was the important contributor of Lake Taihu during summer rainy period. Using parallel factor analysis, four fluorescent components were indentified from CDOM three-dimensional excitation-emission matrix spectra. Component 1 and component 2 were protein-like fluorescence components. Component 3 and component 4 were humic-like fluorescence components. Humic-like fluorescence component gradually decreased from the mouth to the open water, while the protein-like fluorescence components significantly increased suggesting that river input mainly brought the humic-like fluorescence organic matter.The results showed that BP neural network model was superior to a single band model and the first order differential model for CDOM spectral absorption estimation. The relative root mean square error (RRMSE) and mean relative error (MRE) of BP neural network model were 15.3% and 12.1%, respectively, based on an independent validation dataset including 25 samples. Thus, BP neural network model could be better used to estimate CDOM spectral absorption in Lake Taihu.