Remote Sensing Of Algal Column-integrated Biomass For Inland Waters Based On Soft Classification

The rapid increase of algal biomass in lakes is an important phenomenon of lake eutrophication.Chlorophyll a is a major pigment of algae and its concentration is usually used to indicate the algae biomass.Recently,remote sensing has been regarded as a reliable technology for monitoring algal biomass through satellite data.Compared with the traditional in situ sampling method,the remote sensing data capture the lake information from a wider area relatively and more easily.However,there is no universal Chla algorithm for Case II waters mainly due to the complex bio-optical properties across different systems,especially inland water systems.The issue,together with the regional heterogeneity,restricts the application of remote sensing in inland waters.Besides,the complex vertical structure of inland waters usually makes the remotely sensed information fail to represent the whole water column.Regarding issues mentioned above,this paper proposed an improved Fuzzy C-Means method(namely FCMm)for softly clustering water spectra data;proposed a soft classification-based blending framework for estimating Chla(namely Blend?FCMm);proposed a column-integrated biomass model by considering the different vertical pattern.The work of this paper intends to improve the applicability of remote sensing in optically complex inland waters and to provide technical support for the research on the lake eutrophication process.The main research contents and conclusion of this paper are as follows:(1)Soft classification for water optical data(FCMm).The determination of the fuzzifier parameter(m)of FCM is the key factor.But the default m value usually fails to obtain the rational cluster structure,especially for high dimensional data like water spectra.This paper proposed an improved FCM method by optimizing m value before clustering.Based on a total of 2306 spectral data in 35 study areas around the world,17 optical water types(OWT)were finally obtained on the Sentinel-3 OLCI(Ocean and Land Colour Instrument)band setting(412,442,510,560,620,666,681,708,754,and 866 nm)and the optimized m value 1.5.The assessment using OLCI data in Lake Hongze,Lake Erhai,and Lake Taihu(and its time-series data)shows reasonable spatial patterns from FCMm results.(2)Chla estimation through a soft classification-based blending framework(Blend?FCMm).An adaptive algorithm assessment method was proposed to evaluate the performance and return the score of published Chla algorithms for each OWT.Based on the evaluation result,optimal algorithms were assigned to all OWTs and used to produce the final Chla result by algorithms blending.The validation through in situ data and satellite match-up data show Blend?FCMm a good performance across all water types with absolute mean error0.204 and 0.236 lg?g/L as well as effective ratio 90.99%and 84.804%,respectively.Compared with any optimal algorithm and previous algorithm blending frameworks,Blend?FCMm performs better.The application of Blend?FCMm on satellite data indicates that Blend?FCMm can synchronously estimate reasonable Chla in productive,suspended matter-dominated,and relatively clean inland waters.The time-series image data in Lake Taihu shows that Blend?FCMm could provide Chla maps both in quality and quantity,compared with other algorithms.(3)Estimating CIB in Lake Taihu by considering different vertical patterns.Two distinct patterns were observed,i.e.,the surface accumulated and the uniform/mixed distributed,by combining the in situ data and the simulation of the algae growth model.A model for estimating penetration depth of water column was proposed to refine the“surface”depth that remote sensing could be observed(with absolute mean error 0.071 m assessed by the in situ data).Based on the refined surface concentration of Chla,the CIB model was built by fitting the logistic function to depth and CIB model coefficients for each vertical pattern.Comparing with previous models,the proposed CIB model performs better across all depths with an absolute mean error of 10.527 mg/m~2.By applying the CIB model to OLCI images,the temporal and spatial distribution of CIB in Lake Taihu was obtained,showing obvious regional and seasonal differences.The total biomass of Lake Taihu was calculated through the interpolated time-series data from 2016 to 2020.The average value of total biomass is 89.89 t with the minimum and maximum as 20.12 t and 350.77 t,respectively.The monthly biomass is higher in summer and autumn,but lower in spring and winter.The change of biomass was significantly positively correlated with accumulated precipitation(r=0.578)and air temperature(r=0.792)but significantly negatively correlated with air pressure(r=-0.726).The effect of wind speed on biomass has no significant correlation on the monthly scale but has an inhibitory effect on the biomass on the daily scale.