A Novel Four-band Semi-analytical Algorithm For Remote Sensing Estimating Chlorophyll-a Concentration In Turbid Water

As one of important water quality parameters,Chlorophyll-a(Chla)concentration is the main indicator of water eutrophication.Establishing a Chla concentration remote estimation model,which has higher robustness and lower error,is the focus of water color remote sensing.The optical properties of turbid water are complex and varied.Different optically active substances and the radiation transmission of light in water(such as absorption and scattering)affect the remote sensing reflectance(Rrs)measured above the surface water,and the remote sensing estimation of the Chla concentration.Based on the relationship between inherent optical properties and Rrs,this study proposed a four-band semi-analytical algorithm for estimating Chla concentration in turbid water(denoted as FATW),and simulated the backscatter coefficient using the data collected from field observation and laboratory.Then,the study estimated the Chla concentration of Taihu Lake in autumn(Sept.-Nov.)with FATW model,validated the results and evaluated the robustness.The main contents and conclusions are as follows:(1)A four-band semi-analytical algorithm for estimating Chla concentration in turbid water(FATW)In this study,it was considered that there is a deviation while using the absorption of yellow matter(denoted as aym(?),the sum of dissolved organic matter absorption acDOM(?)and non-pigmented particulates absorption as(?))at the second band ?2 to remove the absorption at the first band ?1.So a new third band ?3 and a new fourth band?4 were introduced to correct the absorption deviation and backscattering deviation respectively.Meanwhile,the rationality of correction was proved using Lagrange Mean Value Theorem.(2)Backscattering coefficient simulation and its improvement of Taihu Lake waterBased on the data of Rrs and absorption coefficient,the backscattering coefficient of 50 samples of Taihu Lake water in 2016 was calculated using the bio-optical model by Gordon and the method by Li Junsheng.Then,the results were used as references to compare with QAA model and IIMIW model.It was considered that neither models could not simulate the backscattering coefficient of Taihu Lake water accurately.The QAA simulation was smaller significantly,and the IIMIW simulation was bigger significantly at longer wavelength.To make the backscattering simulated consistent with the reported,IIMIW model was improved,and then was used as input parameter of FATW model.(3)Estimation of Chla concentration in Taihu Lake in autumnFATW model and improved IIMIW model were applied to estimate Chla concentration of Taihu Lake in autumn(from Sept.to Nov.)with 83 modeling samples and 42 verification samples.The optimal bands of FATW were determined by maximizing the correlation coefficient between bands combination value and Chla concentration,and the results were as follows:the first band ?1=670nm,the second band ?2=690nm,the third band ?3 746nm and the fourth band ?4=767nm.Both the modeling accuracy(determination coefficient(R2)=0.88,root mean square error(RMSE)=7.62?g/L,mean absolute error(MAE)=6.06?g/L,mean relative error(MRE)=28.32%)and the verification accuracy(R2=0.83,RMSE=9.06?g/L,MAE=6.90?g/L,MRE=47.21%)were better than the band ratio model and three-band model.(4)Robustness evaluation of Chla concentration estimation modelsTo simulate numerous and multi-batch samples,the existing samples were randomly sampled 10000 times in the specific range of Chla concentration.The relative root mean square error(the root mean square error divided by the average of Chla concentration,denoted as R-RMSE)and the optimal band position were recorded in every random sampling.The average,variance and standard deviation of R-RMSE were calculated to evaluate the robustness of error,and the cumulative frequency in the wavelength±2nm was counted to evaluate the robustness of the optimal band position.It was considered that both the robustness of error and the optimal band position were stronger than the band ratio model and three-band model.For the water of Taihu Lake in autumn,the optimal band positions of FATW model were as follows:the first band?1 at 670±2nm,the second band ?2 at 691 ±2nm,the third band ?3 at 746±2nm and the fourth band ?4 at 767±1nm.The cumulative frequency were 96.7%,99.9%,98.2%and 99.4%respectively.To reduce the influence of yellow matter absorption(aym)and backscattering on the estimation of Chla concentration,this study proposed,established and verified the FATW model.It shows that FATW model reduced the error of Chla estimation,improved the robustness of the estimation error,and had good applicability due to the strong robustness of optimal band positions.The idea of FATW model construction is supported by bio-optical theory,and is valuable for the construction of Chla concentration estimation models in other turbid water.