988 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 50, NO. 3, MARCH 2012 Estimation of Chlorophyll a Concentration Using NIR/Red Bands of MERIS and Classification Procedure in Inland Turbid Water Yunmei Li, Qiao Wang, Chuanqing Wu, Shaohua Zhao, Xing Xu, Yanfei Wang, and Changchun Huang Abstract—The classification criteria are established to classify the water of Taihu Lake into four classes based on above-water remote sensing reflectance (R rs ), i.e., types A to D. Among the four water types, type A spectra represented the case of waters where algal blooms or aquatic plants appeared, while type B is referred to the water with high suspended matter concentration and low chlorophyll a concentration (C chla ). Both types A and B were not suitable for retrieving C chla from image data. Hence, three-band, four-band, and two-band ratio algorithms were constructed to retrieve C chla from water types C and D. The obtained results showed that the relation trends between C chla and R rs were different between type C and type D waters. By using Medium Resolution Imaging Spectrometer images, acquired on November 11, 2007 and November 20, 2008, the C chla of Taihu Lake was mapped by band 9/band 7 models; it could be concluded that the C chla mainly ranged from 0 to 20 mg · m 3 , accounting for 83.70% of the whole lake area in 2007 image, while the area was 86.63% in 2008 image. The estimation accuracies varied from different C chla ranges. The mean absolute percent errors obtained by band 9/band 7 models were 106.23%, 56.79%, 38.04%, 33.80%, and 58.74% for the ranges 0 mg · m 3 <C chla <=5 mg · m 3 , 5 mg · m 3 <C chla <= 10 mg · m 3 , 10 mg · m 3 < C chla <= 20 mg · m 3 , 20 mg · m 3 <C chla <= 30 mg · m 3 , and 30 mg · m 3 <C chla , respectively. Correspondingly, the root-mean-square errors were 5.02, 4.45, 5.59, 8.72, and 32.55 mg · m 3 , respectively. Index Terms—Chlorophyll a concentration, Medium Resolution Imaging Spectrometer (MERIS), near infrared (NIR)/red model, Taihu Lake, water classification. I. I NTRODUCTION A S THE THIRD largest freshwater lake in China, Taihu Lake is very important for water supply to major cities along the lakeshore. However, water pollution of the lake is Manuscript received November 1, 2010; revised February 26, 2011 and May 1, 2011; accepted July 17, 2011. Date of publication September 15, 2011; date of current version February 24, 2012. This work was supported in part by the National Natural Science Foundation of China under Grant 40971215, by National Doctoral Fund under Grant 20093207110011, and by the Special Project of High-Resolution under Grant E0203/1112. Y. Li, X. Xu, Y. Wang, and C. Huang are with the Key Laboratory of Virtual Geographic Environment, Ministry of Education, Nanjing Normal University, Nanjing 210046, China (e-mail: liyunmei@njnu.edu.cn; 68698500@qq.com; 357008144@qq.com; huangchangchun_aaa@163.com). Q. Wang, C. Wu, and S. Zhao are with the Satellite Environment Application Center, Ministry of Environmental Protection, Beijing 100029, China (e-mail: wangqiao@sepa.gov.cn; wu.chuanqing@sepa.gov.cn; zshyytt@126.com). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TGRS.2011.2163199 becoming more and more serious in recent years, which has caused lake eutrophication. This problem severely threatens the normal functioning of the lake. As such, it is essential that the water quality should be regularly monitored for tracking the anthropogenic effects and sustainably managing water re- source. Fortunately, as a key indicator of water quality, chloro- phyll a concentration (C chla ) can be inversed from remotely sensed data, which is crucial for the continuous monitoring of water quality. Generally speaking, the C chla can be derived from the blue and green spectral bands [1], [2] in open-ocean Case-1 water. However, the method is not very suitable for inland productive turbid waters, because of the overlapping and uncorrelated absorptions by dissolved organic matter and nonalgal particles in the blue spectral region [3]–[6]. Thus, spectral algorithms which are based on reflectance in the red and the near-infrared (NIR) spectral regions are regarded to be preferable for estimating C chla in turbid productive waters [6]–[9]. Dall’Olmo et al. [6] validated the potential of ap- plying NIR/red reflectance ratios to estimate C chla via Sea- viewing Wide Field-of-view Sensor and Moderate Resolution Imaging Spectroradiometer data. Gons [8] developed a simple semianalytical method using wavebands centered at 665, 705, and 775 nm and later reexamined the performance by setting wavelength 708.75 nm instead of 705 nm [10]. Dall’Olmo and Gitelson [11] developed a three-band semianalytical al- gorithm for estimating C chla . Moses et al. [12] validated a three-band model and a two-band model which used Medium Resolution Imaging Spectrometer (MERIS) reflectances in the red and NIR spectral regions for estimating C chla in inland, estuarine, and coastal turbid productive waters. Some studies show that three-band model is better than the two-band ratio [13], [14]. Three-band algorithm is based on radiative transfer (RT) theory according to absorption and scattering properties of water constituents. The relationship between C chla and three band is as follows: C chla R 1 rs (λ 1 ) R 1 rs (λ 2 ) × R rs (λ 3 ) (1) where R rs (λ i ) stands for remote sensing water reflectance at wavelength λ i (i =1, 2 or 3), usually in the visible range of 400–800 nm. The efficiency of the algorithm has been proved in various lakes and reservoirs with variable optical properties [12]–[15], and the algorithm is also validated in Taihu Lake [16], [17]. 0196-2892/$26.00 © 2011 IEEE