Atmospheric correction of HJ-1 CCD imagery over turbid lake waters

We have presented an atmospheric correction algorithm for HJ-1 CCD imagery over Lakes Taihu and Chaohu with highly turbid waters. The Rayleigh scattering radiance (Lr) is calculated using the hyperspectral Lr with a wavelength interval 1nm. The hyperspectral Lr is interpolated from Lr in the central wavelengths of MODIS bands, which are converted from the band response-averaged Lr calculated using the Rayleigh look up tables (LUTs) in SeaDAS6.1. The scattering radiance due to aerosol (La) is interpolated from La at MODIS band 869nm, which is derived from MODIS imagery using a shortwave infrared atmospheric correction scheme. The accuracy of the atmospheric correction algorithm is firstly evaluated by comparing the CCD measured remote sensing reflectance (Rrs) with MODIS measurements, which are validated by the in situ data. The CCD measured Rrs is further validated by the in situ data for a total of 30 observation stations within ± 1h time window of satellite overpass and field measurements. The validation shows the mean relative errors about 0.341, 0.259, 0.293 and 0.803 at blue, green, red and near infrared bands. ©2014 Optical Society of America OCIS codes: (010.1285) Atmospheric correction; (010.0010) Atmospheric and oceanic optics; (010.0280) Remote sensing and sensors. References and links 1. Y. Li, Y. Xue, X. He, and J. Guang, “High-resolution aerosol remote sensing retrieval over urban areas by synergetic use of HJ-1 CCD and MODIS data,” Atmos. Environ. 46, 173–180 (2012). 2. H. R. Gordon and M. Wang, “Retrieval of water-leaving radiance and aerosol optical thickness over the oceans with SeaWiFS: a preliminary algorithm,” Appl. Opt. 33(3), 443–452 (1994). 3. A. Morel and L. Prieur, “Analysis of variations in ocean color,” Limnol. Oceanogr. 22(4), 709–722 (1977). 4. D. A. Siegel, M. Wang, S. Maritorena, and W. Robinson, “Atmospheric correction of satellite ocean color imagery: the black pixel assumption,” Appl. Opt. 39(21), 3582–3591 (2000). 5. M. Wang and W. Shi, “Estimation of ocean contribution at the MODIS near-infrared wavelengths along the east coast of the U.S.: Two case studies,” Geophys. Res. Lett. 32(13), L13606 (2005). 6. R. A. Arnone, M. Sydor, and R. W. Gould, Jr., “Remote sensing reflectance of case 2 waters,” Ocean Optics XIII 2963, 222–227 (1996). 7. X. He and D. Pan, “A practical method of atmospheric correction of SeaWiFS imagery for turbid coastal and inland waters,” Proc. SPIE 4892, 494–505 (2003). 8. S. J. Lavender, M. H. Pinkerton, G. F. Moore, J. Aiken, and D. Blondeau-Patissier, “Modification to the atmospheric correction of SeaWiFS ocean color images over turbid waters,” Cont. Shelf Res. 25(4), 539–555 (2005). 9. J. Ding, J. Tang, Q. Song, and X. Wang, “Atmospheric correction for Chinese coastal turbid waters using iteration and optimization method,” J. Remote Sens. 10, 732–741 (2006). 10. C. Hu, K. L. Carder, and F. E. Muller-Karger, “Atmospheric correction of SeaWiFS imagery over turbid coastal waters: a practical method,” Remote Sens. Environ. 74(2), 195–206 (2000). 11. K. G. Ruddick, F. Ovidio, and M. Rijkeboer, “Atmospheric correction of SeaWiFS imagery for turbid coastal and inland waters,” Appl. Opt. 39(6), 897–912 (2000). 12. P. Dash, N. Walker, D. Mishra, E. d’Sa, and S. Ladner, “Atmospheric correction and vicarious calibration of Oceansat-1 Ocean Color Monitor(OCM) data in coastal case 2 waters,” Remote Sens. 4(12), 1716–1740 (2012). 13. M. Wang, “Remote sensing of the ocean contributions from ultraviolet to near-infrared using the shortwave infrared bands: simulations,” Appl. Opt. 46(9), 1535–1547 (2007). #206330 $15.00 USD Received 14 Feb 2014; revised 14 Mar 2014; accepted 14 Mar 2014; published 27 Mar 2014 (C) 2014 OSA 7 April 2014 | Vol. 22, No. 7 | DOI:10.1364/OE.22.007906 | OPTICS EXPRESS 7906 14. M. Wang, W. Shi, and J. Tang, “Water property monitoring and assessment for China’s inland Lake Taihu from MODIS-Aqua measurements,” Remote Sens. Environ. 115(3), 841–854 (2011). 15. G. M. Hale and M. R. Querry, “Optical constants of water in the 200-nm to 200-micrometer wavelength region,” Appl. Opt. 12(3), 555–563 (1973). 16. L. Gross-Colzy, S. Colzy, R. Frouin, and P. Henry, “A general ocean color atmospheric correction scheme based on principal components analysis: Part I. performance on Case 1 and Case 2 waters,” Proc. SPIE 6680, 668002 (2007). 17. T. Schroeder, I. Behnert, M. Schaale, J. Fischer, and R. Doerffer, “Atmospheric correction algorithm for MERIS above case-2 waters,” Int. J. Remote Sens. 28(7), 1469–1486 (2007). 18. C. Hu, F. E. Muller-Karger, S. Andrefouet, and K. L. Carder, “Atmospheric correction and cross-calibration of LANDSAT-7/ETM+ imagery over aquatic environments: A multiplatform approach using SeaWiFS/MODIS,” Remote Sens. Environ. 78(1–2), 99–107 (2001). 19. J. Li, B. Zhang, Z. Chen, and Q. Shen, “Atmospheric correction of CBERS CCD images with MODIS data,” Sci. China Ser. E 49, 149–158 (2006). 20. J. Chen, J. Fu, and M. Zhang, “An atmospheric correction algorithm for Landsat/TM imagery basing on inverse distance spatial interpolation algorithm: a case study in Taihu Lake,” IEEE J. Sel. Top. Appl. Earth Observ. Remote Sens. 4(4), 882–889 (2011). 21. A. G. Dekker, R. J. Vos, and S. W. Peters, “Comparison of remote sensing data, model results and in situ data for total suspended matter (TSM) in the southern Frisian lakes,” Sci. Total Environ. 268(1–3), 197–214 (2001). 22. C. Hu, Z. Chen, T. D. Clayton, P. Swarzenski, J. C. Brock, and F. E. Muller-Karger, “Assessment of estuarine water-quality indicators using MODIS medium-resolution bands: initial results from Tampa Bay, Florida,” Remote Sens. Environ. 93(3), 423–441 (2004). 23. A. Morel, Y. Huot, B. Gentili, P. J. Werdell, S. B. Hooker, and B. A. Franz, “Examining the consistency of products derived from various ocean color sensors in open ocean (Case 1) waters in the perspective of a multisensor approach,” Remote Sens. Environ. 111(1), 69–88 (2007). 24. M. Zhang, J. Tang, Q. Dong, Q. Song, and J. Ding, “Retrieval of total suspended matter concentration in the Yellow and East China Seas from MODIS imagery,” Remote Sens. Environ. 114(2), 392–403 (2010). 25. J. Chen, W. Quan, G. Yao, and T. Cui, “Retrieval of absorption and backscattering coefficients from HJ1A/CCD imagery in coastal waters,” Opt. Express 21(5), 5803–5821 (2013). 26. J. Chen and W. Quan, “An improved algorithm for retrieving chlorophyll-a from the Yellow River Estuary using MODIS Imagery,” Environ. Monit. Assess. 185(3), 2243–2255 (2013). 27. L. Guo, P. Xie, L. Ni, W. Hu, and H. Li, “The status of fishery resources of lake Chaohu and its response to eutrophication,” Acta Hydrobiol. Sin. 31, 700–705 (2007). 28. J. L. Mueller, C. Davis, R. Arnone, R. Frouin, K. L. Carder, Z. Lee, R. G. Steward, S. Hooker, C. D. Mobley, and S. McLean, “Above-water radiance and remote sensing reflectance measurements and analysis protocols,” in Radiometric Measurements and Data Analysis Protocols, Vol. III in Ocean Optics Protocols for Satellite Ocean Color Sensor Validation, revision 4, (2003), Chap. 3, pp. 21–30. 29. C. D. Mobley, “Estimation of the remote-sensing reflectance from above-surface measurements,” Appl. Opt. 38(36), 7442–7455 (1999). 30. R. Ma, J. Tang, J. Dai, Y. Zhang, and Q. Song, “Absorption and scattering properties of water body in Taihu Lake, China: absorption,” Int. J. Remote Sens. 27(19), 4277–4304 (2006). 31. M. Rijkeboer, A. G. Dekker, and H. J. Gons, “Subsurface irradiance reflectance spectra of inland waters differing in morphometry and hydrology,” Aquat. Ecol. 31(3), 313–323 (1998). 32. D. Gurlin, A. A. Gitelson, and W. J. Moses, “Remote estimation of chl-a concentration in turbid productive waters Return to a simple two-band NIR-red model?” Remote Sens. Environ. 115(12), 3479–3490 (2011). 33. Y. Zhang, L. Feng, J. Li, L. Luo, Y. Yin, M. Liu, and Y. Li, “Seasonal–spatial variation and remote sensing of phytoplankton absorption in Lake Taihu, a large eutrophic and shallow lake in China,” J. Plankton Res. 32(7), 1023–1037 (2010). 34. C. C. Trees, R. R. Bidigare, D. M. Karl, L. V. Heukelem, and J. Dore, “Fluorometric chlorophyll a: sampling, laboratory methods, and data analysis protocols,” in Biogeochemical and Bio-optical Measurements and Data Analysis Protocols, Vol. V in Ocean Optics Protocols for Satellite Ocean Color Sensor Validation, revision 5 (2003), Chap. 3, pp. 15–25. 35. J. D. H. Strickland and T. R. Parsons, A Practical Handbook of Seawater Analysis, 2nd ed. (Board of Canada, 1972). 36. M. Wang, “The Rayleigh lookup tables for the SeaWiFS data processing: accounting for the effects of ocean surface roughness,” Int. J. Remote Sens. 23(13), 2693–2702 (2002). 37. M. Wang, “A refinement for the Rayleigh radiance computation with variation of the atmospheric pressure,” Int. J. Remote Sens. 26(24), 5651–5663 (2005). 38. C. Cox and W. Munk, “Measurement of the roughness of the sea surface from photographs of the sun’s glitter,” J. Opt. Soc. Am. 44(11), 838–850 (1954). 39. M. Wang and S. W. Bailey, “Correction of sun glint contamination on the SeaWiFS ocean and atmosphere products,” Appl. Opt. 40(27), 4790–4798 (2001). 40. H. R. Gordon and M. Wang, “Influence of oceanic whitecaps on atmospheric correction of ocean-color sensors,” Appl. Opt. 33(33), 7754–7763 (1994). 41. W. W. Gregg and K. L. Carder, “A simple spectral solar irradiance model for cloudless maritime atmospheres,” Limnol. Oceanogr. 35(8), 1657–1675 (1990). 42. R. E. Eplee, W. D. Robinson, S. W. Bailey, D. K. Clark, P. J. Werdell, M. Wang, R. A. Barnes, and C. R. McClain, “Calibration of SeaWiFS. II. vicarious techniques,” Appl. Opt. 40(36), 6701–6718 (2001). #206330 $15.00 USD Received 14 Feb 2014; revised 14 Mar 2014; accepted 14 Mar 2014; published 27 Mar 2014 (C) 2014 OSA 7 April 2014 | Vol. 22, No. 7 | DOI:10.1364/OE.22.007906 | OPTICS EXPRESS 7907 43. H. R. Gordon and D. J. Castaño, “Aerosol analysis with the Coastal zone color scanner: A simple meth


Introduction
The satellite constellation including HJ-1A and HJ-1B, launched on Sep.6,2008, is designed for monitoring the environment and forecasting the disaster [1].Two charge-coupled device(CCD) cameras (named CCD1 and CCD2) aboard each satellite, with a spatial resolution about 30m and four bands including blue, green, red, and near infrared (NIR) ones, have the potential of monitoring the color of inland waters with high spatial variation of optical properties.
Due to the complexity of water optical properties, atmospheric correction (AC) is still a challenge for the application of satellite remote sensing in monitoring the color of inland waters.The standard AC algorithm for case 1 waters [2] fails when applied to inland area, which show characteristics of case 2 waters (the definitions of case 1 and 2 waters were described by [3]).The failure results from invalid assumption of black water at NIR bands [4,5].Many algorithms have been developed for the atmospheric correction of case 2 waters using data not only from ocean color sensors (such as SeaWiFS, MODIS and MERIS) but also from sensors primarily developed for monitoring land targets (such as TM).The algorithms for ocean color sensors can be classified into three categories.The first one is the iterative algorithms [6][7][8][9], where the spectral relationship is used to calculate the NIR contribution in order to use the AC algorithm based on the black water assumption.The second one is based on the assumption of low variation of aerosol [10][11][12].The third one is the shortwave infrared (SWIR) scheme [13,14] with black water assumption in SWIR wavelengths [15].Other AC algorithms are developed by means of a multi-parameter inversion using the complete spectral information at visible to NIR bands and coupled atmospheric and oceanic radiative transfer model.The inversion is based on principle components analysis (PCA) [16] or neural networks (NN) [17] techniques.The AC algorithms for data from sensors like TM are developed using the aerosol information either provided by ocean color sensors [18,19] or in situ data [20] to calculate the aerosol scattering radiance.
Satellite remote sensing has been used to monitor the water quality in oceanic and coastal regions [21][22][23][24][25][26].However, the algorithms developed for oceanic and coastal waters fail when applied to lake waters due to the complicated optical properties, which result from the combined effects of shallow water and sediment resuspension.On the other hand, the AC algorithms described above cannot be applied to CCD (CCD means both CCD1 and CCD2 aboard HJ-1A and HJ-1B) due to the difference of band specifications.An AC algorithm is in need for CCD imagery when applied to quantitatively monitor the optical properties of lake waters.It is challenged to develop the AC algorithms for CCD over lake waters, primarily due to the poorly documented optical properties of fresh waters and the optical path variation resulted from the lake altitudes.The challenge is also due to the adjacency effects and spatial resolution.The latter is addressed in this study.CCD doesn't meet the minimum requirement for the atmospheric correction with at least two NIR or SWIR bands.As a result, other data have to be used in developing the AC algorithm.
The overall aim of this paper is to develop an AC algorithm for CCD imagery over turbid lake waters.The accuracy of the algorithm is firstly evaluated by comparing with MODIS measured spectra, which are validated using in situ data.The algorithm is further validated by in situ data.Using the CCD measured R rs , the development of an algal bloom occurred in Lake Taihu is monitored.The potential application in quantifying the concentration of water's component and the limitations of the algorithm are discussed.

Study area
Lake Taihu with an area about 2338km 2 (see the geographical location in Fig. 1(a)) is the third largest freshwater lake in China.The water in the lake is polluted by nutrient-laden river runoff and algal bloom is frequently occurred, affecting the water usage of nearby residents.As a result, there is an urgent need for monitoring the water quality and understanding the biological, optical, and ecological process in the lake.Lake Chaohu well known for fishery is one of the five freshwater lakes in China, with an area about 775km 2 .The lake is semi-enclosed since a floodgate was built in the southwest exit in 1962.Water quality is deteriorating primarily due to the terrestrial inputs and also due to the decrease of self-purification resulted from the construction of the floodgate.The fish species have decreased 40% since 1980s due to the pollution [27].The water quality monitoring is in need for promoting the fishery development and for the water usage in daily life for nearby residents.

Remotely sensed data
The imagery of CCD1 and CCD2 (see the band specifications in Table 1), with a swath width about 360km and a quantization 8bits, is provided for free by China Center for Resources Satellite Data and Application(www.cresda.com).It should be noted that the relative spectral     [28] (see the geographical locations of the observation stations in Fig. 1 and the cruises information in Table 2).The spectroradiometers with a bare fiber optic cable produces a 25° field of view.The R rs measurements were carried out at a viewing zenith and azimuth angles about 40° and 135°, respectively [29], which are fit for avoiding the ship shadow and the sun glint.The R rs derivation from the measurement of the spectroradiometers is the same as that described by [24].It is noted that the Fresnel reflectance (equivalent to 0.025 for a viewing zenith angle about 40°) was used as the sky radiance reflectance (ρ) in calculating R rs .
The measured R rs (Fig. 3) shows spectral signatures of case 2 waters.The R rs is relatively low in blue wavelengths, resulted from the absorption of organic particulate and colored dissolved organic matter (CDOM) [30].Two troughs are observed in the wavelength ranging from 600 to 700nm.The first one, around 625nm, results from secondary peak of absorption from accessory pigments such as cyanophycocyanin [31].The second one at 675nm is related to the chlorophyll-a absorption maximum in red wavelengths [32].The monthly mean R rs (in Mar., Apr., May, Aug., and Oct.) in Lake Taihu is also shown in Fig. 3(j).Seasonal variation is obviously shown in Fig. 3.The fluorescence peaks, in the wavelength range of 690-715nm, are more marked in summer, spring, and autumn than that in winter.The peaks wavelength is becoming long with the increase of chlorophyll-a concentration (Chl-a), which is higher in the phytoplankton growth season (spring-summer-autumn) [33] than in winter.It is noted that the spectra marked by ´#´ in Fig. 3(e), are from the measurements over algal bloom-laden (j) The monthly mean spectra calculated using the in situ R rs collected over Lake Taihu.Please note that the measurements in (f) and (g) are from Lake Chaohu and the others are from Lake Taihu.
water with the concentration of total suspended matter (C TSM ) and Chl-a about 219.0 mg/l and 471.63 mg/m 3 , respectively.The Chl-a were measured using a fluorometer by following the protocol described by [34] and the C TSM were measured by means of the weighing method described by [35].

Atmospheric correction of CCD imagery
Scattering radiance due to air molecule (L r ) and aerosol (L a ) account for over 90% of the topof-atmosphere (TOA) radiance (L t ) received by a remote sensor in the visible wavelengths [36].L w is less than 10%.As a result, the atmospheric correction, which removes atmospheric path radiance from L t , is a key procedure in quantitative remote sensing of water color.L t in the wavelength λ is given by Eq. ( 1) [2]: L g is the direct sun glint.L f is the scattering radiance due to whitecaps.T and t are direct and diffuse transmittance at sensor viewing direction.L r can be calculated with an uncertainty within 0.1% in blue wavelengths and within 0.05% in green to NIR wavelengths [37].For the calculation of L g , sun glint coefficient is derived using the model proposed by [38].L g can be neglected if the coefficient is smaller than the threshold (0.0001 in this study).Otherwise, L g is calculated by means of the method presented by [39].L f is calculated using the method developed by [40].The scattering reflectance ρ r , ρ a , ρ g , ρ f , and ρ w are given by 0 s / (F cos ) where x means r, a, g, f, and w.F 0 is the instantaneous extraterrestrial solar irradiance F´ reduced by two trips through the ozone layer, i. e., θ v is the zenith angle of a vector from the pixel to the sensor and θ s is the sun zenith angle.τ oz is the ozone optical thickness given by The specific absorption coefficients k oz are presented by [41] and the ozone contents U oz are from http://oceandata.sci.gsfc.nasa.gov/Ancillary/Meterological.

Rayleigh scattering
The scattering radiance due to air molecule (named Rayleigh scattering) of CCD is converted from the Rayleigh scattering radiance at MODIS bands, which are calculated using the Rayleigh lookup tables (LUTs).The dependency in λ −4 is used in the conversion [18].The conversion factor β, from response-averaged Rayleigh optical thickness to that in the central wavelength is given by Eq. ( 5): τ r (λ 0 ) is Rayleigh optical thickness in the central wavelength λ 0 of MODIS band.<τ r (λ)> MOD is the response-averaged Rayleigh optical thickness at MODIS band, calculated by Eq. ( 6). ( S(λ) is the RSR of MODIS bands [42].Using Eqs. ( 5)-( 7), β are estimated as Since the Rayleigh scattering reflectance ρ r ∝ τ r ∝ λ −4 [43], once we get <ρ r (λ)> MOD from the Rayleigh lookup tables (LUTs) in SeaDAS6.1, ρ r (λ 0 ) can be calculated using β, from which ρ r (λ) in the wavelength ranging from 400 to 900nm can be approximated using λ −4 functions.It is noted that <ρ r (λ)> MOD is derived by multiplying Rayleigh coefficients with F 0 .The coefficients are derived from Rayleigh Stokes vector for a fourier expansion in relative azimuth.The Stokes vector is interpolated from the vector in the LUTs using wind speed, solar, and sensor zenith angle.For the wavelengths longer than 900nm, ρ r (λ) are calculated by means of the single scattering algorithm [44].Using Eq. ( 8), ρ r (λ) are converted to the Rayleigh scattering radiance L r , from which L r at the bands of CCD are derived by means of band response-averaged calculation [45].

Aerosol scattering
Using the aerosol optical thickness (τ a ) retrieved from MODIS (see the validation of τ a in Appendix A) by means of the SWIR atmospheric correction algorithm described by [46], L a at CCD bands are calculated as follows.MODIS imagery on the same day as CCD is processed using the SWIR scheme and τ a at MODIS bands (see the specifications of MODIS bands at http://modis.gsfc.nasa.gov/about/specifications.php) are derived for the entire lake.
To minimize the effect of outliers on the mean value, a uniformity screen presented by [47] is applied to τ a for a box of 9 × 9 pixels indicated as ´E´ in the central lake in  α at the band 531nm is used due to the unique value for each model in the aerosol LUTs in SeaDAS6.1 [14], which are composed of eight relative humidity (rh)-specific families of models, each family spanning ten fine-to-coarse mode size fractions from 0.0 to 0.95.The models have been developed for atmospheric correction based on the range of single scattering albedos and aerosol size distributions retrieved from maritime AERONET sites (see the detailed description at http://oceancolor.gsfc.nasa.gov/REPROCESSING/R2009/aerosols/).20 models are firstly selected from the aerosol LUTs which closely bracket the rh derived from the meteorological data.The 2 models which are bracketing the MODIS retrieved Angstrom exponent are ultimately selected.Using (869) a τ , the aerosol single scattering albedo and the aerosol phase functions, the aerosol single scattering reflectance (ρ as ) at MODIS band 869nm can be derived.It is noted that the albedo and phase functions are derived by means of linear interpolation between the two models based on α. ρ as in the wavelength ranging from 400 to 900nm are interpolated from ρ as (869) by means of the spectral relationship for ρ as in the aerosol LUTs.ρ as at CCD bands are calculated from integration over the band width and are converted to ρ a using the relationship between ρ a and ρ as .All the relationships are derived by means of the linear interpolation as described above.L a (λ CCD ) is calculated from L a = ρ a F 0 cosθ s / π.The flowchart for deriving L a at CCD bands is shown in Fig. 4.
L w can be derived after subtracting L a , L r , L g , and L f from L t and R rs is calculated from Eq. (10).s ( ) ( ) / ( ( )cos ( )) where t s (λ) is the diffuse transmittance from sun to water given by [48] #

Assessment of CCD measured R rs
The accuracy of the AC algorithm developed in this study is firstly evaluated by comparing CCD measured R rs with MODIS measurements, which are validated by in situ data.CCD measured R rs is further validated using the in situ data from the observation stations within ± 1h time window of satellite overpass and field measurements.

Comparison between CCD and MODIS measured R rs
We present the comparison between CCD and MODIS measured R rs along four transects indicated in Fig. 1.CCD measured R rs is derived using the AC algorithm described in section 3. MODIS measured R rs is derived using the SWIR scheme detailed by [46].It should be noted that MODIS data at the bands with spatial resolutions 500m and 1000m were remapped with 250m by means of the bilinear interpolation method for the two lakes with relatively small area.MODIS measured R rs is validated using the in situ data from the observation stations within ± 1h time window of satellite overpass and field measurements [49].It should be noted that the in situ data was simulated by means of band response-averaged calculation) [45] using the measured hyperspectral R rs and the RSR of MODIS.MODIS measured R rs were derived by averaging a box of 3 × 3 pixels centered at the location of field measurements.The uniformity screen described in section 3.2 was applied to calculate the mean value.shows the validation results for a total of 43 observation stations and Table 3 shows the corresponding mean relative error (MRE) and STD.Both Fig. 5 and Table 3 show the lower accuracy of MODIS retrieved R rs in short wavelengths than in long wavelengths, probably due to the difference of the representative aerosol model with the true model.The difference results in the error of L a at all bands and the error increases with the decrease of wavelengths #206330 -$15.00USD Received 14 Feb 2014; revised 14 Mar 2014; accepted 14 Mar 2014; published 27 Mar 2014 [4].The lower accuracy is also possibly due to the significant degradation at blue bands.The degradation is a common occurrence as satellite instruments age in the harsh space environment.It was reported that the degradation at short wavelengths is larger than at long wavelengths for Terra MODIS [50].It is noted from Fig. 5 the tendency of smaller MODIS measured R rs than in situ data, which is probably resulted from the small ρ (0.025 in the study) used in calculating the in situ R rs .It was reported that ρ increases from 0.026 with wind speed 0m/s to approximately 0.043 when wind speed is 15m/s under the assumption of a viewing zenith and azimuth angles about 40° and 135° and a clear-sky radiance distribution for θ s about 30° [28].The small ρ results in the higher R rs than the true value.It is noted from [29] that the relative difference between R rs calculated using ρ equivalent to 0.022 and 0.034  is about 5% with the same observation geometry as that described above for a wind speed of 10m/s, θ s about 30° and for a case 1 water body with Chl-a of 2.0 mg/m 3 in a clear sky.The error in the in situ R rs used in the present study, resulted from ρ, is certainly smaller than that for case 1 waters due to that the effect from ρ on R rs decreases with the increase of water turbidity under the same sky condition.The R rs used for matchup comparison is collected under a cloud free sky condition, when the satellite image can be used.
Figure 6 shows the comparison between CCD and MODIS measured R rs at blue, green, and red bands.The relatively large difference between Figs. 6(a) and 6(b) results from the wavelength difference, with the central wavelengths about 488 and 500nm for MODIS and CCD, respectively.The comparison is further carried out quantitatively along transects in Lakes Taihu and Chaohu (Fig. 7).R rs at 500nm for MODIS in Figs.7(a) and 7(b) is interpolated from R rs at MODIS bands 488 and 555nm.It is reasonable to assume a linear relationship for R rs ranging from 488 to 555 nm, as shown in Fig. 8(a).The comparison is not carried out at the NIR band of CCD, due to the saturation of MODIS bands 748 and 869nm, which bracket the central wavelength of the NIR band and to the difficulty of interpolation in the wavelength range of 740 to 860nm as shown in Fig. 8(b).Figure 7 shows a better consistency at red and green bands than at blue bands.The difference between CCD and MODIS measured R rs results from the difference of RSR (as shown in Fig. 2).The effect of the RSR difference is shown in Fig. 9, where the R rs spectra are derived by means of bandequivalent calculation method using the RSR and in situ data.The difference in Fig. 7 is also possibly due to the spatial variation resulted from the uncertainties in geometric positioning and the temporal variation since the color of the shallow water with a mean depth about 1.9m [51] is easily affected by wind-driven resuspension.In order to evaluate the stability of the water color in Lake Taihu, the spectra were collected with 1h time interval from 0:45 to 7:45 (UTC time) on May 2, 2010 at the station located at (31.405°N, 120.034 °E).The spectra variation is shown in Fig. 10, from which we can see that the variation may reach 30% during 7 hours.Although ρ might play a role in the spectra variation, its effect on R rs is small for turbid waters as discussed above.As a result, we can make certain that the spectra variation is largely resulted from the variation of water optical properties.
Please note from Fig. 7 the larger difference near the coast than off the coast, possibly due to the adjacency effect.It was reported that the contribution from the adjacent pixels to the L t of the objective pixel increases with the increase of spatial resolution [52].As a result, CCD measured R rs is possibly higher than MODIS measured R rs near the coast.

Comparison between CCD measured and in situ R rs
CCD imagery over Lakes Taihu and Chaohu was processed using the AC algorithm described in section 3, from which R rs at CCD bands were derived and were compared with in situ data for the observation stations within ± 1h time window of satellite overpass and field measurements [48].The in situ data was simulated by means of band response-averaged calculation [45] using the measured hyperspectral R rs and the RSR of CCD.CCD measured R rs were derived by averaging a box of 3 × 3 pixels centered at the location of field measurements.The uniformity screen was applied to calculate the mean value.
Figure 11 shows four examples of the matchup comparison between CCD measured and in situ hyperspectral R rs , along with the Rayleigh corrected R rs and the CCD measured R rs using the aerosol provided by Aqua MODIS. Figure 12 further shows the matchup comparison for a total of 30 observation stations and Table 4 provides the corresponding MRE and STD.It should be noted in Fig. 12 and Table 4 that the difference of the RSR for the four CCD sensors is neglected.The first band for these four sensors is named blue band.The second, third, and fourth ones are green, red, and NIR bands, respectively.Both Figs.11 and 12 and Table 4 show better results at blue, green, and red bands than at NIR band.The poor performance at NIR band primarily results from the low signal-to-noise ratio which decreases with the low signal over water and also possibly due to the low quantization of 8 bits [10].As shown in Fig. 5, the tendency of smaller CCD measured R rs than in situ data is Fig. 12.Comparison between CCD measured and in situ R rs for the observation stations within ± 1 hour time window of satellite overpass and field measurements.The solid circles mean that R rs is retrieved using the algorithm developed in this study and the hollow circles are for R rs retrieved using the Flaash model.The solid line is 1:1 line.The number of stations (N) and the correlation coefficient(R) are also shown.R_flsh means for the Flaash model.also observed in Fig. 12.The difference in Fig. 12 is also possibly due to: (1) the difference of τ a (869) over the box of 3 × 3 pixels centered at the location of field measurements with (869) a τ derived from MODIS data.The difference of τ a (869) results from the temporal and spatial variation.The variation of τ a (869) from 0 to 8h (UTC time) in four seasons is shown in Fig. 13 by averaging all the data in the season.For example, the mean τ a (869) at 2h in autumn is calculated by averaging all the data from 2 to 2:59h.τ a are from the measurements of sun photometer CE-318 at the site named Taihu provided by AERONET (see the geographical location in Fig. 1(b)).The measurements are level 2.0 data, which are cloud-screened and quality-assured (http://aeronet.gsfc.nasa.gov/new_web/index.html).It is noted from Fig. 13 that τ a (869) is higher in spring-summer than that in autumn-winter, which was also observed by [53].(2) the temporal variation of R rs as shown in Fig. 10.The CCD measured R rs compare reasonably well with in situ data considering the turbid waters with normalized water leaving radiance at NIR band ranging from ~0.1 to over 5 mw cm −2 μm −1 sr −1 and the high temporal variation of optical properties.
The performance of the AC algorithm is further compared with the algorithm that has been established for CCD over water.The Flaash model in ENVI was used for the atmospheric correction of CCD data over waters [54,55].The visibility required by Flaash was derived from the Terra MODIS measured τ a (551).Following the steps described by [54,55], the R rs was calculated using the Flaash model and was validated by the in situ data for the observation stations within ± 1h time window of satellite overpass and field measurements.It is noted that τ a (551) is retrieved from MODIS using the AC algorithm presented by [46].Both Fig. 12 and Tables 4 and 5 show that the algorithm developed in this study has a better performance than the Flaash model.
The effect of uncertainties in MODIS measured τ a (869) on the CCD measured R rs is assessed by comparing the R rs retrieved using (1 ± MRE τ )τ a with that using τ a when α remains unchanged.The effect of uncertainties in α on CCD measured R rs is assessed by the comparison of R rs retrieved using α and (1 ± MRE α )α when τ a (869) remains unchanged.The combining effects of uncertainties in τ a and α are assessed by comparing the R rs retrieved using (1 ± MRE α )α and (1 ± MRE τ )τ a with R rs retrieved using α and τ a .MRE τ and MRE α are the MRE values in Table 7 for τ a (869) and α, respectively.Table 6 shows the effects of uncertainties in τ a (869) and α on CCD measured R rs using data on Sep.21, 2011 as a test image.It should be noted that the CCD measured R rs is derived by averaging the value for a box of 100 × 100 pixels indicated as ´E´ in Fig. 1(b).CCD measured R rs at blue band is more sensitive to the uncertainties in both τ a (869) and α than at other bands, resulted from relatively large wavelength difference with 869nm since the aerosol reflectance is interpolated from ρ a (869).

Discussions and conclusion
We have presented an atmospheric correction algorithm for CCD imagery over turbid waters in Lakes Taihu and Chaohu.Using the hyperspectral Rayleigh scattering reflectance and the relative spectral response, the scattering radiance due to air molecule at CCD bands is calculated by means of the band-equivalent calculation.The hyperspectral reflectance is interpolated from ρ r in the central wavelengths of MODIS bands, which are converted from the band response-averaged ρ r calculated from Rayleigh LUTs.The aerosol model is determined using relative humidity and Ångström coefficient calculated using aerosol optical thickness at bands 531 and 869 nm, which are retrieved from MODIS data by means of the SWIR atmospheric correction scheme.The aerosol scattering radiance at CCD bands are interpolated from L a (869) calculated using the aerosol model and τ a .The R rs retrieved from CCD data using the atmospheric correction algorithm is compared with MODIS measurements, which are validated by in situ data.The CCD retrieved R rs is further validated using the in situ data for a total of 30 observation stations within ± 1h time window of satellite overpass and field measurements.The validation shows much noise at NIR band, resulted from the low signal-to-noise ratio and from the low quantization of 8 bits.CCD measured R rs compare reasonably well with in situ data considering the complexity of optical properties and can be used to monitor the water quality in Lakes Taihu and Chaohu.Only through combining with retrieval models, the atmospheric correction algorithm can be used for quantitatively monitoring the water quality such as the turbidity, C TSM , and Chl-a.Red band has been shown to work well in calculating C TSM .A robust linear relationship was established between reflectance at MODIS band 645nm (with the wavelength ranging from 620 to 670 nm) and C TSM [56,57].The red band was used to provide a robust and C TSMsensitive algorithm for MERIS, MODIS and SeaWiFS [58] and a C TSM retrieval model using a single red band was developed for SEVIRI (Spinning Enhanced Visible and InfraRed Imager) [59].It is shown in Fig. 14  and R rs at the red band of HJ-1A CCD1.The R rs and C TSM in Fig. 14 are collected during the cruise over Lake Taihu in Oct., 2004.Water turbidity was retrieved from R rs at MODIS band 645nm by means of a linear model [60].
Blue bands should be avoided for retrieving Chl-a due to the strong absorption of CDOM, which is abundant in lake waters.A lot of research has been carried out for retrieving Chl-a in case 2 waters using red and NIR bands [61].established a three-band reflectance model for assessing Chl-a in turbid productive waters.Two red bands in the range of 670-700nm and one NIR band around 750nm are used in the model.Using measured R rs and Chl-a, three bands with central wavelengths 666, 688, and 725nm were selected by means of optimization and iteration for retrieving Chl-a in Lake Taihu [62].An improved four-band model had a better performance than the three-band model when applied to Lake Taihu [63].The central wavelengths for the four bands are 663, 693, 705, and 740nm.Overall, at least two red bands and one NIR band are needed for the retrieval model of Chl-a in productive waters.It is noted that one red band and one NIR band are included in CCD.A linear relationship between R rs at MODIS red bands 645 and 667nm was established using in situ data [60] for producing hyperspectral data set.R rs at band 640nm was simulated using R rs at 667nm, blue, and green bands in order to apply the quasi-analytical algorithm (QAA) to SeaWiFS and MODIS data [64].As a result, CCD data along with the models describing the relationship between reflectance at the red band of CCD and at another red band can be used to retrieve Chl-a in case 2 waters.
The AC algorithm is carried out with the assumption that the aerosol is spatially and temporally stable over the area of interest.τ a is an important parameter for affecting the accuracy of the CCD measured R rs .The effect is obviously shown in Fig. 11 that the CCD measured R rs using Terra MODIS derived τ a matches better with in situ R rs than that using Aqua MODIS, because of the closer overpass time of Terra than Aqua with HJ-1A/1B.The algorithm is also limited by τ a , which is provided by MODIS data.It should be noted that the revisit time for MODIS aboard Terra satellite over Lakes Taihu and Chaohu is one day.The revisit time for CCD aboard HJ-1 is two days.As a result, MODIS data can provide concomitant aerosol information for the atmospheric correction of CCD.It should also be noted that the CCD measured R rs for the pixels near the coast are doubted due to adjacency effect, which is high for remotely sensed data with high spatial resolution.
Although CCD is designed for monitoring the land targets, this study shows that the CCD data can be also applied to quantify the quality of case 2 waters (such as water turbidity, total suspended matter and chlorophyll-a) with proper atmospheric correction algorithm and

Fig. 1 .
Fig. 1.(a) The geographical location of Lakes Taihu and Chaohu.(b) The geographical location of the observation stations for the field measurements in Lake Taihu.The black, red, green, and blue triangles are for the stations in Oct., 2008, Mar., 2009, Apr., 2009, Apr., 2010, respectively.The black, red, and blue circles are for the stations in Mar., 2011, May, 2011, and Aug., 2011, respectively.(c) As in (b) but for the field measurements in Lake Chaohu.The black circles and triangles are for the stations in June, 2009 and May, 2013, respectively.

Fig. 4 .
Fig. 4. The flowchart for deriving aerosol scattering radiance.λ CCD and λ MOD are the bands of CCD and MODIS, respectively.

Fig. 1 (Fig. 5 .
Fig. 5. Comparison between MODIS measured and in situ R rs .The solid line is 1:1 line.The number of stations (N) and the correlation coefficient(R) are also shown.

Fig. 8 .
Fig. 8. R rs spectra ranging (a) from 480 to 560nm and (b) from 740 to 860nm collected during the cruise in Apr., 2009.The dotted lines in (a) are the fitted lines.

Fig. 9 .
Fig. 9.The response-averaged R rs calculated using the in situ data and RSR for the bands of MODIS and CCD.

Fig. 10 .
Fig. 10.(a) R rs collected during 0:45 to 7:45 with 1h time interval at the station located at (31.405°N, 120.034 °E).(b) The relative difference (RD) for R rs at CCD bands, simulated by means of band-equivalent calculation using the spectra in (a) and the RSR.RD = (R max -R min )/ R max , where R max and R min mean the maximum and minimum values of R rs among the 8 spectra.

Fig. 11 .
Fig. 11.Comparison between CCD measured R rs and in situ data collected on (a) Apr. 30, 2010, (b) May 7, 2011, (c) May 2, 2010, and (d) June 15, 2009 for the observation stations within ± 1h time window of satellite overpass and field measurements.It is noted that CCD measured_A means that R rs is derived using τ a retrieved from Aqua MODIS and CCD measured_T means from Terra MODIS.Rayleigh corrected R rs is also included.

Fig. 14 .
Fig. 14.Relationship between R rs at the red band of HJ-1A CCD1 and C TSM .

Fig. 15 .
Fig. 15.(a) CE-318 measured τ a .(b) The fitting of binomial distribution for τ a indicated as ´b´ in (a).(c) and (d) are the same as (b) but for the τ a indicated as ´c´ and ´d´, respectively.

Table 6 . Effects of uncertainties in τ a (869) and α on CCD measured R rs .
−0.449 means the relative error of CCD measured R rs at the blue band resulted from the error in τ a (869).