The statistic inversion algorithms of water constituents for the Huanghai Sea and the East China Sea

Acta Oceanologica Sinica 2004, Vol. 23, No. 4, p.617~626http://www .oceanpress.com.cnE- mail:hyxbe@263.netThe statistic inversion algorithms of water constituentsfor the Huanghai Sea and the East China SeaTANG Junwu', WANG Xiaomei, SONG Qingjun', LI Tongji?,CHEN Jiezhong', HUANG Haijun', REN Jingping41. National Satellite Ocean Application Service (NSOAS), State Oceanic Administration, Beijing 100081, China2. National Ocean Technology Center(NOTC), State Oceanic Administration, Tianjin 3001 11, China3. Hong Kong University of Science and Technology, Hong Kong, China4. Institute of Oceanology, Chinese Academy of Sciences, Qingdao 266071, ChinaReceived 11 March 2004; accepted 20 August 2004AbstractA group of statistical algorithms are proposed for the inversion of the three major components of Case -II waters inthe coastal area of the Huanghai Sea and the East China Sea. The algorithms are based on the in situ data collected inthe spring of 2003 with strict quality assurance according to NASA ocean bio optic protocols. These algorithms arethe first ones with quantitative confidence that can be applied for the area. The average relative error of the inversedand in situ measured components' concentrations are: Chl-a about 37%, total suspended matter (TSM) about 25%,respectively. This preliminary result is quite satisfactory for Case II waters, although some aspects in the modelneed further study. The sensitivity of the input error of 5% to remote sensing reflectance (Rrs) is also analyzed andit shows the algorithms are quite stable. The algorithms show a large difference with Tassan' s local SeaWiFSalgorithms for different waters, except for the Chl-a algorithm.Key words: ocean color sensing, Case -II water algorithms, statistic model1 Introductionwaters, and no universal Case -II model is avail-able so far. Another reason is that the standardThe inversion of constituents' concentra-atmospheric correction algorithm suitable totions for Case -II waters is one of the difficultCase -I waters fails over Case -II waters to sepa-tasks and also a hot topic of ocean color sens-rate accurately the aerosol scattering and wa-ing (Sathyendranath, 2000). The reasons areter-leaving radiance signals, because the Wa-multifold. One is the intrinsic complexity ofter-leaving radiances in NIR bands are not 0.the Case- -II waters. Many independently vari-The operational Case -II atmospheric correc-able components with variable localized param-tion中国煤化工ccuracy and adapt-eters influence the optic characteristics of theabilTYHCNMHG:aters has not been* Corresponding author, E- mail: jwtang@mail.nsoas.gov.cnIn China, the development of Case -II wa-618TANG Junwu et al. Acta Oceanologica Sinica 2004, Vol. 23, No. 4, p.617-626ter color sensing has also been retarded by theneuron network (NN), optimization and princi-measurement technologies until recently. In thepal component analysis (PCA), etc. Because ofyear 2000, under the support of HY-1(firstthe difficulties in Case- -II water atmosphericmarine satellite of China, means Ocean-1) sat-correction, some researchers proposed theellite ground application system, the Nationalmethod of direct inversion with top-of- atmo-Satellite Ocean Application Service (NSOAS)sphere ocean color signals with NN or optimi-invested about half million US dollars to importzation technologies based on the fact that aero-ocean color and bio- optic instruments of firstsol scatterings of different bands are followingclass and state-of art performance in the world.a much simple or“lat" way in most atmosphereAfter more than 3 a persistent efforts to im-conditions (Lee et al., 2003; Doerffer andprove the technologies of in situ measurements,Schiller, 1998; Lee et al, 2002).data processing and analysis, the quality of quan-Tassan (1994) proposed a set of empiricaltitative ocean color data set, including opticlocal algorithms for the coastal waters of theproperties and the three major constituents' con-Gulf of Naples in the Mediterranean Sea to re-centrations (Chl-a, suspended matter, the yel-trieve the concentrations of the three majorlow substance), has been improved greatly andcomponents, chlorophyll-a, suspended sedimentmade it possible for us to establish local algo-and the yellow substance or CDOM. The bands'rithms in typical China coastal waters.selection and combination are derived by theCase-II water algorithms can be classi-AOPs spectra and the IOPs features of differ-fied into three types (Sathyendranath, 2000):ent components, so the model is of good guid-(1) Empirical algorithms, which are tradition-ance to other waters. The algorithms proposedally band ratios and regression based algorithmsin this paper are partly learnt from Tassan'sand regress directly the apparent optic proper-models.ties (AOPs) or AOPs' ratios, such as water-leaving radiances, remote sensing reflectance2 Cruise area and data setor diffuse reflectance, with concentrations ofwater constituents. Although the model's gen-The Huanghai Sea and the East China Seaerality and adaptability to diverse Case -II wa-Case- -II spring cruise was carried out in Aprilter bodies are limited, it may be the most prac-2003. See Fig.1 for the cruise area and casttical and applicable one before we can obtainstations.better inherent optic properties (IOPs) and ana-The data set and the measurement tech-1ytical algorithms; (2) semi- analytical algo-nologies are given below. Most of the measure-rithms, which are based on the geophysical re-ments were taken by strictly following NASAlations of the IOPs and water components'SIMBIOS ocean optic protocols (Mueller andconcentrations, combining with some empiri-Fargion, 2002).cal relations; this type of algorithm is the most2.1中国煤化工; (AOPs)promising one in the near future but it needsaccurate measurements of the IOPs. Currently,YHC N M H Grofling and above-IOPs analyses are difficult tasks and not accu-water measurements. In the high turbidity watersrate enough, and it seems still difficult for us tothe above-water measurement is the major one,establish a reasonable semi-analytical algorithmsince the profiling method has a large error be-for gga数avaters; (3) other models, such ascause of strong absorption (self-shadowing) andTANG Junwu et al. Acta Oceanologica Sinica 2004, Vol. 23, No. 4, p.617~626619119° 120° 121° 122° 123° 124° 125° 126° 127° 128° E38°N37036°_Qingsao34°-33°320-A310.30°-Shanghai '29°Fig. 1. Area and cast stations of 2003 spring cruise in the Huanghai Sea and the East China Sea.difficulty in determining the near surface extrapo-2.3 Atmosphere optic propertieslation interval in data processing. Two profilingFour sunphotometers, one CIMEL CE317systems and two above-water systems were usedand three Solarlights Microtops- -IIs with WMOin the experiment and the data from hyper-spec-standard aerosol, ozone and water vapor channels,trometer ASD FieldSpec Dual VNIR data are usedfor this study. In clear or moderately turbid waters,were used to measure direct sun irradiance tothe profiling and the above-water results are able toderive aerosol optic thickness and ozone and va-por amount. The instruments were calibrated inbe matched within 15% in visible bands, and thehighland before cruise.two above- water results are within 10% at moststations.2.4 The three major ocean color components'concentrations2.2 Inherent optic properties (IOPs)The Chl-a and CDOM concentrations wereTwo spectrophotometers (GBC Cintra20,obtained by strictly following NASA bio-opticCarry-100) were independently operated by dif-protocols, and Chl-a was analyzed by twoferent groups (NOTC, HKUST) for the absorp-methods: in situ fluorometry with Turner- 10,tion spectra of total particles, inorganic particlesand HPLC with samples kept in a nitrogen tankand the yellow substance (or CDOM). Totalin lar中国煤化工absorption and beam attenuation coefficientswere measured by Wetlabs AC9, and the back-YHCN M H Gr (TSM) concen-scattering was measured by Hobilabs HS6. Thetrations, the water samples were filtered withAC9 and HS6 were calibrated by the manufac-0.45 μm pore size filter and vacuum filtrationsystem. After the water-sample filtering, theturers before the cruise.620TANG Junwu et al. Acta Oceanologica Sirica 2004, Vol. 23, No. 4, p.617-626container was flushed with pure water and the with trail measurement before cruise.water inside the container filtered again. Then,the filter pad was flushed with 50 cm2 distilled3 Data analysiswater for 3 times to move away the salt. The3.1 The measurements and analyses of threeweighting of the dry-weight of filter-pad wascomponents' concentrationscarried out with an electronic analytic scale with0.01 mg/dm3 accuracy. The blank filter and3.1.1 Chl-a concentrationssampled filter-pad were scaled several times andThe results of inter-comparisons of Chl-atwo successive weighting readings should bemeasurements show that there exists a big dif-within 0.01 mg/dm3.ference between the concentrations derived byNext step, the sediment (inorganic particle)different methods, and systematic differenceconcentrations were determined. The sampledfilter-pad was put inside the crucible and burntexists between HPLC and the traditional fluoro-metric technology in this experiment, while awith ethanol and then the crucible was coveredquite large difference also exists between twoand put into the 500。C muffle oven to be com-HPLC measurements. Refer to Figs 2 and 3 forbusted for 1 h. The sample was re-scaled againthe comparisons of CASIO fluorometric Chl-ato get inorganic sediment concentration after itwith HKUST's and NOTC's HPLC resultscooled down.respectively, as well as NOTC and HKUST's2.5 Other environmental parametersHPLC result. (CASIO: Institute of Oceanology,YSI cruising measurements of water qualityChinese Academy of Sciences. HKUST: Hongparameters (pH, temperature, salinity, Chl-a,Kong University of Science and Technology.NOTC: National Ocean Technology Center, Stateturbidity), CTD profile, water color, transparency,Oceanic Administration).GPS and meteorological parameters (air pressure,In the following figures, average relativetemperature, wind directions) wave also acquired.abs(x'-x)For data quality assurance, the followingerror is calculated asxmeasures were adopted:The two HPLC results have a 45% rela-(1) All critical parameters, such as the AOPs,tive error, and the biggest relative error maythe IOPs, concentrations of water constituents,be larger than 500%. Since the HPLC technol-were measured by two or three independent in-ogy for Chl-a analysis in China has just begunstruments and/or investigators; parallel waterrecently, the fluorometric Chl-a concentrationssamples were kept for post-cruise analysis forwere used to derive the Chl-a algorithm in or-Chl-a in land-based laboratories.(2) All the instruments were calibrated be-der to establish a more stable and widely ac-fore cruise and all the AOP instruments wereceptable algorithm.also tracked in situ during the cruise with3.1.2 TSM concentrationsSatlantic SQM- -II, a very stable light source.中国煤化工tal TSM from two(3) If the measured parameter was statedinve.MHC N M H Ger consistent thanin NASA ocean optic protocols, the protocol andthose of Chl-a, the average relative error wasrecommendations were followed mandatorily.only 17%, and larger errors occurred in low(4) All investigators were asked to readTSM concentrations (<3 mg/dm) and the larg-carefully the NASA protocols and be trainedTANG Junwu et al. Acta Oceanologica Sinica 2004, Vol. 23, No. 4, p.617~626621CASIO Fluoro Ch-a and HKUST HPLC -Chl-a comparison.Total suspended matter (TSM) comparison.Huanghai Sea and East China Sea, 200304 cruise100.01000.0HPLC- Chla equals 0.5473 Flu-Chl-a plus 0.35average relaive err 17%average relaive err 0.28, 号100.0里10.0只昌10.01.00.110.01 00.0CASIO Chl-a concentration/ug:dm -North branch TSM concentration mgdm-3Fig. 2. Comparison of fluorometric(CASIO) andFig. 4. Two independent total suspended matterHPLC(HKUST) Ch1-a concentrations.(TSM) concentrations.Nearly identical results of ag (400) from the twoCASIO Fluoro- .ChI-a and NOTC HPLC- Ch-a comparison.Huanghai Sea and Fast China Sea, 200304 cruisegroups (HKUST and NOTC) can be seen in Fig.5. The range of variation of CDOM concentra-average relaive err 0.40tions in our cruising area is 0.05~0.50 mr', against0.01~0.70 mr', the reported values of CDOM of望10.0most coastal waters in the world, and in someHUUST and NOTC CDOM comparison.0::0.5p011000.4CASIO Chla-a concentration/ugdm-3Fig. 3. Comparison of fluorometric(CASIO) and导0.3)HPLC(NOTC) Chl-a concentrations.E 0.2est relative error was about 200%, as can beseen from Fig. 4. For TSM large than 5 mg/dm', in general case, 80% or more of the total0.00.20.30.5weight of TSM is from inorganic minerals orHJYST ag (400) /m-sediment. So, within the error budget of Case-Fig. 5. Comparison of two independent CDOMII ocean color sensing, it is reasonable to takeconcentrations.the TSM as sediment for moderately and highlyturbid waters.extreme cases even ag (440) may be as high as3.1.3 Yellow substance (CDOM) concentrations19中国煤化工es 64~70), but theseex||YHC N M H Gcur at severely pol-The ag (400) is used as the CDOM con-luted waters in very close and small seashorecentration in this paper because the larger dif-areas.ference occurred in comparing the results fromAs commonly recognized, the CDOM spec-tw5毋数骚ndent measurements at ag (440).tra absorption can be described as an exponen-622TANG Junwu et al. Acta Oceanologica Sinica 2004, Vol. 23, No.4, p.617~626tial decreasing curve, ag()=ag(400) exp[-Sxmethod are given in Fig.7 (measured with ASD(- 400)], where s values from the cruise are givenFieldSpec Dual VNIR 350~ 1050 nm).in Fig. 6. They span a large range, and withoutSpectra of Huanghai Sea and East China SeaCDOM S values (0 m)200304 snring cnuise0.07Huanghai Sea and East China Sea, 200304 cruise0.06Ave equals 0.017 6, Min 0.013 3, Max 0.027 30.050.025. 0.040.020s 0.030.020.0150.01日0010(0.005020406080-0.01350450 550 650 750 850 950 1050Station No.Wave length/mFig. 6. The CDOM S values of Huanghai Sea andFig. 7. Typical Rrs spectra of Huanghai Sea andEast China Sea (NOTC result).East China Sea (ASD FieldSpec Dual VNIRspectrometer data).explicit trends along with the cruise lines or wa-ter regions. The average S value was derived byThe main uncertainty for the above-waterregression of the 0 m data of Huanghai Sea andmethod is the sky light reflectivity of waterthe East China Sea. The S values from two inde-surface, p。 w, and we found that the recom-pendent measurements are 0.017 6 for the NOTCmended value of 0.028 may be too big underand 0.018 4 for the HKUST respectively. Interna-calm sea and clear sky situation. As one cantionally reported values of S are 0.011~0.019 andvalidate it by inspecting the NIR (>850 nm) Rrs0.014 is taken as a commonly accepted average.in clear waters, the value 0.028 was just usedfirstly in most cases and NIR Rrs data were re-3.2 Apparent optic data analysischecked whenever the water color number wasThe AOPs data used to derive the models inwithin 6~9. Another way to verify the relationthis paper are mainly based on above-surfacewas using Rrs~bb/(aw+bb) and see if the Rrsmeasurements, because the profiling methodratios in NIR bands are inverse to the ratios ofmeets some difficulties in high-turbidity waters, pure water absorptions of these bands sincesuch as strong absorption and self-shadowingthe pure water absorption is very high and domi-effect, the light level attenuates too fast to con-nant and backscattering is nearly uniform in NIRduct accurately data smoothing and extrapola-range(Tang et al, 2004; Tang, 1999). Thesetion to derive just beneath water surface (0 m)treatments of the water surface reflectivity wereparameters. The viewing geometry setting oftaken中国煤化工SA protocols.above-water measurement is (40°, 1 35°) basedTHCNMHGon NASA ocean optic protocols (Mueller and4 Statistical algorithmsFargion, 2002).The remote sensing reflectance (Rrs) spec-On the basis of the data set and data analy-tra fro毋数辖e stations by the above-watersis described above, and learning from theTANG Junwu et al. Acta Oceanologica Sinica 2004, Vol. 23, No. 4, p.617~626623models proposed by Tassan(1994), the follow-nate the influence from yellow substance anding models and algorithns are established.sediment, refer to Formula (1).Define X :=[R(,)/R(A)] [R(h, ,)/R(A,)]*,4.1 Chl-a inversion algorithmsand4.1.1 Tassan modellgC= C.+c,1g(X )+c.2xlg(X),(1)Tassan( 1994) model was based on the ra-where A=443 nm; h=555 nm; h =412 nm; n,=490tio of bands located at Chl-a absorption and re-nm; a is a region dependent constant. The resultflectance peaks, and used second ratio of bandsof this algorithm is shown in Fig. 8 for the wholecruising area after deleting some abnormal stations.located at the wings of Chl-a absorption to elimi-Tassan algorithm resultTassan model result for color number 6~9 stationsHuanghai Sea and East China SeaHuanghai Sea and East China Sea, 200304 cruise100.010.0p◆910.0是y.1.01.0”gg0.110Measured Chl-a concentration/μg m-3Measured ChI-a concentration/ug m-3Fig. 8. Tassan ChI-a algorithm results for all stations (a) and for low-turbidity stations (b). a. N=80, r=0.7, averagerelative error equals 0.3. X. =(R./Rsx) (Rx2LR.o)", lgC=cg+clgX .+c2x1X, where a= -1, c.=0.5403,c=-1.997 7,c,=1.051 3. b. N=36, r=0.9, average relative eror equals 0.1. X.<(Rl!ss (Rux(Roo)", lgC=c+C|lX.+e,x12X,where a=-1, c=0.524 0, c= -2.490, c,=0.310 3.It can be seen from the result that theselected the 5 10 nm band to correct sedimentTassan model is usable in the Huanghai Sea andinfluence and used a quadric formula. Thisthe East China Sea, but the local parameters, a, .makes a lttle better improvement (see Fig. 9).Co c and C2, need adjustment.As the sediment load seriously affects theIn order to see ifTassan model is more suit-Chl-a retrieval, we also made a sub-algorithmable to the mid- and low-turbidity waters, a re-for the ocean color number 6~9 stations cover-gress is made for the ocean color number 6~9ing the mid- and low-turbidity waters (coveringstations, which have much lower sediment loadsmost area of the Huanghai Sea and the East China(TSM less than 5 mg/dm). As shown in Fig.Sea) and color number 10~21 stations covering8b, the result is much better than that in Fig. 8a.the mi中国煤化工(covering somepart Oolor number 6~94.1.2 Modified Tassan Chl-a modelstatioTYHCNMHGoandtheaver-Considering that there was still some re-age relative error equals 0.17, even the regionalsponse of Chl-a at 490 nm band in Case -II, weindex a is the same as Tassan's (1994). In color624TANG Junwu et al. Acta Oceanologica Sinica 2004, Vol. 23, No.4, p.617-626NSOAS Chl-a algorithm result[R(2)/R(A)]", where b is a region dependentHuanghai Sea and East China Sea, 200304 cruisefactor, a=555 nm, a=670 nm, A=490 nm, and100.0pthe sediment concentration is obtained bylgS=s。+s,xlgX.(2)In our cruising area, with adjusted local para-言室10.0meters, this model's 2=0.79 and the average rela-tive error equals 53%, much lower than expected.So, a new TSM model is proposed here.1.0The new TSM model works very well,as shown in Fig. 10. During our model testing,even not deleting any data point, the r=0.920.110.0100.0and the average relative error equals 30%, COV-In situ fluorometric Chl-a concentration/μ g.dmering 0.6~1 760.0 mg/m3 concentrations.Fig. 9. Modified Tassan Chl-a from Huanghai Seaand East China Sea, 2003 spring cruise. 78 stations,Total Suspeded Matter (TSM) Algorithm Resultr=0.7, average relative error equals 0.3. X =(R;/Huanghai Sea and East China Sea, 200304 CruiseR) (RIR.o)", lgC=co+cJgX.+c,x1*X, where a=10 000.0-1, c=0.477 80, c= -2.858, c2= -0.084.1000.0number 10~21 stations (mid- and high-turbid),we can only derive an algorithm with r=0.71and the average relative error equals 0.45(omitted here). These indicate that (1) the TassanChl-a model is more suitable for the mid- to low-turbidity Case -II waters; (2) in turbid waters,10.0 100.0~ 1000.0 10000.0more data and model work are needed to im-Measured TSM concentration/mg.m3prove the Chl-a retrieval accuracy; and (3) sta-tistical or empirical models may not work wellFig. 10. New TSM model for Huanghai Sea and Eastin high turbid Case- -II waters.China Sea. All 83 stations, r=0.92, average relative4.2 Total suspended matter inversion algo-error equals 0.3. Ig==<5<(