Full report

Original page 1

NASA Technical Memorandum 104566, Volume 4 SeaWiFS Technical Stanford B. Hooker and Elaine Report Series R. Firestone, Editors Volume 4, An Analysis of GAC Sampling Algorithms: A Case Study Charles R. McClain, Eueng-nan and Gary Fu November 1992 Yeh, (NASA-TM-IO4566-Vol-4) SEAWi_S N93-13382 TECHNICAL REPORT SERIES. VOLUME AN ANALYSIS ALGO&ITHMS: 24 p OF GAC SAMPLING A CASE STUDY (NASA) Unclas ,3_/48 0130519

Original page 2

_p -. =

Original page 3

NASA Technical Memorandum 104566, Volume 4 SeaWiFS Technical Report Series Stanford B. Hooker, Editor Goddard Space Flight Center Greenbelt, Maryland Elaine R. Firestone, Technical Editor General Sciences Corporation Laurel, Maryland Volume 4, An Analysis of GAC Sampling Algorithms: A Case Study Charles R. McCiain Goddard Space Flight Center Greenbelt, Maryland Eueng-nan Yeh and Gary Fu General Sciences Corporation Laurel, Maryland National Aeronautics and Space Administration Goddard Space Flight Center Greenbelt, Maryland 20771 1992

Original page 4

Original page 5

C.R. McClain, E. Yeh, and G. Fu ABSTRACT The Sea-viewing Wide Field-of-view Sensor (SeaWiFS) instrument will sample at approximately a 1 km resolution at nadir which will be broadcast for reception by realtime ground stations. However, the global data set will be comprised of coarser four kilometer data which will be recorded and broadcast to the SeaWiFS Project for processing. Several algorithms for degrading the one kilometer data to four kilometer data are examined using imagery from the Coastal Zone Color Scanner (CZCS) in an effort to determine which algorithm would best preserve the statistical characteristics of the derived products generated from the one kilometer data. Of the algorithms tested, subsampling based on a fixed pixel within a 4x4 pixel array is judged to yield the most consistent results when compared to the one kilometer data products. 1. INTRODUCTION Early in the SeaWiFS design phase, questions arose regarding what the best scheme would be for producing reduced resolution global area coverage (GAC) data from the high resolution local area coverage (LAC) data generated by the scanner. The primary motivation for conand the Rayleigh radiance at 443nm. L2GAC also allows the user to specify the atmospheric correction algorithm, the AngstrSm exponents, the ozone optical depths (default is computed from the Total Ozone Mapping Spectrometer (TOMS) Dobson Unit value at the center of the scene), the sensor calibration, the land/cloud threshold (750 nm), the haze threshold (670nm), the water radiance sealing, sidering schemes other than a fixed pixel subsampling (a and the water radiance product (upwelled subsurface wapredefined element in a pixel array) was to maximize the number of cloud-free pixels. Other considerations included reduction of sensor noise (average-value techniques) and error introduced by high aerosol concentrations or clouds with low albedo (least-value techniques). A comprehensive study of GAC sampling techniques using Landsat and Advanced Very High Resolution Radiometer (AVHRR) data has been published by Justice et al. (1989) who found fixed pixel subsampling to be the best representation of the original full resolution data. In this study, an analysis was performed on a single scene from the southeastern U.S. coast. The scene was ter radiance or normalized water-leaving radiance). Ringing correction was not applied to the data (Mueller 1988). L2GAC generates standard SEAPAK 512x512 pixel images by filling in the 4x4 pixel area with constant values. It is important to note that any pixel that fails either the land and cloud or haze threshold test is excluded from the analysis. In the case of fixed pixel subsampling where the pixel fails a threshold test, the entire 4x4 pixel area is assigned a grey level of either 0 (land) or 255 (cloud) to indicate an invalid pixel which is then excluded from the statistical analyses. The methods tested are described in Table 1, with the method numbering convention being the selected because it encompasses both Case 1 and Case 2 same as in the L2GAC program. waters (Morel and Prieur 1977). Of the five methods used in this study, the fixed pixel subsampling produced the best statistical fidelity to the LAC product. The other approaches can show significant deviations from the LAC statistical properties. 2. DATA PROCESSING Various methods for reducing the resolution of an image were tested in order to compare their adequacy in retaining the statistical characteristics of the high resolution derived products. The SEAPAK (McClain et al. 1991) program L2GAC provides these methods as an option for the user while processing CZCS level-2 products. The GAC data analyses used in this study operate on 4x4 (pixelxline) blocks of data. L2GAC supports the AVHRR GAC generation scheme which operates on a 5 x 3 block of data, but it was not considered for this study. The level-2 products include the water-leaving radiances at 443, 520, and 550 nm; the aerosol radiance at 670 nm; the pigment concentration; Table 1. GAC generation methods and the mechanisms used for each. Method Generation Mechanism 1 Average level-2 product (level-2 products are generated and then averaged). 1 2 Fixed pixel subsampling (pixel [2,2] from the 4 x 4 array). 3 !Mean radiance product (level-2 products are generated from mean radiances). 4 Lowest 670nm radiance pixel. 5 NOAA/AVHRR subsampling which uses a 5x3 array of data. 2 6 Lowest 750 nm radiance pixel. 7 Pigment concentration derived from mean Lw values as computed in Method 1. 8 Pigment concentration derived from mean Method 1 log(concentration) values. 1. Implemented only on the data processing system. 2. Not used in this study.

Original page 6

An Analysisof GACSamplingAlgorithms:A CaseStudy In thecasesofaverage-valueandleast-valuealgorithms, Table 2. Level-2 processing parameters. thepixelsthat fail theflagcriteriaareignored.Thenum- Parameter Value berofvalidpixelswithinthe16pixelarraymaybeasfew Level-1 Image Name: SNG:5106Bx.IMG asoneandthe arraywill still be assigneda validvalue. Y4NG:5106B-0-L2x.IMG Thus,implementationof average-radiance-value(Method Level-2 Image Name: Processing Day/Time: 15-Oct-1991/14:21:33 3) andleast-radiance-value(Methods4 and6) algorithms Orbit Number: 5106 wouldrequireon-boardprocessingfor GACdatastorage 10.000 Tilt Angle: onthespacecraft.Methods1,7,and8 canonlybeimple- Sensor Gain: 1 mentedon theground. Wavelengths: 443, 520, 550, 550, 670 nm Toprovidesomebackgroundonthe standardanalysis 1979/301/18.766 Scene Year/Day/Time: methods(Gordonet al. 1983),the total radiancereceived Thresholds by theCZCS is governed by the equation (1) Pigment Algorithm: Two Channel ! Lt(A) = t(A)Lw(A)+ L_(A) +L.(A), Land/Cloud (750nm): 21 counts Haze (670 nm): 255 counts Water Radiances: Normalized where, )_ is the wavelength, Lt is the total radiance, Lw is Water Rad. Range: 0.0, 3.0 the water-leaving radiance, t is the diffuse transmittance is Mean Solar Flux: 186.96, 187.02, 186.81, of the atmosphere, L_ is the aerosol radiance, and Lr the Rayleigh radiance. Lr depends upon the orientation between the sun, Earth, and satellite, and, for this analysis, Lw(670) is assumed to be zero. of 0.0125 La(A) is related to L_(670) through an expression the form (2) Yes where, n(A) is the /ngstrSm exponent. For the present Rayleigh Calculations: Exact 2 153.09 Optical Thicknesses Ozone: 0.0011, 0.0144, 0.0279, Rayleigh: 0.237, 0.123, 0.098, 0.044 Angstr6m Coefficients: 0.0, 0.0, 0.0 Epsilon Coefficients: 1.0, 1.0, 1.0 ILT Record Option: Solar Zenith Angle: 44.491 ° (at center) Satellite Zenith Angle: 11.370 ° (at center) analyses, the/ngstrSm exponents were assumed to be 0. Solar Azimuth Angle: 95.673 ° (at center) In this case, aerosol radiance equals La(670) multiplied by a ratio of the solar constants times an exponential function of the solar and spacecraft zenith angles and the ozone optical thicknesses. From the water-leaving radiances, pigment concentration is calctflated using an equation of the form [ Lw(.X) ], (3) [chlorophyll a + phaeophytin] = a [Lw) where, pigment concentration is considered to be the sum of the concentrations of chlorophyll a and phaeophytin, is a positive number, and fl is negative. A CZCS scene of the United States East Coast, covering the area approximately from 70-90°W and from 26for threshold. In all cases, the high values are isolated, oc- 34°N, on October 28, 1979, was processed using L2GAC this test. The full-resolution level-2 products were generated by another SEAPAK program, L2MULT. The inputs All SEAPAK's scaling convention). Most cases are associated used in the level-2 processing are provided in Table 2. wavelength dependent parameters in the table are given in order of increasing wavelength, and three digit day references are references to the sequential day of the year (February 1 being the 32nd day of the year). The method Satellite Azim. Angle: 290.502 ° (at center) Total Rad. Correction: Method of R. Evans a Water Rad. Iteration: None 4 1. Gordon et al. 1983. 2. Gordon et al. 1988. 3. Unpublished. 4. Smith and Wilson 1981. 3. RESULTS COLOR PLATES 1-8 (see envelope on back of cover for all PLATES) show the pigment images for the full-resolution processing, Methods 1-4, and Methods 6-8, respectively. Of these images, the most striking is PLATE 2 for Method 1 which shows GAC blocks with high pigment values in the Gulf Steam and Sargasso Sea regions. Apparently, these high values are artifacts of the imperfect cloud detection cur in the vicinity of a cloud, and always produce pigment values of 39 mg m -3 (a grey level of 254 based on with the small scattered clouds over the Gulf Stream and not the larger cloud bank along the eastern portion of the image. Therefore, the problem does not appear to be related to sensor ringing, but is possibly related to subpixel of Evans was used for total radiance correction which is size clouds. Generally, the effect results in low normalunpublished but briefly discussed in McClain et al. (1992). - 2 ized water-leaving radiances in the 443, 520, and 550 nm

Original page 7

C.R.McClain,E.Yeh,and(3.Fu Table 3. Statisticalsummaryof variousmethods.Analysesof the full resolutionarelabeled"Full". Because theGAC methods fill the entire 16 pixel block with constant values, the "Valid Count" is not the true number of independent pixel )airs. Parameter Cross-correlation Cross-correlation Mean Standard Valid Between Methods Coefficient Value Deviation Count LWN(443) Full 2 1.527 0.759 131,413 1.518 0.757 131,344 2 and 0.777 1.584 0.780 164,160 2 and 0.772 1.566 0.798 164,160 2 and 0.755 1.752 0.848 164,160 2 and 0.760 1.686 0.831 _ 164_160 __ LWN(520) Full 2 0.807 0.386 131,413 0.801 0.378 131,344 2 and 0.568 0.874 0.447 164,160 2 and 0.567 0.867 0.453 164,160 2 and 0.581 1.033 0.555 164,160 2 and 0.531 0.936 0.478 164_160 LWN(550) Full 2 0.505 0.235 131,413 0.503 0.231 131,344 2 and 1 0.757 0.511 0.223 164,160 2 and 3 0.755 0.507 0.225 164,160 2 and 4 0.677 0.558 0.217 164,160 2 and 6 0.687 0.547 0.234 164_160 La(670) Full 2 0.467 0.367 131,413 0.468 0.363 131,344 2 and 0.512 0.527 0.290 164,160 2 and 0.511 0.525 0.291 164,160 2 and 0.261 0.310 0.226 164,160 2 and 0.283 0.351 0.223 164,160 Pigment Full Concentration 2 1.320 5.343 131,413 1.364 5.471 131,344 2 and 1 0.411 2.046 6.149 164,160 2 and 3 0.358 1.405 5.537 164,160 2 and 4 0.268 1.108 4.664 164,160 2 and 6 0.300 1.176 4.790 164,160 2 and 7 0.356 1.370 5.462 164,160 2 and 8 0.341 1.294 5.026 164,160 derived products. Because of the lower number of valid parison of the full-resolution and Method 2 histograms values in a GAC block containing clouds, the weight of show high fidelity as would be expected, but with some the high value is amplified which biases the average pigment value of the scene to higher values and dramatically modifies the frequency distribution as discussed below. Histograms of the level-2 images were generated using the SEAPAK program HIST. Pixels flagged as being land, clouds or saturated in the 670 nm band were excluded from the analyses. The histograms in Figs. 1-4 compare the frequency distributions of the normalized water-leaving radiances [LwN(443), LWN(520), LwN(550)] and aerosol scatter due to the fact that Method 2 has a greatly reduced number of valid samples. Methods 1, 3, 4, 6, 7, and 8 all show elevated peaks at low pigment concentrations. Also, note the histogram minimum at 1.5 mg m -3 in all pigment histograms except Methods 1 and 8. This is due to the algorithm switching mechanism in the twochannel bio-optical algorithm (Denman and Abbott 1988; Muller-Karger et al. 1990). Methods 3, 4, 6, and 7 clearly tend to over estimate pigment concentrations for the range radiance [La(670)] for the full-resolution and Method 2 above 1.5 mg m -3. Method 1 produces greatly exaggerproducts. The histograms of Figs. 5-11 compare the frequency distributions of pigment concentration for the fullresolution data and the seven GAC methods. The comated values at high concentrations while Method 8 results in reasonably good estimates in this concentration range. Scatterplots of the data products from Methods 1, 3, 4, 3

Original page 8

An AnalysisofGACSamplingAlgorithms:A CaseStudy 0.50 FULL RESOLUTION * x.x** METHOD 2 0.40 0.30 < x ),_o¢ x x x x (9 x x 0 I-- 0.20 x J / _'_lx x x DO x x x "F 0.10 / x 0'000 .06 .... "i-I _)! .lC) ........ 1.00|......... LWN 445 1.40 ..-..1.05 < r'r" 0.70 (9 O l--- U') T 0.35 x x 0.00 000 0.50 1.00 LWN 520 * x i X x) Xx x x x x x x x Xx x X X I .)l .... 1 [',2 O0t l l'' rTI I2' .) ........ 3.00| FULL RESOLUTION x×x×. METHOD 2 1.50 2.00 2.50 3.00 (rnW,/¢rn2/urn/sr) Figs. 1 and 2. Histograms comparing the distributions of LwN(443) [top] and LwN(520) [bottom], respectively, as derived from the full resolution and Method 2 analyses. 4

Original page 9

C.R.MeClain,E. Yeh,andG. F_a 2.10 1.75 1.40 cx <£ rY 1.05 (9 O I--- 0.70 ill x x 0.35 0.00 I [ I J j I J I I J I J I J j I I" [J'_ 0 O0 0.50 1.00 FULL RESOLUTION xxxxx METHOD 2 ]" _ [_ r I TT1 1 • 1 ' I ....... J ..... , • [, I 1.50 2.00 2.50 3.00 LWNSSO(n W/cr 2/ m/ r) 1.5 x x x 1.2 _0.9 <£ Et/ (_9 © 1__0.6 x 09 T 0.3 xxx x t x x x FULL RESOLUTION _xx METHOD 2 0.0 i ! ! i i i f I ! I i i i i i ! i I ! i t i ! ! ! i i | ! i ! 0.0 0.5 AEROSOL 1.0 I.5 2.0 (mW/cm2/um/sr) Figs. 3 and 4. Histograms comparing the distributions of LwN(550) [top] and La [bottom], respectively, as derived from the full resolution and Method 2 analyses.

Original page 10

An Analysisof CACSamplingAlgorithms:A CaseStudy 1.5 x x_ x <£ 13/0.8 x (.9 O t-- (J3 T 0.4 0.0 I 2 FULL RESOLUTION xxxxx METHOD 1 x x x x x x x x x x x x x x x x x x x x 5 4 5 6 PIOMENT (MO/MS) K XX" 0 1 2 FULL RESOLUTION xxx. METHOD 2 X X 5 4 5 6 PIGMENT (MG/MS) Figs. 5 and 6. Histograms comparing the pigment concentrations of the full resolution analysis with those obtained from Methods 1 [top] and 2 [bottom], respectively. 6

Original page 11

C.R. McClain, E. Yeh, and G. b"u 1.6 ..-..1.2 CEO.8 (.9 0 l-or) T 0.4 X X xx 0.0 0 1 2 FULL RESOLUTION x,(x ,((METHOD 3 x x x x x 3 4 5 PIGMENT (MG/M3) x x O0 0 1 2 -- FULL RESOLUTION ,,,(xxxMETHOD 4 X 3 4 5 6 PIGMENT (MC/M3) Figs. 7 and 8. Histograms comparing the pigment concentrations o[ the full resolution analysis with those obtained from Methods 3 [top] and 4 [bottom], respectively.

Original page 12

An Analysisof GACSamplingAlgorithms:A CaseStudy 1.6 p_1.2 x < (V 0.8 (D O kbO :I: 0.4 XxX 0.0 1 2 -- FULL RESOLUTION ****x METHOD 6 x xx Xx x X X X 3 4 5 PICMENT (MC/M,3) 1.6 x ! x x F F_1.2 xx x (. 0.8 o U) 0.4 x FULL RESOLUTION xxxxx METHOD 7 xx x Xx x Xx x x x x x x x x x )_x 0.0 0 1 2 x x x 3 4 5 6 PIOMENT (MC/MS) Figs. 9 and 10. Histograms comparing the pigment concentrations of the full resolution analysis with those obtained from Methods 6 [top] and 7 [bottom], 8 respectively.

Original page 13

C.R. McClain, E. Yeh, and G. Fu 1.6 _1.2 :5 .< rYo.8 0 0 I-- U-) I 0.4 x FULL RESOLUTION }( _(x x x METHOD 8 x x x x xx >( x x x xx Xx 0.0 0 I 2 x x 3 4 5 6 PICMENT (MO/MS) Fig. 11. Histogram comparing the pigment concentrations of the full resolution analysis with those obtained from Method 8. 3.0 METHO02 L_H 443 Fig. 12. Scatterplot comparing the LwN(443) obtained from Methods 1 and 2.

Original page 14

An Analysisof GACSamplingAlgorithms:A Case Study 3.0 2.2 04 :Z: 3 --J 1.5 ..... ," i ",, ,..,° ,'o i • .. • r." :."" . • ." . . - .2 " .',-:.',. ..: :. :";.:--, .: /. . ". • : % o "i':" , • .:",". "" • :" .....:. ,4+.: . L.. I",", O ,:'..'!i_ ",...t...... • -r" I--- .... • ",.'j W • ",,:5 :';.:" "° "_ i 0.8 _ • :.,t: ..; I% ,, I 1 . .. :.,- . - "..'" • .::.C... •:.:,,?!. °l,sI ; ! .2E-02 1.2E-02 0.1 METHOD2 LWN52e 1.5 2.2 3.( Fig. 13. Scatterplot comparing the Lw:(520) obtained from Methods 1 and 2. 3.0 L . . . 2.2 M-$ "7" ::m J 1.5 | " 0 • ",% • ° % - °,-• %" ., .o; o o**, ," d. :;%" -r- • ,.::::.-.,.:. .- ..:::L_. • LLI 'i:.':-: 0.8 . . ...:-,j" .. ,,," ":';.>., i :-:i,.-,?.,.,...- 1.2E-02 . 1.2E-02 0.8 METHOD2 LWH550 °° ., ,. ...... . . . 1.5 2.2 3.0 Fig. 14. Scatterplot comparing the Lwlv (550) obtained from Methods 1 and 2. 10

Original page 15

C.R.McClain,E. Yeh,andG. Fu i _ 2.5 1.9 o o L_ 1.2 • °,° :,.'. :...:: :.....:.... -'.:.. .. • , "' "k:': .... .' .,, :-'.'.. : , 7. " : 0 _ . • o ', . ,:. D-- ,.@' g_ * L'." ,'-" :,:¢- ,'. ,...-..., •.""•"'":"-"_ ,:L;'__-" : ' ¢', L :.. :1._ " :'.: ...... :"'. ,. ,... ,¢-.- ..., • : ..;- , . ; . . . .. : :2.,,; , . • 0.6 • ,.:,'.' 1.0E-02 1.0E-02 0.6 METHOD2 AEROSOL _ ".,.. ".,,...'.:. , .. ,.-* "-a . . :s..j t : •.:" ":.. :,..... : . ¢ .•..".: -o ... .. • ".'.. '.. 1.2 1.9 2•5 Fig. 15. Scatterplot comparing the La obtained from Methods 1 and 2. 9.8 7.5 I-- :z: M 5.0 • . ::E: F-- "I-- ....:;!:::::::., - ..., 2.5 " ::':".::.:'i" "J:" • -:," , • ..,_ ..;;:;: •.,,'" . .E.,...•"*='" ".; i" ". • " •. ::.-r._.. '.. • . • , . • 4.1E-02 2.5 METHOD2PIGMENT • o: .• °. .°• :, :" .: .... : • . . , . . . 5.0 7.5 9.8 Fig. 16. Scatterplot comparing the pigment values obtained from Methods 1 and 2. 11

Original page 16

An Analysisof GACSamplingAlgorithms:A Case Study 3.0 2.2 • ......: ...,-:..::.,:,.;.,." ..:.'.,'o:::,'.: •-'.-.Y-':""t,•• oL_,$2_ , " , O .: .s. .,, "° "°,w S, • . .., .-,..'.'.'[.. " • •• A"8 : .. ::_ .;,--:•;... "" :""'-'.'-,;:1,•S.. • ' '" "t.'.'V'''" " !..:""':"''"" • • ,:,•s ]• , ,.o °s " • " ". ,"':':";";','tJ.a_l_l_,_.s= -:'f"" .'' Z .. : .... ,..... ;:.. .. i. ". : s # ."#. .1 .......,•_ ;l_I, i':, ,....... _ ,s. ... .:o • . o • : . :.:::. :,:.| .:.,-:...-.r ;- O -r" • : .,,-'. ::::'.,::'.... : i--- LID T- • . "'." ;'.."'=':*,l.."r" • .. ...,,,,.:,.• .. 0,8 "# ..:,i'.'::',.""" 12E-e2 -;'; - !.2E-02 0.8 METHOD 2 LWH 443 1.5 2.2 3•0 Fig• 17. Scatterplot comparing the LWN(443) obtained from Methods 3 and 2. 3.0 2.2 N :Z: 1.5 r,31 l==l o :.?,, -t,-. I--- LIJ , °, oo ,¢ ..,=- 0.8 • ";.',• :. • |° ° "• , o.•! 1 2E-e2 1 2E-02 0•8 METHOD 2 LWH 20 1 • • ° • . o- .. ° • ,' o .-..: :.-:.; ".., .'.,.,;,..,..,...,.'." ,';..:.%.': ..., • Yb: - T "" "°°v'.| " 1._ !.2 3.8 Fig. 18. Scatterplot comparing the Lw_(520) obtained from Methods 3 and 2, 12

Original page 17

C.R.McClain,E. Yeh,andG. Fu 3.0 2.2 1.5 .. • ....::-'= :.. . . !. 0 -r- . ,..."",.,: :... . ...:...: ;., , , .. L_ • ; , .," . • ,. ,g,::: 0.8 4 °°, ":"iF "lip"__ ° 1.2E-02 t 2E-02 0.8 METHOD2 LWN550 • ° " °%° IT , • ° = '" " .. • .:, i .S 2.2 3.1 Fig. 19. Scatterplot comparing the LwN(550)obtained from Methods 3 and 2. 2.5 i . . . 1.9 J o o ILl 1.2 •" :,...'7'i . . ° ,°" • ° °. .:..:i:::.::,..• 0 "- : *I' ......'.:..:,....: ,.. :, .... ; ....:': • .. -1-- • -,; .. "..k ,:,':,:...%,...::].j, • ,.,:, , • , .:. .,.': .... • .. • . ..:':.:',.1,'-,-ZP,, , • "," •.... o,. .. . " r- ILl •;..,;. _ • M' '"" h_',.::" :- ::'ir,;:':::: "Y. ,-. " ...,..1,: 2,- . . . 0.6 gr.;.::".r.T.,": .'..," :., ,.'..j ::.:..., F' 10E-02 10E-02 0.6 METHOD2 AEROSOL i.:;' .:,,.. L .... :." • ,.... ... • • ::':,".-,,' :: " .. .. •., .: 1.2 .9 2,5 Figs. 20. Scatterplot comparing the L obtained from Methods 3 and 2. 13

Original page 18

An Analysis of CAC Sampling Algorithms: A Case Study 9.8 -- 7.5 I-- ->- W -Jr- M 5.0 • •" ...... • ...;.,,:.: "'•° . .......... O • . ,, --...:" -r- D-- IJJ • . :.;..:-;. ;;'...!..: .'r" .. -:!.:.:.::'; : !" . . 2.5 • "" : ;'" |::1".:"|" , " ;.•s. ,. o,,o, , ••. ". • • . • 4.1E-02 _ . . 4.1E-02 2.5 METHOD2 PIGHEHT ., • | :oo , , , • • : .': . : .o • • .. • °; , ::,.:.: .: : 58 75 98 Fig. 21. Scatterplot comparing the pigment values obtained from Methods 3 and 2. 3.0 • ..:.. ; • ._ .. ',:,.:,:.::.:,,I..,,,.,..v,t.-:'.;.;.• • • .i :, .. ,:.... . L':-:. • • ." , -.. "; o ", _ ' , , o:: -- q,.- - ; " ".:"" :'., .:.;¢,.."(l, •"., :....._ / ,...% _ ..:.o..I- ..: ...,.,:..:'..:::: ....,.,, .....,i °, o • • • . . , .,'$. _ '%:-. - . o,, I"o "o,,12 . a_,".e . ! 4 " A - ° • • • ."..... , ,..:" "..:.,,'x "__'"""'" : • ....• ,_ _ ......._I,%" t-.* i ,...:,,..__,'?'.di . ""° • " "" " t.." • :':...f • ..,......... ",dr" .s4 .,. e. -.', .. - ._ .... • . ":" ,..,..::":,.','..:.•.,,:...,.,,-: '?-" ::3 • ". ,,..:":.,:.,.,....;.$".':'""__ i,"." J • :. ' :':;'":.'.:L¢ I ._ " . ...,....::w_,r,_.. oO.• • ; •. ¢31 <:> • q -rl-laJ o • "o. o • $ ." ' . • • . o... ,Jr- . . .. .::.;.:.:::, 0.8 ¢:.° :'i • "- ! i • o,_ _ o° ° • *0_ ° 1.2E-02 1.2E'02 0.8 HETHOI)2LWH443 .o 1.5 2.2 3.0 Fig. 22. Seatterplot comparing the Lw_(443) obtained from Methods 4 and 2. 14

Original page 19

C.R. McClain, E. Yeh, and G. 3.0 2.2 -- , -..: ............. " , "° • ,? , : ; • . . ,'.', • ';.*, • " *_ • """ ".1' "''. : • .. o . . . ". :.. ....':...' -...':: • - ....-' ".: • , ...:.z...,.." ;:. ¢',,.1 • • ........ :.?.:'..' '..:i:.-. '/ '. • .,...:.. ,. ,' g,,.,.'" ".. z .s_ °. -..'o • Olo_ "' .. _ I .J •"' .'L4"":-'.'c';'-' , J'.<.:'":.. •". ,'.. .,,."...-2.:,,.-.:..'.7-,; ",. ". 1.5 ".. ," :. ._. ., _.,m, ,,:. _,'m ' o :-" '.','-":'":'"'"''"'.'. a.",.:-"i":-...: "-r" • ,..;..' :'.. l-..- Lid 0.B • • -.".'." : .7 i.2E-02 i.2E-e2 0.8 METHOD2 LWH520 ,.:... • ,, . . 1.5 2.2 3.E Fig. 23. Scatterplot comparing the Lw(520) obtained from Methods 4 and 2. 3.0 2.2 ..II 1.5 o •.. . •"- -rl-m,m • .. - _ ql. , " ,.: .,. . . : . ". -",,:,.-..,:" ,8 . • .i-..,.i.__.--.;. " l_m •. .,,.'a_i,, .. ..: .:... 1.2E-02 _".l 1.2E-02 0.8 METHOD2 LWH 550 l _ ;..: .. • ", ". .:. . 1.5 2.2 3.0 Fig. 24. Scatterpiot comparing the LWN(550) obtained from Methods 4 and 2. 15

Original page 20

An Analysis of GAC Sampling 2.5 1.9 o o L_ 0 Im METHOD_ AEROSOL Algorithms: A Case Study ii :i .9 2.5 Fig. 2S. Scatterplot comparing the L_ obtained from Methods 4 and 2. 9.8 7.5 I--- Z klJ 12_ 5.0 .....• • , ,-,, -- .,° ..... , i i i • , • • o. • ° °. : ..::" ." -- .. • , °., • . . •., • ,, ': :::, : :',...'', * °. . , ........ :. iii . ,." ;.| , ;o-|,,,, . ...:.;,, T- • |,.:. ; ,, .':'.:'i.,;;'i'.;:':i:'.:.,'1: ' "" , 2.5 ": "'" " ; - F:,:"i :.' .... . : -':. • ;..-'::: . . : :'. .,: .. •.',.:"', ;':'.,.,:!!"' :!'..!;. '" " " :" ":i 4.1E-02 _1 4.1E-02 2.5 METHOD2 PIGMENT 5.0 7:' 9.8 Fig. 26. Scatterplot comparing the pigment values obtained from Methods 4 and 2. 16

Original page 21

C.R.MeClain,E. Yeh,andC. Fh 3.g • . 2.2 • - . '.; - --:. -.....---- _ .rr*,-.--M'.;= • ... ".,'.'. ,";" • ;'P- " ." ., ".: -'1:, mrw.-'. • .'...,,. ,:-;,,,.,-.e..-., ." . :..-.:"'-"------4".14.,,", • . ... ..,, .... ...,;,....,t.,-j.. • ,; ;-:.| .... %'$ , ;.: ,, t . , .... ".,,, ...';..." ,." . ": . • -..-::..;,,.-'.'i';.;:.::" .. .' ''J",.i_l . ." ,-,.,,,:,...*J¢.I::,>'...:.• • • ,.,:.:,,. ,:: ,.l::."-;IL:I;,!,:-i_C.,,=.,.,..!':'": "" z ..,:..,E,,"T,,:.,igGrope",.• --- • " " "" "". :T_.'",i::',f,,i-:.;. ,-.:..-..:':''-:"": . .. ... ..- 1.5 . . ...-:.'.._, '. o '-i- !.-- 1.1.11 • • ..' , • . .I_:# ,.e. t:J."i'"."' 1.2E-O2 g.8 METHOD 2 LWN 443 '.,:.". 1.5 2.2 3.g Fig. 27. Scatterplot comparing the LWN(443) obtained from Methods 6 and 2. 3.0 I | | | 2.2 • • ; , .'. .! • .. :. ,:.. : • : z , • :;.;.',: ..1 , -" 1.5 %o ! .z,,,,.;: .:::.,:,'....," ,::::Nge.!:,/. ,%','... , • o ], r.,-;.. "-r .. :.-.,.._ • .'_ .4 ;.;.,¢ • W ...... .#. ,;'.*,,.",;. 0.8 ::.::.....'_ •. ..:'":'-': i : .._ .::',:i<?: 1.2E-O2 1.2E-g2 0.8 METHOD2 LWH529 • "° -° | '' • .° • • =, • . . , 11° .. .o, • • ,° •...'.. ,,,'...'." I %, ".!:,.::.'' .. .'d .'..;,' °; :.:....,:.. ,.,,.,'.:":' ;o ; • • " . . . 2.2 3.g Fig. 28. Scatterplot comparing the LwN(520) obtained from Methods 6 and 2. 17

Original page 22

An Analysis of GAC Sampling 3.0 2.2 "7" :=3 -.J 1.5 M:I °s • " o oq,¢° O "r" 'o "w " -.°% •:...," :;.': .. I'-- ILl .. "- :.'...'_ ia_,'.'.:ka'.. 0.8 ,. :" :,,. -:.:¥.- °- 1.2E-02 1.2 E-02 e.e METHOD2 LWN 550 Algorithms: A Case Study m . . . ° . . . °" : °: • , -. • 1.5 2.2 3.1 Fig. 29. Scatterplot comparing the LwN(550) obtained from Methods 6 and 2. 2.5 i.9 ..I 0 o .J 1.2 o • . .:-; .;.. $% • ,, •o• ":.:-- .'., . .',:'j__x,'.,. 0.6 ., .... WX_an," , , • , , :,," ,.'/ :. .. ,.T,::';... ,.:.,:: ,;ir_,o,,..,. ". ,.: •.-I.L.,.:...*.... '**, o. " . * S ., '% • 1.0E-02 1.0 E- 02 0.6 METHOD2 AEROSOL o, i:" o, • ,..-... -. . • ...:.,. , :.: ::,: . ,¢! s_... :... : bo_," *%-v ,. • 41,"...,:.....J, :,:. ¢" . ,, ° %$ • 1.2 1.9 2.5 Fig. 30, Scatterplot comparing the L_ obtained from Methods 6 and 2. 18

Original page 23

C.R.McClain,E. Yeh,andG. Fu 9.8 7.5 I-- -j,- IJ.I B'--I ICL. 5.0 ¢=1 .'. ".';o'..: -r- : F-m . • ,° .... .: .. : . . .":: .:: ILl -- . . .. • : • • :- : . • ..:.-., ..... :-., ,| .;; ..... • • ....'m:":.:C 2.5 ...- .:...j:..,!':.'" • A " : " %;"1" " " i • : " 4.1E-02 4.1E-02 2.5 METH092 PIGHEHT ... • :. • . 5.0 7.5 9.8 Fig. 31. Scatterplot comparing the pigment values obtained from Methods 6 and 2. 9.8 !" 7.5 | Z LIJ 22 eJ I---I Q. 5.0 r,- .'..': :: :.:" ": ,:::::': .... ..... :;: ,: .:. .. . "i .:::.:"F'::':'," z : ::':.:i;V •,- "" .:---;:;:i;:.';.. :: : i,i 2.5 •. ..':.:..:!:::.. ,," . ":'. ".':'.:';:::i-:,." • .,_ .. :..: • .'. f . :.::: 4.1E-02 4.1E-02 2._ • ° . . ° | | , , • ° ° i " • , ° • , , •°° , ° • °° • ,i oo °_ ;.:;." ." . • - 5.0 7,5 9.8 METHOD 2 P I GHEHT Fig. 32. Scatterplot of pigment values from Methods 7 and 2. 19

Original page 24

An Analysisof GACSamplingAlgorithms:A CaseStudy 9,B i ..... , . I-- 7.5 W • ¢, ¢0 iL , .... ::: .... ...,:;::ii :". ". ' "r" ",. • -" '::.:::::L::,:;.. .' •: •,';I .-I ..... I-laJ 2.5 ,,'.. '".!.';;.'.;;i':,." .'" • .' ". •. :...:.::::...... 4.1E-e2 4.1E-02 2.5 HETHO Fig. 33. Scatterplot of pigment Scatterplots of the data products from Methods 1, 3, 4, 6, 7, and 8 versus Method 2 are shown in Figs. 12-16, 17-21, 22-26, 27-31, 32, and 33, respectively. Only pigment comparisons are shown for Methods 7 and 8 because both are derived from Method 1 products. Method 2 was • " • "- 5.0 7.5 9.8 2 PIGHEHT values from Methods 8 and 2. correlation coefficients for these cases are meaningless. 4. DISCUSSION As indicated in the histograms, Table 3, and the scatused as the baseline for evaluation because it most closely terplots, the average-value and least-value techniques tend represents the the full resolution image as shown in the to overestimate the normalized radiances with the leasthistograms. The SEAPAK program SCATT was used to value value methods performing the worst. Additionally, generate the scatterplots. SCATT excludes any pixel pair the relative increases in normalized water radiances are that includes a value outside the range of valid values. All highest at 520 nm and lowest at 550nm. However, as inmethods tend to yield high values of LWN(443), especially dicated in the pigment histograms, this does not necessar- Methods 4 and 6, the least-value methods. For LWN(520), ily hold for all water masses in the scene. The tendency the bias towards high values is especially pronounced with in the pigment range above 1.5 mg m -s is to overesti- Method 4. As expected, Methods 4 and 6 strongly bias mate the concentrations and requires the relative increase L,(670) towards low values. Finally, the pigment scat- in Lw(550) be greater than in Lw(520). As expected, terplots do not indicate any pronounced biases, which are the least-value methods bias the aerosol radiances towards better illustrated in the frequency distribution plots and lower values. On the other hand, the average-value methquantified in Table 3. ods bias the aerosol radiances toward high values. Also, The last analysis performed used the SEAPAK pro- the least-value methods underestimate the mean pigment gram CORCO to determine the image's first and second concentrations. statistical moments and the correlation statistics between In summary, the fixed pixel subsampling gives the best Methods I, 3, 4, and 6 versus Method 2. The statistics representation of the full resolution data for GAC product are presented in Table 3. The statistics for full resolution generation• The explanation for why the water-radiances products (no subsampling; labeled "Full" in Table 3) and are biased in one direction, or the other, depending on Method 2 products were computed using the same image water mass, is more involved and is beyond the scope of for the two image inputs required by CORCO, so cross- this analysis. 20

Original page 25

C.R. McClain, E. Yeh, and G. Fu CLOSSARY AVHRR Advanced Very High Resolution Radiometer CZCS Coastal Zone Color Scanner GAC Global Area Coverage LAC Local Area Coverage NOAA National Oceanic and Atmospheric Administration SEAPAK Software package developed at NASA/Goddard Space Flight Center which ingests,displays,and processes data from the CZCS SeaWiFS Sea-viewing Wide Field-of-view Sensor TOMS Total Ozone Mapping Spectrometer REFERENCES Denman, K.L. and M.R. Abbott, 1988: Time evolution of surface chlorophyll patterns from cross spectrum analysis of satellite color images, J. Geophys. Res., 93) 6,789-6,798. Gordon, H.R., D.K. Clark, J.W. Brown, O.B. Brown, R.H. Evans, and W.W. Broenkow, 1983: Phytoplankton pigment concentrations in the Middle Atlantic Bight: Comparison of ship determinations and CZCS estimates, AppL Opt., 22, 20-36. , J.W. Brown, and R.H. Evans, 1988: Exact Rayleigh scattering calculations for use with the Nimbus-7 Coastal Zone Color Scanner, Appl. Opt., 27) 862-871. Justice, J.O., B.L. Markham, J.R.G. Townshend, and R.L. Kennard, 1989: Spatial degradation of satellitedata, Int. J. Remote Sensing, 10) 1,539-1,561. McClain, C.R., M. Darzi, J. Firestone, E. Yeh, G. Fu, and D. Endres, 1991a: SEAPAK Users Guide, Version 2.0,Vol. I-- System Description, NASA/Goddard Space Flight Center, NASA Tech. Memo. 100728, 158pp. , M. Darzi, J. Firestone, E. Yeh, G. Fu, and D. Endres, 1991b: SEAPAK Users Guide, Version 2.0, Vol. II--Descriptions of Programs NASA/Goddard Space Flight Center, NASA Tech. Memo. 100728, 586 pp. , W.E. Esaias, W. Barnes, B. Guenther, D. Endres, S.B. Hooker, G. Mitchell, and R. Barnes, 1992: Calibration and Validation Plan for SeaWiFS, NASA Tech. Memo. 10_566, Vol. 3, S.B. Hooker and E.R. Firestone, Eds., 41 pp. Morel, A., and L. Prieur, 1977: Analysis of variations in ocean color) Limnol. Oceanogr., 22, 709-722. Mueller, J.L., 1988:Nimbus-7 CZCS: Electronic overshoot due to cloud reflectance, Appl. Opt., 27) 438-440. Muller-Karger, F.E., C.R. McClain, R.N. Sambrotto, and G.C. Ray, 1990: A comparison of ship and CZCS-mapped distributious of phytoplankton in the Southeastern Bering Sea, J. Geophys. Res., 95) 483-499. Smith, R.C., and W.H. Wilson, 1981: Ship and satellite biooptical research in the California Bight, Oceanography from Space, J.F.R. Gower, Ed., Plenum Press, 281-294. 21

Original page 26

Form Approved REPORT DOCUMENTATION PAGE OM8No.070.0I Publicreporltngburdenforthiscollectionofinformationisestimatedtoaverage1hourperresponse,Includingthetimeforreviewinginstruclions,searchingexistingdatasources,gathering andmaintainingthedataneeded,andcompletingandreviewingthecolleclionofinformation.Sendcommentsregardingthisburdenestimateoranyotheraspectofthiscollectionof information,Includingsuggestionsforreducingthisburden,to WashingtonHeadquartersServices,DirectorateforInformationOperationsandReports,1215JeffersonDavisHighway,Sulle 1204, .Adin_lton,VA222024302,andto theOfficeof ManagementandBudget.PaperworkReductionProject(0704.O188),Washington,DC 20503. 1. AGENCY USE ONLY (Leave blank) 2. REPORT DATE November 1992 4. TITLE AND SUBTITLE SeaWIFS Technical Report Series Volume 4, An Analysis of GAC Sampling Algorithms: A Case Study 6. AUTHOR(S) Charles R. McClain, Eueng-nan Yeh, and Gary Fu Series Editors: Stanford B. Hooker and Elaine R. Firestone 7, PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) Laboratory for Hydrospheric Processes Goddard Space Flight Center Greenbelt, Maryland 20771 3. REPORT TYPE AND DATES COVERED Technical Memorandum 5. FUNDING NUMBERS 970.2 8. PERFORMING ORGANIZATION REPORT NUMBER 93B00016 AND ADDRESS(ES) 10. SPONSORING/MONITORING 9. SPONSORING/MONITORING AGENCY NAME(S) National Aeronautics and Space Administration Washington, D.C. 20546-0001 11. SUPPLEMENTARY NOTES E. Yeh, G. Fu, and E. Firestone: General Sciences Corporation, 12a, DISTRIBUTION/AVA!LABlUTY STATEMENT Unclassified - Unlimited Subject Category 48 13. ABSTRACT (Maximum 200 words) AGENCY REPORT NUMBER TM-104566, Vol. 4 Laurel, Maryland. 12b. DISTRIBUTION CODE The Sea-viewing Wide Field-of-View Sensor (SeaWiFS) instrument will sample at approximately a l-kin resolution at nadir, which will be broadcast for reception by realtime ground stations. However, the global data set will be comprised of coarser, 4-km data, which will be recorded and broadcast to the SeaWiFS Project for processing. Several algorithms for degrading the 1-km data to 4-kin data are examined using imagery from the Coastal Zone Color Scanner (CZCS) in an effort to determine which algorithm would best preserve the statistical characteristics of the derived products generated from the 1-km data. Of the algorithms tested, subsampling based on a fixed pixel within a 4x4 pixel array is judged to yield the most consistent results when compared to the 1-kin data products. i4. SUBJECT TERMS 15. NUMBER OF PAGES 20 Oceanography, SeaWiFS, Algorithms, GAC, SEAPAK, Pigment Concentration 17. SECURITY CLASSIFICATION 118.SECURITY CLASSIFICATION OF REPORT OF THIS PAGE Unclassified Unclassified ,r NSN 7540-01-280-5500 16. PRICE CODE 19. SECURITY CLASSIFICATION 20. LIMITATION OF ABsI HACT OF ABSTRACT Unclassified Unlimited Standard Form 298 (Rev, 2-89) PrescribedbyANSISV:l.230-18,2N-102