USF Libraries
USF Digital Collections

Autonomous optical measurements in Bayboro Harbor (Saint Petersburg, Florida)

MISSING IMAGE

Material Information

Title:
Autonomous optical measurements in Bayboro Harbor (Saint Petersburg, Florida)
Physical Description:
Book
Language:
English
Creator:
Du, Chunzi
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla.
Publication Date:

Subjects

Subjects / Keywords:
Backscatter
Light absorption
Phytoplankton
Remote sensing
Dissertations, Academic -- Marine Science -- Masters -- USF   ( lcsh )
Genre:
government publication (state, provincial, terriorial, dependent)   ( marcgt )
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )

Notes

Summary:
ABSTRACT: Estimating with precision coastal marine properties such as primary production, particulate and dissolved carbon, and red tide concentrations is a challenging but important part of marine research. It benefits not only the local communities, but also provides an important input to various global biogeochemical modeling efforts. Due to the complexity of coastal environments resulting from temporal variability of tidal and riverine influences, it is useful to develop and deploy an automated sensor network that provides real-time feedback. It can be used to validate remote sensing models to retrieve in-water constituents, and provide calibration and validation for atmospheric correction of satellite sensors. For turbid waters, satellite observations in the infrared part of the spectrum can not be used to estimate atmospheric aerosol concentration because the water is not black as is found for clearer waters.This research contribution introduces a modeling effort for a turbid coastal harbor area using a semi-analytical hyperspectral remote sensing algorithm for Case 2 waters to process data from the Autonomous Marine Optical System (AMOS). Retrieved results are then compared with field sample measurements showing satisfactory closure between measurements and theory. A time series of AMOS data over a one-month time span is examined, revealing significant variations in biological activity. A sensitivity analysis of the model is performed to expose the limitations and possible improvements to AMOS measurements in the future.
Thesis:
Thesis (M.S.)--University of South Florida, 2005.
Bibliography:
Includes bibliographical references.
System Details:
System requirements: World Wide Web browser and PDF reader.
System Details:
Mode of access: World Wide Web.
Statement of Responsibility:
by Chunzi Du.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 68 pages.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 001709507
oclc - 68621542
usfldc doi - E14-SFE0001384
usfldc handle - e14.1384
System ID:
SFS0025704:00001


This item is only available as the following downloads:


Full Text
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001709507
003 fts
005 20060614112133.0
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 060510s2005 flua sbm s000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0001384
035
(OCoLC)68621542
SFE0001384
040
FHM
c FHM
049
FHMM
090
GC11.2 (Online)
1 100
Du, Chunzi.
0 245
Autonomous optical measurements in Bayboro Harbor (Saint Petersburg, Florida)
h [electronic resource] /
by Chunzi Du.
260
[Tampa, Fla.] :
b University of South Florida,
2005.
502
Thesis (M.S.)--University of South Florida, 2005.
504
Includes bibliographical references.
516
Text (Electronic thesis) in PDF format.
538
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
500
Title from PDF of title page.
Document formatted into pages; contains 68 pages.
520
ABSTRACT: Estimating with precision coastal marine properties such as primary production, particulate and dissolved carbon, and red tide concentrations is a challenging but important part of marine research. It benefits not only the local communities, but also provides an important input to various global biogeochemical modeling efforts. Due to the complexity of coastal environments resulting from temporal variability of tidal and riverine influences, it is useful to develop and deploy an automated sensor network that provides real-time feedback. It can be used to validate remote sensing models to retrieve in-water constituents, and provide calibration and validation for atmospheric correction of satellite sensors. For turbid waters, satellite observations in the infrared part of the spectrum can not be used to estimate atmospheric aerosol concentration because the water is not black as is found for clearer waters.This research contribution introduces a modeling effort for a turbid coastal harbor area using a semi-analytical hyperspectral remote sensing algorithm for Case 2 waters to process data from the Autonomous Marine Optical System (AMOS). Retrieved results are then compared with field sample measurements showing satisfactory closure between measurements and theory. A time series of AMOS data over a one-month time span is examined, revealing significant variations in biological activity. A sensitivity analysis of the model is performed to expose the limitations and possible improvements to AMOS measurements in the future.
590
Adviser: Dr. Kendall L. Carder.
653
Backscatter.
Light absorption.
Phytoplankton.
Remote sensing.
690
Dissertations, Academic
z USF
x Marine Science
Masters.
773
t USF Electronic Theses and Dissertations.
4 856
u http://digital.lib.usf.edu/?e14.1384



PAGE 1

Autonomous Optical Measurements in Ba yboro Harbor (Saint Petersburg, Florida) by Chunzi Du A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science College of Marine Science University of South Florida Major Professor: Kendall L. Carder, Ph.D. Gabriel A. Vargo, Ph.D. Edward Van Vleet, Ph.D. Date of Approval: November 15, 2005 Keywords: Backscatter, Light Abso rption, Phytoplankton, Remote Sensing Copyright 2005, Chunzi Du

PAGE 2

ACKNOWLEDGEMENTS There are many people I would like to th ank who have supported me greatly in one way or another during my study in the College of Marine Science. First and foremost, I would like to thank my advisor Dr. Kendall Carder for sharing with me his passion for ocean optics, providing me the oppo rtunity to learn my own interests, and patiently offering encouragement. I would also like to thank my committee members, Dr. Gabriel Vargo and Dr. Edward Van Vleet, for their support and guidance. Their classes were instrumental in providing me with the biological and chemical framework necessary to interpret my optical findings. My husba nd and sons, I thank for providing balance to my life outside of the lab, so that I could put things into perspective. My fellow lab mates were also very supportive and deserve many thanks. My friend Jennifer Cannizzaro has helped me in countless ways. She taught me how to operate the in situ optical sensors, helped with data collection and processing and supplied helpful editorial comments. David E nglish kindly helped me with data sampling and instruments. Jim Ivey helped me with programming issues. Financial support for this work wa s provided by NASA (NAS5-31716) and ONR (N00014-97-1-0006 and N00014-96-1-5013) funding.

PAGE 3

i TABLE OF CONTENTS LIST OF TABLES.............................................................................................................. ii LIST OF FIGURES........................................................................................................... iii ABSTRACT....................................................................................................................... vi 1. INTRODUCTION ........................................................................................................1 1.1 Background.......................................................................................................1 1.2 Objectives and Approach..................................................................................5 2. DATA AND METHODS..............................................................................................7 2.1 Study area.........................................................................................................7 2.2 AMOS...............................................................................................................8 2.3 Remote sensing reflectance R rs ( )..................................................................10 2.3.1 AMOS radiometer............................................................................10 2.3.2 Spectrix radiometer..........................................................................12 2.4 Absorption......................................................................................................12 2.5 Backscattering.................................................................................................14 3. THEORY.....................................................................................................................15 4. RESULTS....................................................................................................................21 4.1 Chlorophyll a concentration...........................................................................21 4.2 Semi-analytic R rs ( ) model ............................................................................22 4.2.1 Original model parameters...............................................................23 4.2.2 Modified model parameters for Bayboro Harbor.............................26 4.2.3 Sensitivity analysis ..........................................................................28 4.3 Validation of AMOS R rs ( ) data....................................................................31 4.4 AMOS Time Series Analysis..........................................................................36 5. DISCUSSION..............................................................................................................46 6. CONCLUSIONS..........................................................................................................50 LIST OF REFERENCES...................................................................................................52

PAGE 4

ii LIST OF TABLES Table 1 Symbol definitions.......................................................................................9 Table 2 AMOS sampling schedule.........................................................................10 Table 3 Statistical results obtained comparing measured versus modeled absorption and backscattering co efficients for Bayboro Harbor (5/2004 ~ 7/2005). Model values were retrieved using the Lee et al. [1998, 1999] optimization technique with original and newly improved model parameters applied to Spectrix R rs ( ) data. Type 2 linear regression and RMSE valu es were calculated from logtransformed data ........................................................................................25 Table 4 Sensitivity test regression results. (a g slope 0.014,0.017,0.020)................29 Table 5 Sensitivity test regression results. (A 0 A 1 _WFS vs A 0 A 1 _Bayboro) .........31 Table 6 Regression results between measured R rs ( ) by direct Spectrix versus AMOS R rs ( ) at 440, 570 and 640nm wavelengths........................33 Table 7 Regression results be tween AMOS and Spectrix R rs ( ) modeled and measured values of a ph (440), a g (440), b bp (555). Error estimates RMSE log10 are consistent with those of Carder et al. (2004), although the other statistics are worse. .....................................................39

PAGE 5

iii LIST OF FIGURES Figure 1. AMOS sampling location in Ba yboro Harbor (St. Petersburg, FL)............7 Figure 2. Left: AMOS above-water un it with extending radiometer, solar panel, rechargeable battery pack; Right: AMOS underwater unit with fluorometer, transmi ssometers and HydroScat-2 Backscattering Sensor ...............................................................................10 Figure 3. Instruments for experime nts: a) Spectrix, a 512-channel spectroradiometer; b) Turner 10-AU-005 fluorometer; c) PerkinElmer Lambda 18 spectrophotometer........................................................14 Figure 4. Examples of phytoplankton, detrital and gelbstoff absorption spectra and the absorption spectra due to pure water [ Pope and Fry, 1997]..................................................................................................17 Figure 5. Example of particulate backscattering spectra and the backscattering spectra due to pure water [ Morel, 1974]............................18 Figure 6. Distribution of chlorophyll a concentrations obs erved during this study period (May 2004 to July 2005) at Bayboro Harbor (Saint Petersburg, Florida)....................................................................................21 Figure 7. Remote-sensing reflectance spectral measurements collected during the study period (May 2004 to July 2005) from Bayboro Harbor. Measurements were obtained using a 512-channel spectral radiometer (Spectrix).................................................................................22 Figure 8. Selected modeled R rs ( ) curves derived by the original Lee et al. [1999] optimization model parameters compared to directly measured Spectrix R rs ( ) curves................................................................23 Figure 9. Optimization-derived a) a g (440), b) a ph (440), c) a total (440), d) b bp (555) values obtained from Spectrix R rs ( ) data compared to measured values. Original model parameters [ Lee et al., 1999] were used. One-to-one lines (das h line) are shown along with type 2 linear regression functions (t hick solid) cal culated on logtransformed data.........................................................................................24

PAGE 6

iv Figure 10. One example of phyt oplankton absorption spectra, a ph ( ). Thick solid line is a ph ( ) measured in Bayboro Harbor, dash line is the modeled a ph ( ) derived using the original parameters [ Lee et al., 1998], dots line is the modeled a ph ( ) derived using the modified A 0 A 1 parameters for Bayboro Harbor......................................................26 Figure 11. Optimization-derived a) a g (440, b) a ph (440), c) a total (440), d) b bp (555) values obtained from Spectrix R rs ( ) data compared to measured values. Model parameters optimized for Bayboro Harbor were used. One-to-one lines (das h line) are shown along with type 2 linear regression functions (t hick solid) cal culated on logtransformed data.........................................................................................27 Figure 12. Sensitivity test results showing effects of changing a g slopes on model outputs for a) a g (440) and b) a ph (440). Gelbstoff slopes examined were 0.014, 0.017 and 0.020 nm -1 One-to-one lines are shown.........................................................................................................29 Figure 13. Sensitivity test results showing effects of changing A 0 ( ) and A 1 ( ) parameters on model outputs for a) a g (440) and b) a ph (440). Phytoplankton absorption parameters from Lee et al. [1998] for the West Florida Shelf (WFS) and fr om Bayboro Harbor data (AMOS) collected during this study are compared. One-to-one lines are shown.........................................................................................................30 Figure 14. Remote-sensing reflectance spectra from the AMOS sensor, May 2004 (hourly between 15:00 and 19:00 GMT). Measurements were collected in Bayboro Harbor (St. Petersburg, Florida)..............................32 Figure 15. Comparisons between Sp ectrix and AMOS remote-sensing reflectance spectra measured in Bayboro Harbor (St. Petersburg, Florida) during May 2004..........................................................................33 Figure 16. AMOS versus Spectrix remote-sensing reflectance values at 440, 570 and 640nm. Measurements were collected from Bayboro Harbor (St. Petersburg, Florid a) in May 2004. Linear best-fit regression lines (solid) are shown along with a one-to-one line (dotted).......................................................................................................34 Figure 17. An example of remote-sensi ng reflectance spectra obtained by the AMOS and Spectrix sensors from Bayboro Harbor (St. Petersburg, Florida) on May 6, 2004. Excess blue light is removed from the AMOS R rs ( ) using an effective Rayle igh-like correction term incorporated into the Lee et al. [1999] optimization model.......................35

PAGE 7

v Figure 18. Corrected AMOS versus Sp ectrix remote-sensing reflectance values at 440, 570 and 640nm obtaine d from Bayboro Harbor (St. Petersburg, Florida) in May 2004. AMOS R rs ( ) data were corrected by incorporating an effective Rayleigh-like correction term into the Lee et al. [1999] optimization model. Linear best-fit regression lines (solid) are shown along with a one-to-one line (dotted).......................................................................................................36 Figure 19. AMOS R rs ( ) derived a) a g (440), b) a ph (440), c) b bp (555) values for Bayboro Harbor (St. Petersburg, FL) May 2004. Values were derived using the Lee et al. [ 1999] optimization model modified for Bayboro Harbor. The boxes are measured values................................37 Figure 20. Measured values compared to optimization model outputs. a) a g (440); b) a ph (440); c) b bp (555).........38 Figure 21. Measured chlorophyll a concentrations from Bayboro Harbor compare to: a) measured a ph (440); b) modeled a ph (440) Figure 22. Fluorescence line heights (FLH). Height above an imaginary line between 670 and 750nm. 1 =670nm, 2 =690nm, 3 =750nm ......40 Figure 23. Measured chlorophyll a concentrations from Bayboro Harbor compared to FLH.......................................................................................41 Figure 24. Chlorophyll a concentrations using FLH method (marked circles) derived by AMOS R rs ( ) data for May 2004 at Bayboro Harbor. Directly measured chlorophyll a concentrations are marked with squares........................................................................................................42 Figure 25. Meteorological parameters of May 2004 near study area. a) rainfall data (from National Weather Servic e at Saint Petersburg Station; b) hourly wind speed from buoy located on West Florida Shelf. Bar height represents wind speed range, with middle dots represent the average daily wind speed (b from NOAA CO-OPS website, for St. Petersburg, Florida location)................................................................43 Figure 26. AMOS underwater instrument s measurements from May 7 to May 21 at Bayboro Harbor: a) beam -c (660nm); b) uncalibrated chlorophyll fluorescence by fluorometer...................................................44

PAGE 8

vi Autonomous Optical Measur ements in Bayboro Harbor Chunzi Du ABSTRACT Estimating with precision coastal marine properties such as primary production, particulate and dissolved carbon, and red ti de concentrations is a challenging but important part of marine research. It bene fits not only the local communities, but also provides an important input to various globa l biogeochemical modeling efforts. Due to the complexity of coastal environments result ing from temporal vari ability of tidal and riverine influences, it is us eful to develop and deploy an automated sensor network that provides real-time feedback. It can be used to validate remote sensing models to retrieve in-water constituents, and provide calibration and validation for atmospheric correction of satellite sensors. For turbid waters, satelli te observations in the infrared part of the spectrum can not be used to estimate atmo spheric aerosol concentration because the water is not black as is f ound for clearer waters. This research contribution introduces a modeling effort for a turbid coastal harbor area using a semi-analytical hyperspectral remote sensing algorithm for Case 2 waters to process data from the Autonomous Marine Optical System (AMOS). Retrieved results are then compared with field sample measurements showing satisfactory closure between measurements and theory. A time series of AMOS data over a one-month time span is examined, revealing significant

PAGE 9

vii variations in biological activity. A sensitivity analysis of the model is performed to expose the limitations and possible improvement s to AMOS measurements in the future.

PAGE 10

1 1. INTRODUCTION 1.1 Background Global climate change is becoming an in creasingly discussed topic due to the huge impact it has on our daily lives. Although associations between global warming and regional climate patterns such as frequency a nd magnitude of hurricanes in the Atlantic Ocean have yet to be established, the need for better understanding of carbon cycles on global scales is evident. To support such ta sks, traditional field spot-type sampling of oceanic environment will not be sufficient. Observations from space with sensors onboard satellites or aircraft will provide th e only synoptic coverage with sufficient temporal and spatial resolution that can be us ed in analytical mode ls for predictions [e.g. Esaias et al. 1998] of global primary productivity and dissolved organic carbon fluxes from rivers. Less than 10% of the light measured by a satellite ocean color sensor originates from beneath the ocean surface. The majority of received light is due to atmospheric absorption and scattering. Consequently, accura te atmospheric correction is critical in remote sensing applications since a small mistake will result in large errors when estimating correct water leaving radiances. Th e performance of retrieval algorithms and the accuracy of derived quantities are strongl y influenced by atmospheric corrections. Atmospheric correction algorithms have to c ope with the reality in coastal waters that infrared wavebands treated as atmosphe ric only in open-ocean waters may contain

PAGE 11

2 a non-negligible and variable signal from the sea as well. Classical atmospheric correction schemes assume that the water-l eaving radiance is zero in the near-infrared part of the spectrum [ Gordon and Wang 1994]. However, recent experiences with spaceborne data (e.g. SeaWiFS and MODIS) and ship-based optical measurements, clearly indicate that this assumption is not valid over turbid coastal waters [ Hu et al., 2000; Siegel et al. 2000]. The principle contri buting factor is high con centrations of scattering constituents that cause the water-leaving signa l in the near-infrared part of the spectrum (>700nm) to be significantly greater than zero (i.e. not black). Therefore, it is highly desirable to have the ability to provide ground truth for atmos pheric correction of satellite ocean color imagery. A network of aut onomous optical sensors that measure downwelling irradiance and water-leaving radiance just above the sea surface may provide the necessary ground truth data to im prove atmospheric corrections of coastal satellite ocean color data. Derivation of in-water optical proper ties (e.g. absorption, backscattering, and chlorophyll concentrations) from water-l eaving radiance data requires accurate processing algorithms. Using the spectral in formation from the light reflected from beneath the sea surface or th e water-leaving radiance (L w ( )), many in-water properties have been successfully retrieved empirically or analytically, incl uding diffuse attenuation coefficients [ Austin and Petzold 1981; Stumpf and Pennock 1991], chlorophyll concentrations [ Carder et al., 1999; Gordon et al., 1983; O'Reilly et al. 1998], mass concentrations of suspended sediments [ Bukata et al. 1991; Doerffer and Fisher, 1994], and bottom depths for waters shallower than ~30 m [ Lee et al. 1999, 2001]. These properties can provide important input assessing the status of the water environment

PAGE 12

3 [ Jerlov 1976], and help to better understand the oceanic photosynthetic process[ Kirk, 1994; Marra et al., 1992; Platt and Sathyendranath 1988], as well as heat transfer [ Lewis et al. 1990; Morel and Antoine 1994]. The initial success of the coastal zone color scanner (CZCS, 1978-1986) chlorophyll algorithm [ Gordon et al., 1983] profoundly enriched our knowledge of the global distribution of phytoplankton, especi ally in the open ocean environments [ Mitchell, 1994]. A better understanding of in-wat er optical properties later led to improved algorithms for a series of next-gener ation sensors, such as the widely-used Seaviewing Wide Field-of-View sensor (SeaWi FS, 1997-present) and Moderate-Resolution Imaging Spectrometer (MODIS). To provide better quantification models to interpret remotely sensed signals, two water types namely Case 1 and Case 2, were introduced by Morel and Prieur, [1977], and refined later by Gordon and Morel, [1983]. By definition, Case 1 waters are those waters in which phytoplankton is the principle agen t responsible for vari ations in optical properties of the water, while Case 2 waters are influenced not ju st by phytoplankton and related particles, but also by other substan ces that vary independently of phytoplankton (e.g. inorganic particles in suspended state and colored dissolved organic matter, CDOM, or gelbstoff). Case 1 waters are often found in the open ocean where influences from the land and seafloor are minimal. This type of water covers most of the oceanic environment (up to 90%). However, they are usually le ss productive compared to the Case 2 type coastal waters. Due to the fact that Case 2 waters are primarily characterized by several opticallyactive substances which vary independently of each other and in many cases, are

PAGE 13

4 accompanied by relatively high levels of sca ttering, algorithms developed for Case 1 waters can not be applied to Case 2 waters [ Carder et al., 1986, 1991]. Many commonlyused algorithms for Case 1 waters are ba sed on correlations between some simple function of ocean-color signals at tw o or three wavebands and chlorophyll a concentration [ Gordon et al., 1983; Morel and Prieur, 1977]. Different algorithms are required for Case 2 water types because ther e are more optical components influencing the measured spectra. Also, due to the overla pping of absorption and scattering spectra, variations in radiance or re flectance can not be related directly to any one component, and the contributions by individual constituents have to be derived simultaneously [ Neumann et al., 2000]. Lastly, in shallo w coastal regions and harbor areas with water depth less than 30m, bottom reflection effects may have to be included in the algorithms. Such complications imply no generic algorith ms for all Case 2 water types will work. Instead, individual models and algorithms may have to be developed to meet needs for specific regions. In recent years, hyperspectral remote sensing has gained much attention, especially in Case 2 water applications, a nd has revealed subtle information that was previously undiscovered. For example, sp ectral signatures of di fferent phytoplankton classes or species can be f ound from hyperspectral sensors [ Bidigare et al. 1989; Hoepffner and Sathyendranath 1993; Millie et al. 1997]. A semi-analytical (SA) model was developed by Lee et al. [1998] for hyperspe ctral remote sensing needs. Briefly, the Lee et al. [1998] SA model provides accurate results comparable to Monte Carlo simulation and Hydrolight appr oaches but with much less co mputational needs. In the mean time, the SA model allows quick i nversion of in-water constituents, including

PAGE 14

5 bottom effects [ Lee et al. 1999]. Also, because of the fact that the SA model is not strongly dependent on choices of scattering phase functions, it is best suited for use with Case 2 water types, especially in a very turbid environment like Bayboro Harbor (St. Petersburg, FL). 1.2 Objectives and Approach The west Florida shelf (WFS) has been se lected by many agencies as a study site for long-term monitoring by instrumented pl atforms and underwater vehicles, aircraft, spacecraft, and monthly ship surveys. It is a region where numerical models of circulation and phytoplankton dynamics are be ing developed. An array of automated, continuous sensors could provide investigators with a cont inuous record of optical conditions during rapidly changing events such as storms, plankton blooms, tidal flushing, and upwelling. They can also provi de boundary conditions for the bio-optical models being developed for predicting primary production and optical properties for the WFS and provide optical data to calibrate and ex plain variations in satellite imagery. From an economics view point, it is rather expensive and unpr actical to deploy a research vessel at a fixed location for extended peri ods of time for monitoring purposes. From a research point of view, an autonomous a rray of sensors could greatly enhance our understanding of river blooms and the temporal variability of red tides [ Cannizzaro et al. 2002]. The Autonomous Marine Optical System (AMOS) was developed at the University of South Florida to measure hype rspectral remote-sensing reflectance spectra and water column measurements of down welling irradiance, b ackscattering, beam

PAGE 15

6 attenuation, and chlorophyll fluorescence. It was deployed in Ba yboro Harbor (Saint Petersburg, Florida) intermittently between May of 2004 and July 2005. The primary hypothesis of this research is that by carefully adjusting model parameters, the semi-analytical hyperspectra l remote-sensing model by Lee et al. [1999] can be used to successfully retrieve in-water optical properties (e.g. phytoplankton absorption (a ph ( )), gelbstoff absorption (a g ( )) and particle backscattering (b bp ( )) from above-water remote-sensing measurements in tu rbid coastal environments. It is expected that new model parameters will be needed for this very turbid type of water, and parameters derived from the west Florida shelf will not perform as well. R rs ( ) data collected using a 512-channel, ha nd-held radiometer (Spectrix) is used for this purpose. The secondary hypothesis is that AMOS provides accurate R rs ( ) data, comparable to that from Spectrix measurements. A validation an alysis is performed. Lastly, the improved R rs ( ) model modified to perform accurately in Bayboro Harbor is applied to validated AMOS R rs ( ) data, and a monthly time series is examined.

PAGE 16

2. DATA AND METHODS 2.1 Study area AMOS was developed w ith funding by the Defense University Research Instrumentation Program (DURIP) and was or iginally deployed near Port Manatee in Tampa Bay [ Steward and Carder, 2002]. For this study, it was placed in Bayboro Harbor (Saint Petersburg, Florida) on a piling located on the southeast corner of the USF College of Marine Science (CMS) (Fig.1). Bayboro Harb or comprises two connected basins with USF AMOS Figure 1. AMOS sampling location in Bayboro Harbor (St. Petersburg, FL). 7

PAGE 17

8 an average depth of 6 meters. It is bounded on three sides by developed shoreline. Both basins are connected to Tampa Bay through a narrow dredged shipping channel of 7.3m depth. These basins receive storm water runoff (largely from Salt Creek), and nearby facility discharges, such as from U. S. Coastal Guard St. Petersburg Station, Albert Whitted Municipal Airport, and City of St. Petersburgs Municiple Sewage Treatment Plant. The two basins are separated by an extr usion of land that cont ains buildings of the University of South Florida College of Marine Science (Fig. 1). The site was chosen in order to measure Case 2 waters in a loca tion easily accessible for routine maintenance and collection of validation data. 2.2 AMOS AMOS was built in 2000 by the Center for Ocean Technology (COT) of the University of South Florida (USF) as a pr ototype sampling device. At predetermined times, it make remote sensing reflectance (R rs ( )) measurements above the water surface as well as in-water measurements of optical properties at one or more depths. After sampling, it transmits the information back to a networked archival and processing station. Note that symbol definitions can be found in Table 1. The AMOS installation is composed of a power supply, a master controller, an above-water remote-sensing radiometer, and an in-water sub-controller for an underwater E d ( ) sensor and inherent optic al property (IOP) instrument s (Fig. 2). A solar panel recharges the battery power supply, so that neither power, nor communication cables are needed between AMOS and the shore. At sc heduled times throughout the day (Table 2) AMOS measures down-welling irradiance, upwel ling sky radiance, irradiance at depth,

PAGE 18

9 Table 1. Symbol definitions Symbols Description Units A Absorption coefficient (=a w +a ph +a d +a g ) m -1 a w Absorption coefficient of pure water m -1 a p Absorption coefficient of particulates (=a ph +a d ) m -1 a ph Absorption coefficient of phytoplankton m -1 a d Absorption coefficient of detritus m -1 a g Absorption coefficient of gelbstoff m -1 A Empirical shape coefficient for power law function (y=Ax B ) b b Backscattering coefficient m -1 b bw Backscattering coefficient of pure water m -1 b bp Backscattering coefficient of particulates m -1 b p Scattering coefficient of particles m -1 B Empirical slope coefficient for power law function (y=Ax B ) c Attenuation coefficient m -1 Chl Chlorophyll a concentration mg m -3 E d Downwelling irradiance W m -2 nm -1 f Water-to-air divergence factor FLH Fluorescence line height W m -2 m -1 sr -1 L G Radiance reflected from a 10% diffu se reflector or gray card L sky Downwelling sky radiance W m -2 nm -1 sr -1 L u Upwelling radiance W m -2 nm -1 sr -1 L w Water-leaving radiance W m -2 nm -1 sr -1 N Refractive index of seawater nL w Normalized water-leaving radiance W m -2 nm -1 sr -1 Q Upwelling irradiance-to-radiance ratio sr -1 R Fresnel reflectance R G Reflectance of a 10% diffuse reflector or gray card R rs Above surface remote-sensing reflectance sr -1 r rs Subsurface remote-sensing reflectance sr -1 S d Spectral slope for detrital absorption spectra m -1 S g Spectral slope for gelbstoff absorption spectra m -1 T Transmittance across the air-sea interface Angstrom exponent describing spectral shape of b bp ( ) chlorophyll fluorescence, attenuation of blue and re d light (470 and 660nm), backscattering of blue and red light (470 and 676nm), and the Global Positioning System (GPS) location and time of sampling. This info rmation is recorded on site and transmitted by radio to a computer at USF where it is arch ived and processed to customary scientific units.

PAGE 19

Figure 2. Left: AMOS above-water unit with extending radiometer, solar panel, and rechargeable battery pack; Right: AMOS underwater unit with fluorometer, transmissometers and back-scattering meters. Table 2. AMOS sampling schedule. 10 Sampling time (EST) No. samples 1:30~10:30 sampling every 3 hours 4 10:30~11:30 sampling every half hour 2 11:30~15:00 sampling every 15 min 14 15:30 sampling 1 16:30 sampling 1 19:30 sampling 1 23:30 sampling 1

PAGE 20

2.3 Remote-sensing reflectance, R rs ( ) 2.3.1 AMOS radiometer The automated measurement of R rs ( ) using a single radiom eter that looks at multiple optical pathways is an important feature of AMOS. A fiber-optic switch allows measurement of the light from a down-welling cosine collector, a sea-surface viewing window, a complementary-angle sky-viewing window, or a terminated light path (to measure dark current). This single spectrometer arrangement allows R rs ( ) spectral ratios to be made without distortions from sharp spec tral features such as Fraunhofer lines. The radiance windows are inclin ed so that the center view th rough these windows is 30 from the vertical, similar to the vi ewing angle that has been used with handheld spectrometers for several years [e.g. Carder and Steward, 1985]. Remote-sensing reflectance (R rs ( )) by AMOS is by definition )( )( ),( d w rsE L AMOSR (1) Light transmitted through the downwelling cosi ne collector provides the spectrometer a direct measurement of downwelling irradiance (E d ( )). This is followed by a spectrometric measure of the water upwelling radiance (L u ( )), and a measure of the sky downwelling radiance (L sky ( )). Combining these generates water leaving radiance (L w ( ), )( )( *022.0 )( )( )(*022.0)()( d sky d u rs rs rsE L E L S TR (2) 11

PAGE 21

12 The effect of skylight on this measured remote sensing reflectance is removed by a factor of 0.022, which is the contri bution of Fresnel reflectance for a 30 o viewing angle [ Mobley 1994]. 2.3.2 Spectrix radiometer A hand-held, 512-channel spectroradiomete r (Spectrix, 350~850nm) (Fig. 3a) was used to measure L u ( ) (upwelling radiance), L G ( ) (radiance reflected from a standard grey diffuse reflector) and L sky ( ) (sky radiance) at Bayboro Harbor intermittently from May 2004 to July 2005 (~1-2 times per w eek) when AMOS was deployed. These measurements were used to estimate E d ( ) and L w ( ). The ratio of L w ( ) and E d ( ) then provided the remote-sensing reflectance [ Lee et al. 1996]. 2.4 Absorption Absorption spectra due to particle s (phytoplankton and detritus), a p ( ) were determined using the quanti tative filter technique [ Kiefer and SooHoo, 1982; Yentsch, 1962]. Seawater samples collected by bucke t within 5 minutes of Spectrix radiance measurements were filtered through 2.5cm GF/F filters. The sample filter and a reference filter wetted with Milli Q wate r were placed on individual gl ass plates (diameter=2.4cm) in a custom-made diffuse transmissometer box. The transmittance of the sample filter, T sample ( ), and the reference filter, T reference ( ), were measured three times each using a custom-made, 512-channel spectroradiometer(~350-850nm).

PAGE 22

Optical densities, OD( ), were calculated as ) )( )( (log)(10 sample refT T OD (3) Particulate absorption spectra were calculated as L OD ap p *)(*3.2)( (4) where is the optical path elonga tion or beta factor, and L is the effective optical pathlength (the area of filter pad divided by volume seawater filtered) The beta factor is an empirical formulation defined as the rati o of optical to geometric pathlength that corrects for multiple scattering inside the filter. In this study, an average of two published beta factor formulations [ Bricaud and Stramski, 1990; Nelson and Robertson 1993] was chosen 5.0)(*6.00.1 pOD (5) Phytoplankton pigments were extracted from the sample filter with ~20-50ml of hot 100% methanol for 10-15 minutes in the dark [ Kishino et al. 1985; Roesler et al. 1989]. Fluorometric chlorophyll a nd pheopigment concentrations were determined using the filtrate using a Turner 10-AU-005 fluoromet er (Fig. 3b) according to the methods of Holm-Hansen et al. [1965]. Light transmission was meas ured again on this extracted filter and the same reference filter to obtain the absorption spect ra of detrital partic les and non-methanolextractable (e.g. water soluble) pigments, a d ( ). The absorption spectra for phytoplankton pigments, a ph ( ), is then calculated as )()()( d p phaaa (6) 13

PAGE 23

Gelbstoff absorption spectra, a g ( ), were measured using filtered seawater obtained using pre-rinsed 0.2um nylon membrane filters. Samples are scanned in 10-cm quartz cells from 200-800nm, using a Perkin-E lmer Lambda 18 spectrophotometer (Fig. 3c) and referenced to Milli Q water. Figure 3. Instruments for experiments: a) Sp ectrix: a 512-channel sp ectroradiometer; b) Turner 10-AU-005 fluorometer; c) Perkin-Elmer Lambda 18 spectrophotometer. 2.5 Backscattering In situ vertical profiles of to tal backscattering measured at 470 and 676nm using a HOBI Labs Hydroscat2 (HS2) were perfor med on four occasions in May 2004. Measurement, calibration, and data processing information for this instrument have been described previously [ Maffione and Dana 1997]. A spectral power function was fit to measured backscattering values at 470 and 676nm in order to obtai n the backscattering coefficient at 555nm. Particul ate backscattering at 555nm, b bp (555), was calculated from total backscattering by subtracting the backscattering coefficient due to pure water [ Morel 1974]. 14

PAGE 24

3. THEORY The semi-analytical (SA) model and optim ization approach of Lee et al. [1998, 1999] retrieves in-w ater properties (a ph ( ), a g ( ), b bp ( ), water depth (H), and bottom albedo ( ( )) from hyperspectral R rs ( ). A brief introduction of the SA model and optimization technique is described below. For optically deep, vertically homogeneous waters, R rs ( ) is dependent on the absorption and backscattering properties of seaw ater and the angular distribution of light within the ocean. Using ra diative transfer theory [ Gordon et al., 1988; Mobley, 1994], R rs ( ) can be expressed as ) (b) (a ) (b ) (Q f n t ) (Rb b 2 2 rs (7) where t is the transmittance across the air-sea interface, n is the i ndex of refraction of seawater, f is an empirical factor that is a function of the solar zenith angle, and Q( ) is the upwelling irradiance-to-radiance ratio, a( ) is the total absorption spectra, and b b ( ) is the total backscattering spectra. By making approximations for these latter terms [ Lee et al. 1998], R rs ( ) can further be related to the subsur face remote-sensing reflectance, r rs ( ), as follows: )) (r5.11( ) (r5.0 )(Rrs rs rs (8) 15

PAGE 25

In optically shallow waters, contributi ons from the bottom can be expressed separately from deep water effects in terms of sub-surface remote sensing reflectance as [ Lee et al. 1999] )9( 92.02.1 1 exp 1 92.02.1 1 exp1 H D H D rrB u C u dp rsrs where r rs dp is the subsurface remote-sensing reflectance for optically deep waters, D u C is the optical-path-elongation factor due to mu ltiple scattering for the water column, D u B is the optical-path-elongation factor fo r the bottom-reflected photons, and is equal to the sum of the absorption and backscattering coefficients. For optically deep waters subsurface remote-sensing reflectance is [ Lee et al. 2004] )10(b bp p b bw w dp rsba b g ba b gr where g w and g p are known model-derived paramete rs for molecular and particle scattering, respectively. Separate terms for particles and molecules are required because the angular distribution for molecula r backscattering due to water, b bw ( ), differs from that of particulate back scattering due to water. Optical path elongation factors for the water column and bottom are [ Lee et al. 1999] )11( 4.5104.1 4.2103.15.0 5.0u Dandu DB u C u respectively, where )12(b bba b u 16

PAGE 26

The absorption coefficient can be exam ined more thoroughly by decomposing it into the sum of its components: a ( ) = a w ( ) + a ph ( ) +a d ( )+ a g ( ) (13) where the subscripts w, ph, d, and g refer to water, phytoplankton, de tritus and gelbstoff, respectively (Fig. 4). Similarly, the back scattering coefficient can be expanded as ) (b) (b) (bbp bw b (14) where the subscripts w and p refer to wate r and particles (phytoplankton and detritus), respectively (Fig. 5). Absorption due to water, a w ( ), and backscattering due to water are constant and well known [ Morel 1974; Pope and Fry 1997]. Terms for chlorophyll and gelbstoff fluorescence and water-Raman scatteri ng are not included in this model. The water column is assumed to be homogeneous and the bottom a Lambertian reflector. Figure 4. Examples of phytoplankton, detrita l and gelbstoff absorption spectra and the absorption spectra due to pure water [ Pope and Fry, 1997]. 17

PAGE 27

Figure 5. Example of particulate backscattering spectra and the backscattering spectra due to pure water [ Morel, 1974]. Combining Eqs. (8-14) provi de a model for deriving a ph ( ), a g ( ), b bp ( ), ( ) and H from R rs ( ). These terms are parameterized be low in order to reduce the number of unknowns. Phytoplankton absorption sp ectra are modeled from a ph (440) as [ Lee 1994] )15( 440ln 4401 0ph ph phaAAaa where A 0 ( ) and A 1 ( ) are empirically derived constant s. This functi on ensures that a ph ( ) curvature changes a ppropriately with a ph (440), taking into consideration the natural variability observed in phytoplankton pigmentation and pigment packaging [ Bricaud et al. 1995]. Absorption spectra due to gelbstoff is modeled from a g (440) as [ Lee et al. 1999] 18

PAGE 28

)16( *)440()()440( S g geaa where S is the spectral slope calculated fo r log-transformed absorption values. Since gelbstoff and detritus both exhibit exponen tially decreasing absorp tion with increasing wavelength, they cannot be derive d independently. Therefore, a g ( ) and a d ( ) are combined and an averag e spectral slope (0.015nm -1 ) is used [ Carder et al., 1989, 1991]. Particle backscattering sp ectra are modeled from b bp (555) as )17( 555 555Y bp bpbb where the reference wavelength 555nm replaces the 400nm value originally used by Lee et al. [1999]. The spectral shape parameter fo r backscattering, Y, is estimated using an empirical relationship from measured R rs (443) and R rs (490) data and values are limited to the 0-2.5 range [ Lee et al. 1999]. Bottom albedo spectra are expressed as )18( *550550 normalized nm where (550) is the bottom albedo co efficient at 550nm, and 550nm-normalized ( ) is a bottom albedo spectrum normalized at 550nm for sand [ Lee et al. 1999]. Since R rs (750) for turbid coastal waters may not be zero [ Hu et al. 2000; Siegel et al. 2000], R rs in ( ) is defined as )19( meas rs in rsRR where R rs meas is the remote-sensing reflectance measured using either the AMOS or Spectrix radiometric sensors. The delta, factor is nonspectral (e.g. white) reflected 19

PAGE 29

20 light representing residual sunglint, cloud light, and sky light brought into AMOS by wave facets and not removed by Eq. 2. Values for a ph (440), a g (440), b bp (550), (550), H and are then derived iteratively using a predictor-corrector optimization scheme until the difference between R rs ( ) in and R rs ( ) mod. is minimized [ Lee et al. 1999]. Parameter input values provided to the model are independent of field measurements.

PAGE 30

4. RESULTS 4.1 Chlorophyll a concentration Chlorophyll a concentrations in Bayboro Ha rbor measured during the study period range between 2.48 and 47.74 mg m -3 with a mean value of 9.47 mg m -3 (Fig. 6). 0 3 6 9 12 15 1 10 100 [chl a](mg m-3)number of observation s Figure 6. Distribution of chlorophyll a concentrations observed during this study period (May 2004 to July 2005) at Bayboro Har bor (Saint Petersburg, Florida). This is in significant contra st to a recent West Florida shelf and Bahamas study where chlorophyll a concentrations ranged from 0.026 to 20.6 mg m -3 with a mean value of 0.66 mg m -3 [ Cannizzaro and Carder, 2005 (submitted)]. The higher mean chlorophyll concentrations observed in Ba yboro Harbor indicates that th is region is highly eutrophic indicating that perhaps a new set of model parameters for the SA model [ Lee et al. 1999] may be needed to adequately describe Bayboro Harbor. 21

PAGE 31

4.2 Semi-analytic R rs ( ) model Between May 2004 and July 2005, 45 remote -sensing reflectance spectra were collected using a Spectrix radiometer from Bayboro Harbor (Fig. 7). Maximal reflectance Figure 7. Remote-sensing reflectance spectra l measurements collected during the study period (May 2004 to July 2005) from Baybor o Harbor. Measurements were obtained using a 512-channel spectral radiometer (Spectrix). is typically observed around 570 nm which is why eutrophic harbor areas are usually greenish in color. A smaller peak ar ound 685 nm is due to chlorophyll fluorescence. The anomalous curve with peak refl ectance at ~700nm corresponds to a K.brevis bloom observed in July 2005 with a chlorophylla concentration of 71.9 mg m -3 Since it is an isolated case and presents very different opt ical characteristics fr om typical in-water constituents in the harbor water, it is not included in this modeling effort. The reflectance at the blue end (~400nm) is low due to the fact that chlorophyll and gelbstoff concentrations are high in the study area. A careful pa rtition of signals at 22

PAGE 32

this wavelength would help to better identify in-water constituents. At the longer wavelengths, especially beyond 700nm, reflect ance values are low due to significantly higher water absorption values (Fig. 4). 4.2.1. Original model parameters In order to determine how well the Lee et al. [1998, 1999] op timization technique works in Bayboro Harbor, the technique was first applied to Spectrix R rs ( ) data using the original model parameters derived from west Florida shelf data. Values for a ph (440), a g (440), b bp (555), (550), H and were derived by minimizing the differences between measured and modeled R rs ( ) data. Figure 8 shows a few examples of these measured and Figure 8. Selected modeled R rs ( ) curves derived by the original Lee et al.[1999] optimization model parameters compared to directly measured Spectrix R rs ( ) curves. modeled curves. It can be seen that they ma tch very well to each other, with the only 23

PAGE 33

exception between 660-740nm, where measured R rs ( ) are always higher than modeled. This is because chlorophyll fluores cence is not included in the SA R rs ( ) model. Relationships between measured a nd model-derived absorption and backscattering values are shown in Figure 9. Type 2 linear regression and root-meanslope=0.359 y_in=-0.193 R2=0.064 RMSlog=0.183 n=450.1 1 10 0.1 1 10 ag(440) meas.(1/m)ag(440) mod.(1/m)a slope=0.555 y_int = -0.193 R2 = 0.232 RMSlog=0.411 n=450.01 0.1 1 10 0.010.1 1 10 aph(440) meas.(1/m)aph(440) mod.(1/m)b slope=0.937 y_int = -0.125 R2 = 0.746 RMSlog=0.156 n=450.1 1 10 0.1 1 10 atot(440) meas.(1/m)atot(440) mod.(1/m)c slope = 3.197 y_int = 4.172 R2 = 0.417 RMSlog = 0.549 n=4 0.001 0.01 0.1 1 0.0010.010.1 1 bbp(555)_meas.(1/m)bbp(555)_mod.(1/m)d Figure 9. Optimization-derived a) a g (440, b) a ph (440), c) a total (440), d) b bp (555) values obtained from Spectrix R rs ( ) data compared to measured values. Original model parameters [ Lee et al., 1999] were used. One-to-one lines (dash line) are shown along with type 2 linear regressi on functions (thick solid) calculated on log-transformed data. square errors calculated on log-transforme d data are shown in Table 3. Total, 24

PAGE 34

25 Table 3. Statistical results obtained comparing measured ve rsus modeled absorption and backscattering coefficients for Bayboro Ha rbor (5/2004 ~ 7/2005). Model values were retrieved using the Lee et al. [1998, 1999] optimization technique with original and newly improved model parameters applied to Spectrix R rs ( ) data. Type 2 linear regression and RMSE values were calc ulated from log-transformed data. N Slope Offset R 2 RMSE log10 a g (440)_orig 45 0.359 -0.193 0.064 0.183 a g (440)_new 45 1.056 -0.066 0.559 0.136 a ph (440)_orig 45 0.555 -0.488 0.232 0.411 a ph (440)_new 45 0.625 -0.225 0.503 0.222 a tot (440)_orig 45 0.937 -0.125 0.746 0.156 a tot (440)_new 45 0.87 -0.097 0.809 0.123 b bp (555)_orig 4 3.197 4.172 0.417 0.549 b bp (555)_new 4 0.818 -0.369 0.377 0.131 phytoplankton and gelbstoff absorption values are typically underestimated as seen by negative y-intercepts. a tot ( ) is modeled more accurately (i.e. lower RMSE log10 ) than a ph ( ) and a g ( ), because it includes the water absorption, resulting in a larger dynamic range. Measured a g (440) values are from 0.99 to 6.14 times higher than a ph (440) values, with an average ratio of 3.25 for a g (440) to a ph (440). This is very typical for a Case 2 harbor, and explains why a g (440) values are modeled more accurately than a ph (440) values. a g (440) values exhibit a smaller dynamic range than a ph (440) values (Fig. 9a, b), which may explain the lower R 2 values shown in Table 3. Since only four b bp (555) values are available (Fig. 9d), statistical results are unreliable. These results indicate that the model para meters derived for the WFS need to be modified to improve optimization derived absorption and backscattering values for Bayboro Harbor.

PAGE 35

4.2.2. Modified model parameters for Bayboro Harbor After careful consideration, it was determined that model parameters for a ph ( ) (A 0 ( ) and A 1 ( ) from Eq. 15) and a g ( ) (S from Eq. 16 ) were the most important parameters requiring change when switching study areas from the WFS to Bayboro Harbor. All of the measured a ph ( ) spectra were used to generate new A 0 ( ) and A 1 ( ) values for Bayboro Harbor. Figure 10 s hows an example of how well modeled a ph ( ) Figure 10. One example of phytopl ankton absorption spectra, a ph ( ). Thick solid line is a ph ( ) measured in Bayboro Harbor dash line is the modeled a ph ( ) derived using the old parameters [ Lee et al. 1998], dots line is the modeled a ph ( ) derived using the modified A 0 A 1 parameters for Bayboro Harbor. spectra can match measured a ph ( ) with the new parameters, compared to results from using previous Lee et al. [1998] model parameters. 26 Gelbstoff absorption slopes between 350500 nm for Bayboro harbor data (May 2004 July 2005) range from 0.0158 to 0.0185 nm -1 with an average value of 0.0174

PAGE 36

nm -1 (n=64). An a g ( ) slope of 0.017 nm -1 was chosen for the modified parameter set instead of 0.015 nm -1 which was used by the Lee model (1999). Using these modified a ph (440) and a g (440) model parameters, the Spectrix R rs ( ) data were re-optimized using the Lee et al. [1998, 1999] technique, and an improved set of a tot (440) a g (440), a ph (440) and b bp (555) values were retrie ved (Fig. 11, Table 3). slope=1.056 y_int=-0.066 R2=0.559 RMSlog=0.136 n=45 0.1 1 10 0.1 1 10 ag(440) meas.(1/m)ag(440) mod.(1/m)a slope=0.625 y_int= -0.225 R2 = 0.503 RMSlog=0.222 n=450.01 0.1 1 10 0.010.1 1 10 aph(440) meas.(1/m)aph(440) mod.(1/m)b slope=0.87 y_int = -0.097 R2 = 0.809 RMSlog=0.123 n=450.1 1 10 0.1 1 10 atot(440) meas.(1/m)atot(440) mod.(1/m)c slope = 0.427 y_int = -0.369 R2 = 0.377 RMSlog = 0.131 n = 40.001 0.01 0.1 1 0.0010.010.11 bbp(555)_meas.(1/m)bbp(555)_mod.(1/m)d Figure 11. Optimization-derived a) a g (440, b) a ph (440), c) a total (440), d) b bp (555) values obtained from Spectrix R rs ( ) data compared to measured values. Model parameters optimized for Bayboro Harbor were used. On e-to-one lines (dash line) are shown along with type 2 linear regressi on functions (thick solid) calcu lated on long-transformed data Compared to Figure 9, large improvements both in data point distribution as well as regression trend lines occu rred once the parameters were modified. The largest 27

PAGE 37

28 improvements are in a ph (440) and b bp (555). Notice that modeled a g (440) and a tot (440) values continue to be somewhat smaller than measured values. The regression results are shown in Table 3 along with RMSE log10 estimates. Results show 12% error in a tot (440), 22% error in a ph (440) estimates, 13% error in a g (440), and 13% error in b bp (555), showing significant improvements ove r original parameters. The RMSE log10 for a g (440) is almost half of that calculated for a ph (440). This is because gelbstoff dominates the absorption in Bayboro Harbor with an average a g (440)/aph(440) value greater than 3. 4.2.3 Sensitivity analysis In order to determine why modeled ab sorption coefficients improved once the model parameters for a ph ( ) and a g ( ) were changed, a sensitivity analysis was performed. Spectrix R rs ( ) data were re-optimized using slightly different a ph ( ) and a g ( ) model parameters than optimal values. Measur ed versus modeled absorption coefficients were then compared. Figure 12 shows that changing th e gelbstoff slope from 0.017 nm -1 to 0.014 nm -1 decreases gelbstoff absorption for high a g (440) values and increas es absorption for low a g (440) values. The opposite is true when a higher a g slope (0.020 nm -1 ) is used. The effects on a ph (440) are similar. A lower a g slope (0.014) causes lower a ph (440) values to decrease and higher values to increase. Detail ed error estimates and regression results are shown in Table 4. Overall, deviations in gelbstoff slopes from 0.017 typically lead to increased errors. The only exception is when an a g slope of 0.014 is used. The RMSE log10 for a g (440) values decreases slightly. However, regression statistics using this slope are worse and a ph (440) values are modeled far less accuratedly (RMSE log10 = 0.365).

PAGE 38

0.1 1.0 10.0 0.1 1 10 ag(440) meas.(1/m)ag(440) mod.(1/m) 0.014 0.017 0.020 a 0.01 0.1 1 10 0.010.1110 aph(440) meas.(1/m)aph(440) mod.(1/m) 0.014 0.017 0.020 b Figure 12. Sensitivity test results showing effects of changing a g slopes on model outputs for a) a g (440) and b) a ph (440). Gelbstoff slopes examined were 0.014, 0.017 and 0.020 nm -1 One-to-one lines are shown. Table 4. Sensitivity test regression results. (a g slope 0.014,0.017,0.020) N Slope Offset R 2 RMSE log10 0.014 45 0.748 -0.088 0.516 0.100 0.017 45 1.056 -0.066 0.559 0.136 a g (440) 0.020 45 1.374 -0.058 0.564 0.196 0.014 45 0.894 -0.303 0.500 0.365 0.017 45 0.625 -0.225 0.503 0.222 a ph (440) 0.020 45 0.469 -0.208 0.304 0.297 29

PAGE 39

A similar sensitivity anal ysis was performed adjusting phytoplankton absorption model parameters A 0 ( ) and A 1 ( ) from Eq. 15. New A 0 ( ) and A 1 ( ) values generated from Bayboro Harbor a ph ( ) sample data (marked AMOS) were compared to the original A 0 ( ) and A 1 ( ) values derived from the WFS [ Lee et al., 1998]. The results are shown in Figure 13 and Table 5 and indi cate that using model a ph ( ) parameters developed from 0.1 1 10 0.1 1 10 ag(440)_meas.(1/m)ag(440)_mod.(1/m) WFS Bayboro a 0.01 0.1 1 10 0.010.1 1 10 aph(440)_meas.(1/m)aph(440)_mod.(1/ m WFS Bayborob Figure 13. Sensitivity test results showing effects of changing A 0 ( ) and A 1 ( ) parameters on model outputs for a) a g (440) and b) a ph (440). Phytoplankton absorption parameters from Lee et al. [1998] for the West Florida Shelf (WFS) and from Bayboro Harbor data (AMOS) collected during this study are compared. One-to-one lines are shown. 30

PAGE 40

31 Table 5. Sensitivity test regression results. (A 0 A 1 _WFS versus A 0 A 1 _Bayboro) N Slope Offset R 2 RMSE log10 WFS 45 1.081 -0.057 0.583 0.132 a g (440) Bayboro 45 1.056 -0.066 0.559 0.136 WFS 45 0.662 -0.200 0.460 0.243 a ph (440) Bayboro 45 0.625 -0.225 0.503 0.222 Bayboro Harbor data causes the error in a ph (440) to decrease slight ly (from 24% to 22%). No significant deviations in a g (440) estimates were observed due to changes in A 0 ( ) and A 1 ( ) values. The sensitivity tests performed indicate that changing gelbstoff slopes affect model outcomes more so than changing the a ph ( ) parameters. This makes sense since gelbstoff dominates the absorption va lues in Bayboro Harbor with average a g (440)/a ph ( ) values greater than three. 4.3 Validation of AMOS R rs ( ) data Beginning in May of 2004, AMOS was depl oyed in the Bayboro Harbor (Saint Petersburg, Florida). Automated R rs ( derived from AMOS measurements for May 2004 (hourly between 15:00 and 19:00 GMT) are show n in Figure 14. Notice that compared to the Spectrix R rs ( curves (Fig. 7), offsets exist amongst many of these spectra. This may be due to the presence of sun glint and/or re moval of too little sky light from the data. Similar to the Spectrix data, reflectance p eaks occur at ~570 and 685 nm. Spectra are slightly noisier due to instrument design (fiber optic cable).

PAGE 41

Figure 14. Remote-sensing reflectance spectr a from the AMOS sensor, May 2004 (hourly between 15:00 and 19:00 GMT). Measurements were collected in Bayboro Harbor (St. Petersburg, Florida). A comparison between several AMOS remote-sensing reflectance spectra and R rs ( collected nearby manually using hand-held Spectrix (within 30 minutes) is shown in Figure 15. Of the 3 stations used to show the variations, the spect ra with the highest reflectivity at 570nm provides the closest match between AMOS and Spectrix data. The spectra with the lowest reflectivity at 570nm provides the worst match. All spectra exhibit peak reflectivity ~ 570nm indicating consistent sp ectral calibrations for both sensors. It can be seen from th is figure also that the AMOS sensor consistently exhibits more reflectance at the blue end (~400 nm) compared to the Spectrix sensor. 32

PAGE 42

Figure 15. Comparisons between Spectrix a nd AMOS remote-sensing reflectance spectra measured in Bayboro Harbor (St. Pe tersburg, Florida) during May 2004. To quantitatively compare R rs ( ) measurements obtained from the AMOS and Spectrix sensors during May 2004, three wa velengths (blue=440nm, green=570nm, and red=640nm) were chosen (Fig. 16). Error es timates obtained from non-log transformed data are listed in Table 6. AMOS R rs ( ) underestimates Spectrix values at 570 and 640nm. RMSE lin estimates are only about 13% for both wavelengths. At 440nm, AMOS overestimates Spectrix R rs ( values (RMSE lin >400%). Table 6. Regression resu lts between measured R rs ( ) by direct Spectrix versus AMOS R rs ( ) at 440, 570 and 640nm wavelengths. N Slope Offset R 2 RMSE lin uncorrected 8 0.651 0.0006 0.888 4.647 440nm corrected 8 0.909 0.0001 0.933 0.438 uncorrected 8 0.921 -6E-05 0.921 0.129 570nm corrected 8 0.967 -3E-04 0.903 0.152 uncorrected 8 0.820 0.0001 0.913 0.126 640nm corrected 8 0.873 3E-05 0.8773 0.150 33

PAGE 43

Figure 16. AMOS versus Spectrix remote-sen sing reflectance values at 440, 570 and 640nm. Measurements were collected from Bayboro Harbor (St. Petersburg, Florida) in May 2004. Linear best-fit regression lines (s olid) are shown along with a one-to-one line (dotted). If a g (440) values are to be modeled successfully from R rs ( ) data, then accurate blue reflectance values are essential since ge lbstoff absorbs blue light strongly (Fig. 4). From Figure 11, recall that a g (440) values were slightly underestimated when derived from Spectrix R rs ( ) data using the Lee et al. [1998, 1999] optimization technique with the model parameters modified for Bayboro Ha rbor. Given that AMOS blue reflectance values are higher than Spectrix re flectance values (Fig. 16), and R rs ( ) is inversely proportional to a( ) (Eq. 7), AMOS modeled a g (440) values would underestimate measured a g (440) values even more than Spectrix modeled a g (440) values. In order to retrieve accurate a g (440) values from the AMOS R rs ( ) data, this excess blue light must first be removed. L ooking back at the actual design of the AMOS and Spectrix radiometers, one large difference is the fiel d-of-view (FOV) whereby the 34

PAGE 44

AMOS sensor has an FOV of ~25 o while the Spectrix has one of only ~10 o Taking this into consideration, perhaps not enough s kylight was subtracted from the upwelled radiance spectra due to wave facets bringing in light reflected from much larger angles, causing the blue reflectance values to be t oo high. In order to solve this problem, an effective Rayleigh-like corr ection term was added to th e optimization technique to remove excess blue light from the AMOS R rs ( data. This term, Ray( ) = Ray(400)(400/ ) 4.1 is subtracted from R rs ( ) meas in Eq.(19) along with Ray(400) and are then iteratively optimized along with a g (440), a ph (440), b bp (555), (550) and H using the Lee et al. [1998] optimization technique (Fig. 17). Figure 17. An example of remote-sensing re flectance spectra obtained by the AMOS and Spectrix sensors from Bayboro Harbor (St. Petersburg, Florida) on May 6, 2004. Excess blue light is removed from the AMOS R rs ( ) using an effective Rayleigh-like correction term incorporated in to the Lee et al. [1999] optimization model. Corrected AMOS R rs ( ) data are then compared to the Spectrix R rs ( ) data again, and results are shown in Figure 18. AMOS R rs ( ) decreased at the blue end (440nm), matching the one-to-one line when compared to the Spectrix R rs ( ). The regression 35

PAGE 45

Figure 18. Corrected AMOS versus Spectrix re mote-sensing reflectance values at 440, 570 and 640nm obtained from Baybo ro Harbor (St. Petersburg, Florida) in May 2004. AMOS R rs ( ) data were corrected by incorpor ating an effective Rayleigh-like correction term into the Lee et al. [1999] optimization mode l. Linear best-fit regression lines (solid) are shown along with a one-to-one line (dotted). results show ten-fold improvements in RMSE lin at 440 nm (Table 6), although, green (570nm) and red (640nm) RMSE lin increase slightly. Recall, how ever, that it is the blue reflectance values that are important for accu rate absorption coefficient retrievals. 4.4 AMOS Time-series analysis Using the modified model parameters discussed in Section 4.2, a time series of a g (440), a ph (440) and b bp (555) values were derived fr om Rayleigh-corrected AMOS R rs ( ) data for May 2004 (Figure 19). Directly measured values are also plotted for validation purposes. 36

PAGE 46

0.000 0.008 0.016 0.024 0.032 5/1/045/6/045/11/045/16/045/21/045/26/045/31/04 Timebbp(555) (1/m) 0.0100 0.1000 1.0000 5/1/045/6/045/11/045/16/045/21/045/26/045/31/04aph(440) (1/m) 0.010 0.100 1.000 10.000 5/1/045/6/045/11/045/16/045/21/045/26/045/31/04ag(440) (1/m) c b a Figure 19. AMOS R rs ( ) derived a) a g (440), b) a ph (440), c) b bp (555) values for Bayboro Harbor (St. Petersburg, FL) May 2004. Values were derived using the Lee et al. [1999] optimization model modified for Bayboro Harbor. The boxes are measured values. Modeled b bp (555) values derived from AMOS R rs ( ) data compare well with measured values exhibiting only a slightly higher RMSE log10 (19%) compared to when 37

PAGE 47

Spectrix derived valu es are used (RMSE log10 = 13%) (Fig. 20c, Table 7). Retrieved values for gelbstoff and phytoplankton absorption s how a similar pattern with Spectrix R rs ( ) retrieved values outperforming the AMOS R rs ( ) retrieved values, but with much larger RMSE log10 (Fig. 20a,b, Table 7). AMOS retrieved a g (440) values may be underestimated 0.1 1.0 10.0 0.1 1 10 ag(440)_meas.(1/m)ag(440)_mod.(1/m) amos spx 0.01 0.10 1.00 10.00 0.010.1110 aph(440)_meas.(1/m)aph(440)_mod.(1/m) 0 0.02 0.04 0.06 00.020.040.06 bbp(555)_meas.(1/m)bbp(555)_mod.(1/m) a b c Figure 20. Measured values compared to optimization model outputs. a) a g (440); b) a ph (440); c) b bp (555). 38

PAGE 48

Table 7. Regression results between AMOS and Spectrix R rs ( ) modeled and measured values of a ph (440), a g (440), b bp (555). Error estimates RMSE log10 are consistent with those of Carder et al. (2004), although th e other statistics are worse. N Slope Offset R 2 RMSE log10 a g (440)_amos 8 1.422 -0.036 0.466 0.271 a g (440)_spx 8 1.484 0.017 0.836 0.151 a ph (440)_amos 8 -0.421 -0.987 0.171 0.363 a ph (440)_spx 8 0.160 -0.527 0.053 0.244 b bp (555)_amos 4 0.824 -0.406 0.301 0.187 b bp (555)_spx 4 0.818 -0.369 0.377 0.131 due to inadequate sky light removal, even after spectra were corrected using the effective Rayleigh-like term. Modeled a ph (440) values are typically underestimated especially when a g (440): a ph (440) values are high (Figure 20b). Chlorophyll concentrations can be retrieved accurately from measurements of a ph (440) if the relationship between them is known [ Bricaud et al., 1995]. Measured a ph (440) data in this study for May 2004 show a strong, positive correlation with chlorophyll concentration (R 2 = 0.964, n = 8) (Fig. 21a). Since modeled a ph (440) values R2 = 0.964 n = 8 0.0 0.1 0.2 0.3 0.4 0.5 05101520 [Chl a ](mg m-3)aph(440)_meas.(1/m) R2 = 0.091 n = 8 0.0 0.1 0.2 0.3 0.4 0.5 05101520 [Chl a ](mg m-3)aph(440)_mod.(1/m) Figure 21. Measured chlorophyll a concentratio n from Bayboro Harbor compared to: a) measured a ph (440); b) modeled a ph (440). 39

PAGE 49

are highly inaccurate exhibiting an RMSE log10 of 36% (Fig. 20b) a weak negative correlation with chlorophyll concentration is observed (R 2 = 0.09, n = 8) (Fig. 21b). This indicates that a ph (440) data modeled from AMOS R rs ( ) data cannot be used to monitor chlorophyll concentrations. Ther efore, an alternative approa ch for deriving chlorophyll concentrations from AMOS R rs ( ) data is needed instead. It has been shown that th e height of the chlorophyll fluorescence peak (~685nm) above background radiances is highly correlated with the chlor ophyll concentrations [ Letelier and Abbott 1996] suggesting that such fl uorescence line heights (FLH) (Fig. 22) may be used to obtain estimates of chlorophyll concentrations from R rs ( ) In this study FLH is defined as ) 670750 )670690(*))750(R)670(R( )670(R()690(RFLHrs rs rs rs (20). 0 0.002 0.004 0.006 0.008 400500600700800 wavelength(nm)Rrs( )(sr-1) 2 3 1 Figure 22. Fluorescence line heights (FLH). Height above an imaginary line between 670 and 750nm. 1 =670nm, 2 =690nm, 3 =750nm. 40

PAGE 50

Chlorophyll concentrations in this stu dy are more highly correlated with FLHs calculated from AMOS R rs ( ) data (R 2 = 0.692, n = 11) (Fig. 23) than with modeled R2 = 0.692 n = 11 0 4 8 12 16 20 0.00000.00030.00060.00090.0012 FLH[Chl a ](mg m-3) Figure 23. Measured chlorophyll a concentration from Bayboro Harbor compared to FLH. a ph (440) data (Fig. 21b). Applying the best-fit linear relationship derived between AMOS FLHs and measured chlorophyll concentrations to AMOS R rs ( ) data for May 2004 results in the time series shown in Figure 24. The time series from AMOS provided interpolations for chlorophyll values between measurements. Meteorological conditions and tide information are shown in Figure 24. Many interesting features in model-derived a g (440), b bp (555) and [Chl a ] (Figs. 19, 24) match nicely with these environmental forci ngs. The higher wind values on May 4 th coincide with higher b bp (555) values, while chlorophyll concentra tions stayed low. This is likely the result of bottom sediment re-suspension caused by the wind. Precipitation (Fig. 25a) shows about 1 inch of rainfall on May 3 rd which could also contribute to increased nutrients along with nutrients released by er osion. These may be responsible for the 41

PAGE 51

0 4 8 12 16 20 5/1/045/6/045/11/045/16/045/21/045/26/045/31/04 time[Chl a](mg m-3) median_chl_mod chl_meas Figure 24. Chlorophyll a concentrations using FLH method (marked circles) derived by AMOS R rs ( ) data for May 2004 at Bayboro Harbor. Directly measured chlorophyll a concentrations are marked with squares. increase of chlorophyll concentration from May 4 th to May 7 th The small peak in a g (440) during this time could be the result of runoff of gelbstoff. Higher b bp (555) from May 12 th to 18 th also match the wind speed information during this period. The large amount of rainfall on May 15 th (Fig. 25a) is probably responsible for the steady [Chl a ] increase until May 19 th when nutrients were brought in from runoff as well as re-suspension, also indicated by higher b bp (555) values. Note the prominent peak shown in a g (440) from May 17 to 19 (Fig. 19) indicative of increased runoff. 42

PAGE 52

0 0.8 1.6 2.4 3.2 05/01/0405/07/0405/13/0405/19/0405/25/0405/31/04 timeprecipitation (in. ) precipitation a 0 2 4 6 8 10 5/1/045/7/045/13/045/19/045/25/045/31/04 timewind speed(m/s ) b Figure 25. Meteorological parameters of May 2004 near study area. a) rainfall data (from National Weather Service at Sa int Petersburg Station; b) hourly wind speed from buoy located on West Florida Shelf. Bar height represents wind speed range, with middle dots represent the average daily wind speed (b from NOAA CO-OPS website, for St. Petersburg, Florida location). Data from AMOS underwater units collect ed during mid-May fu rther validate the pattern observed for the AMOS R rs ( ) derived data (Fig. 26). Measured in situ 43

PAGE 53

0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 5/1/045/6/045/11/045/16/045/21/045/26/045/31/04 TimeBeam C660 (m-1) 0 10 20 30 40 50 60 5/1/045/6/045/11/045/16/045/21/045/26/045/31/04 TimeChl fluoresenc e Figure 26. AMOS underwater instrument m easurements from May 7 to May 21 at Bayboro Harbor: a) beam-c (660nm); b) uncalibrated chlo rophyll fluorescence by fluorometer. chlorophyll fluorescence values match nicely with the chlorophyll a concentrations calculated by the FLH-method (Fig. 24), especi ally for the peak values from May 17 to 19. The transmissometer-measured beam c(660) values show peak values around May 15-18 (Fig. 26a). This elevat ed change is much higher than what the modeled b bp (555) values indicate (Fig. 19c). 44

PAGE 54

45 These pattern co-variations can also be seen towards the end of May, when chlorophyll concentrations b ecome lower along with lower wind speed and lack of rainfall. Brief periods of stronger winds did o ccur at the end of May, but the directions were mostly from north and did not affect our research area significantly.

PAGE 55

46 5. DISCUSSION From May 2004 to July 2005 when AM OS was deployed in Bayboro harbor, significant variability in optical properties was obse rved. Measured chlorophyll a concentrations ranged from 2.5 to 47.7 mg m -3 a ph (440) ranged from 0.1 to 1.9m -1 and a g (440) ranged from 0.5 to 2.4 m -1 In order to derive ab sorption and backscattering coefficients and chlorophyll concentrations accurately from above-water remote-sensing reflectance spectra for such Case 2 waters, accurate R rs ( ) data and a successful R rs ( ) inversion technique are required. Prior to using the AMOS R rs ( ) data to derive a time-series of IOPs and chlorophyll concentrations, however, it was necess ary for this data first to be validated. Reflectance data measured using a hand-held Spectrix radiometer was used for this purpose. Higher R rs ( ) values at 440nm measured by AM OS compared to those measured by the Spectrix sensor indicated that perhaps not enough skylight had been removed from the AMOS upwelled radiance data. While both sensors viewed the water and sky at 30 o from nadir and zenith, respectively, th e larger field-of-view for AMOS (25 o ) compared to the Spectrix (10 o ) sensor, necessitates the use of a hi gher Fresnel reflectance factor with the AMOS data to be removed excess skylight [ Mobley, 1994]. Instead of reprocessing the AMOS R rs ( ) data using a higher Fresnel factor, an effective Rayleigh-like co rrection factor was included in the Lee et al. [1999]

PAGE 56

47 optimization model to allow variable amounts of excess blue-rich li ght to be removed. This correction term forced the AMOS and Spectrix R rs (440) data to agree more closely. These improvements are very important, since the main focus of this study which is to obtain accurate a ph (440) and a g (440) estimates requires a goo d understanding of the blue part of the spectrum. Differences in green (R rs (570)) and red (R rs (640)) reflectance values, however, remained high, and can be attributed to other sources. Spatial and temporal sampling differences must be considered when comparing R rs ( ) values from the AMOS and Spectrix se nsors. Bi-directional reflection due to varying viewing angles may introduce differences in R rs ( ) when surfaces are not Lambertian. Since the Spectrix and AMOS sens ors do not look at the same spot in the sky or water at sampling time, on top of vi ewing solid-angle differences, perfect matches should not be expected when R rs ( ) curves from both sensors are compared side-by-side. Whats more, since Spectrix measurements were made closer to the seawall compared to AMOS measurements, differences in water depth could also cause mismatches, especially in the gree n transparency window. Timing differences between the instrument measurements may also explain the small differences observed in R rs ( ) between the AMOS and Spectrix sensors. Even though Spectrix measurements were made w ithin 30 minutes of AMOS mesurements, solar radiance inputs du e to cloudiness and water conditi ons (wind riffles) can change by the second to introduce differences, especia lly when considering harbor areas with shallow bottoms. Compared to waters of the west Flor ida shelf, Bayboro Harbor is a highly gelbstoff-dominated environment. Ratios of a g (440)-toa ph (440) ranged from 1.0 to 6.7

PAGE 57

48 during this study period with an average value of 3.3, which is highly indicative of a Case 2 water environment. Relative to the total absorption coefficient at 440nm, a ph (440) and a g (440) contributed 24% and 65% to a tot (440), respectively, on av erage. As a result, the Lee et al. [1999] optimization model had to be modified to perform successfully in Bayboro Harbor. Changes made to the a ph ( ) and a g ( ) parameters in the semi-analytic model to more accurately represent the measured Ba yboro Harbor data improved IOP estimates derived from Spectrix R rs ( ) data. Compared to when the original Lee et al. [1999] parameters were used, root-mean-square e rrors generated between log-transformed measured versus modeled a ph (440) data decreased from 41% to 22%. Similarly, errors for a g (440) decreased from 18% to 14% and errors for b bp (555) decreased from 55% to 13%. Retrievals for a g (440) were much more accurate compared to those for a ph (440), again since Bayboro harbor is gelbsto ff-dominated. Modeled values for a g (440) were slightly underestimated perhaps because bottom c ontributions which cause higher green reflectivity were overestimated. Modeled bottom depths were typically much lower than the true depth supporting this theory. In order to put the retrieved erro rs calculated in this study for a ph (440) and a g (440) into perspective, results are compared to errors calculated semi-analytically for a large global data set (n = 656) using only S eaWiFS wavebands (412, 443, 490, 510, and 555nm) [ Carder et al., accepted]. This global data set was made available by the International Ocean Colour Coordinating Gr oup (IOCCG) for an algorithm testing round robin and contained no bottom effects. Compared to IOCCG results, a ph (440) values were

PAGE 58

49 modeled slightly more accurately for the IOCCG data set (RMSE log10 = 19.5%) and a g (440) values were modeled much less accurately (RMSE log10 = 27.9%). Since model retrievals for a ph (440) were not very accurate for this study owing to the gelbstoff-dominated nature of Bayboro Harbor, chlorophyll concentrations were derived from AMOS R rs ( ) data using fluores cence line heights. While the algorithm developed in this study may be highly season -specific (i.e. will only work with AMOS data), site-specific, and time-specific (i .e. other relationships may be observed for different sensors), chlorophyll concentrati ons were derived fair ly accurately (RMSE log10 = 21.5%) using only three R rs ( ) wavebands (670, 690, 750n m). Results from deriving and testing numerous empirical band-ratio algorithms that require SeaWiFS wavebands on a large global dataset (n = 919) show similar errors with RMSE log10 values ranging from 17.2 to 31.1% [ OReilly et al. 1998]. A recent NASA report on ocean color and carbon for the Chesapeake Bay (Case 2) [ Signorini et al. 2005] shows that the best statis tical results obtained for modeled chlorophyll concentrations using the regionally tuned Garver-Siegel-Maritorena (GSM01_CB) semi-analytic algorithm [ Maritorena et al., 2002] yield an absolute percent difference (APD) equal to 68.34%. This SA algorithm requires R rs ( ) data at SeaWiFS wavebands. Chlorophyll concentr ations derived from AMOS R rs ( ) FLH data for May 2004 in Bayboro Harbor were estimated more accurately (APD = 41.26%) compared to the GSM01-CB results.

PAGE 59

50 6. CONCLUSIONS The measured high levels of chlorop hyll concentrations in Bayboro Harbor indicate that it is a hi ghly productive area. Using the origin al parameters for the Lee et al. [1999] SA model to retrieve in-water optical properties (a ph (440), a g (440) and b bp (555)), results in large retrieval errors. New optimi zation model parameters were applied, and retrieval results show much improvement s upporting the hypothesis th at the Lee et al. [1998, 1999] semi-analytical hyperspectral remo te-sensing model could be fitted to work in very turbid Bayboro Harbor water. A sens itivity analysis further demonstrated the effectiveness of the new parameters. AMOS R rs ( ) data from May 2004 were used to evaluate the system performance against in-situ measurements by the hand-held Spectrix sensor. The traditional approach to derive measured R rs ( ) spectra does not work well with AMOS. A new approach is used to calculate AMOS R rs ( ) from measured downwelli ng irradiance and upwelling radiance, with the removal of proper Raleigh scattering and model residual ( ) at 750nm. The results show significant improvements in the blue part of the spectrum, which enables better model estimates. The SA model derived time-series valu es for the month of May 2004, namely a g (440) and b bp (555), correspond nicely to measured values as well as to external environmental changes. Because of the dom inance of absorption by gelbstoff, modeled

PAGE 60

51 a ph (440) values cannot be used to accuratel y estimate chlorophyll a concentrations. However, improved estimates of chlorophyll co ncentration in these turbid waters were obtained from fluorescence line heights. This validates the effectiv eness of AMOS as a tool not only to provide meas urements for high-altitude se nsor-calibration purposes, but also to generate time series for coastal marine ecosystems. The optimization model used demonstrated s lightly biased errors for this specific location and model parameters. This might n eed to be fine-tuned for optimal results. Better results could also arise from better AM OS instrument calibration and its situation over a flat bottom away from a seawall. The success of AMOS results encourages further testing and implementation of such automate d, continuous data sampling stations for wide-area field deployment.

PAGE 61

52 LIST OF REFERENCES Austin, R.W., and T.J. Petzold, The determina tion of the diffuse attenuation coefficient of sea water using the coastal zone color scanner, in Oceanography from Space, edited by J.F.R. Gower, pp. 239-256, Plenum Press, New York, 1981. Bidigare, R.R., J.H. Morrow, and D.A. Kiefer Derivative analysis of spectral absorption by photosynthetic pigments in the western Sargasso Sea, Journal of Marine Research 47 323-341, 1989. Bricaud, A., M. Babin, and A. Morel, Variab ility in the chlorophyll-specific absorption coefficients of natural phytoplankt on: Analysis and parameterization, Journal of Geophysical Research 100 (C7), 13,321-13,332, 1995. Bricaud, A., and D. Stramski, Spectral abso rption coefficients of living phytoplankton and nonalgal biogenous matter: a compar ison between the Peru upwelling area and the Sargasso Sea, Limnology and Oceanography, 35 (3), 562-582, 1990. Bukata, R.P., J.H. Jerome, K.Y. Kondratyev, and D.V. Pozdnyakov, Satellite monitoring of optically-active compounds of inland wa ters: an essential input to regional climate impact studies, Journal Great Lakes Research 17, 470-478, 1991. Cannizzaro, J.P., and K.L. Carder, Estimating chlorophyll a concentrations from remotesensing reflectance data in optically shallow waters, Remote Sensing of Environment 2005 (submitted).

PAGE 62

53 Cannizzaro, J.P., K.L. Carder, F.R. Chen, J. J. Walsh, Z.P. Lee, and C. Heil, A novel optical classification technique for detecti on of red tides in the Gulf of Mexico: application to the 2001-2002 bloom event, in Florida Fish and Wildlife Conservation Commission and Intergovernmental Oceanographic Commission of UNESCO edited by J.H.L. K.A. Steidinger, C.R. Tomas, G.A. Vargo, pp. 4, Intergovernmental Oceanographic Commi ssion of UNESCO, St. Pete Beach, FL, 2002. Carder, K.L., J.P. Cannizzaro, F.R. Chen, and Z.P. Lee, MODIS Semi-Analytic Algorithm for IOP, in Reports of the Internationa l Ocean-Colour Coordinating Group, No. 10 edited by Z.P. Lee, accepted. Carder, K.L., F.R. Chen, Z.P. Lee, S.K. Hawes, and D. Kamykowski, Semi-analytic Moderate-Resolution Imaging Spectro meter algorithms for chlorophyll a and absorption with bio-optical domains based on nitrate-depletion temperatures, Journal of Geophysical Research 104 (C3), 5403-5422, 1999. Carder, K.L., S.K. Hawes, K.A. Baker, R.C. Smith, R.G. Steward, and B.G. Mitchell, Reflectance model for quantifying chlor ophyll a in the pres ence of productivity degradation products, Journal of Geophysical Research 96 (C11), 20,599-20,611, 1991. Carder, K.L., and R.G. Steward, A remote-sensing reflectance model of a red-tide dinoflagellate off west Florida, Limnology and Oceanography, 30 (2), 286-298, 1985.

PAGE 63

54 Carder, K.L., R.G. Steward, G.R. Harvey, and R.B. Ortner, Marine humic and fulvic acids: their effects on remote sensing of ocean chlorophyll, Limnology and Oceanography 34 (1), 68-81, 1989. Carder, K.L., R.G. Steward, J.H. Paul, and G.A. Vargo, Relationships between chlorophyll and ocean color constituents as they affect remote-sensing reflectance models, Limnology and Oceanography, 31 (2), 403-413, 1986. Doerffer, R., and J. Fisher, Concentrati ons of chlorophyll, suspended matter, and gelbstoff in case II waters derived from satellite coastal zone color scanner data with inverse modeling methods, Journal of Geophysical Research 99, 74757466, 1994. Esaias, W., M. Abbott, I. Barton, O.B. Brown, J.W. Campbell, K.L. Carder, D.K. Clark, R.H. Evans, F.E. Hoge, H.R. Gordon, W.M. Balch, R. Letelier, and P.J. Minnett, An overview of MODIS capabilities for ocean science observations, IEEE Trans. Geosci. Remote Sens. 36 1250-1265, 1998. Gordon, H.R., O.B. Brown, R.H. Evans, J.W. Brown, R.C. Smith, K.S. Baker, and D.K. Clark, A semianalytic radiance model of ocean color, Journal of Geophysical Research 93 (D9), 10,909-10,924, 1988. Gordon, H.R., D.K. Clark, J. W. Brown, O.B. Brown, R.H. Evans, and W.W. Broenkow, Phytoplankton pigment concentrations in the Middle Atlantic Bight: comparison of ship determinations and CZCS estimates, Applied Optics 22 (1), 20-36, 1983. Gordon, H.R., and M. Wang, Retrieval of water-leaving radiance and aerosol optical thickness over the oceans with SeaWiFS: A preliminary algorithm, Applied Optics 33, 443-452, 1994.

PAGE 64

55 Hoepffner, N., and S. Sathyendranath, Determination of the major groups of phytoplankton pigments from the absorption spectra of total particulate matter, Journal of Geophysical Research 98 (C12), 22,789-22,803, 1993. Holm-Hansen, O., C.J. Lorenzen, R.W. Ho lmes, and J.D.H. Strickland, Fluorometric determination of chlorophyll, J. Cons. Perm. Int. Explor. Mer 30 (1), 3-15, 1965. Hu, C., K.L. Carder, and F.E. Mller-Ka rger, Atmospheric correction of SeaWiFS imagery over turbid coastal waters: a practical method, Remote Sensing of Environment 74, 195-206, 2000. Jerlov, N.G., Marine Optics, Elsevier, New York, 1976. Kiefer, D.A., and J.B. SooHoo, Spectral absorp tion by marine particles of coastal waters of Baja California, Limnology and Oceanography, 27 (3), 492-499, 1982. Kirk, J.T.O., Light capture by aquatic plants, in Light and Photosynthesis in Aquatic Ecosystems edited by J.T.O. Kirk, pp. 201-218, Cambridge University Press, Cambridge, 1994. Kishino, M., M. Takahashi, N. Okami, and S. Ichimura, Estimation of the spectral absorption coefficients of phytoplankton in the sea, Bulletin of Marine Science 37 (2), 634-642, 1985. Lee, Z., Visible-infrared remote-sensing mode l and applications for oceanic waters, Ph.D. dissertation thesis, University of South Florida, St. Petersburg, Fla., 1994. Lee, Z., K.L. Carder, R.F. Chen, and T.G. Peacock, Properties of the water column and bottom derived from Airborne Visible In frared Imaging Spectrometer (AVIRIS) data, Journal of Geophysical Research 106 (C6), 11,639-11,651, 2001.

PAGE 65

56 Lee, Z., K.L. Carder, and K. Du, Effects of molecular and particle scatterings on the model parameter for remote-sensing reflectance, Applied Optics, 43 (25), 49574964, 2004. Lee, Z., K.L. Carder, C. Mobley, R.G. Stewar d, and J.S. Patch, Hyperspectral remote sensing for shallow waters: I. A semi-analytical model, Applied Optics 37 (27), 6329-6338, 1998. Lee, Z., K.L. Carder, C.D. Mobley, R.G. Stew ard, and J.S. Patch, Hyperspectral remote sensing for shallow waters: 2. Deriving bottom depths and water properties by optimization, Applied Optics, 38 (18), 3831-3843, 1999. Lee, Z., K.L. Carder, T.G. Peacock, C.O. Davis, and J.L. Mueller, Method to derive ocean absorption coefficients from remote-sensing reflectance, Applied Optics, 35 (3), 453-462, 1996. Letelier, R.M., and M.R. Abbott, An analysis of chlorophyll fluores cence algorithms for the Moderate Resolution Imaging Spectrometer (MODIS), Remote Sensing of Environment 58, 215-223, 1996. Lewis, M.R., M. Carr, G. Feldman, W. Esaias and C. McClain, Influence of penetrating solar radiation on the heat budget of the equatorial Pacific ocean, Nature, 347, 543-545, 1990. Maffione, R.A., and D.R. Dana, Instruments and methods for measuring the backwardscattering coefficient of ocean waters, Applied Optics, 36 (24), 6057-6067, 1997. Maritorena, S., D.A. Siegel, and A.R. Peters on, Optimization of a semi-analytical ocean color model for global-scale applications, Applied Optics 41 (15), 2705-2714, 2002.

PAGE 66

57 Marra, J., T. Dickey, W.S. Chamberlin, C. Ho, T. Granata, D.A. Kiefer, C. Langdon, R.C. Smith, K.S. Baker, R.R. Bidigare, and M. Hamilton, Estimation of seasonal primary production from moored optical sensors in the Sargasso Sea, Deep-Sea research 97 7399-7412, 1992. Millie, D.F., O.M. Schofield, G.J. Kirkpatr ick, G. Johnsen, P.A. Tester, and B.T. Vinyard, Detection of harmful algal bl ooms using photopigments and absorption signatures: A case study of the Fl orida red tide dinoflagellate, Gymnodinium breve, Limnology and Oceanography, 42 (5), 1240-1251, 1997. Mitchell, B.G., Coastal zone color scanner retrospective, Journal of Geophysical Research 99 7291-7292, 1994. Mobley, C.D., Light and water: radiative transfer in natural waters 592 pp., Academic Press, San Diego, 1994. Morel, A., Optical properties of pure water and pure sea water, in Optical Aspects of Oceanography edited by N.G. Jerlov, and E. S. Nielsen, pp. 1-24, Academic Press, London, 1974. Morel, A., and D. Antoine, Heating rate with in the upper ocean in relation to its biooptical state, Journal of Physical Oceanography 24, 1652-1665, 1994. Morel, A., and L. Prieur, Analysis of variations in ocean color, Limnology and Oceanography 22 (4), 709-722, 1977. Nelson, J.R., and C.Y. Robertson, Detrital spectral absorption: laboratory studies of visible light effects on phyt odetritus absorption, bacter ial spectral signal, and comparison to field measurements, Journal of Marine Research 51, 181-207, 1993.

PAGE 67

58 Neumann, A., H. Krawczyk, and T. Wal zel, A complex approach to quantitative interpretation of spectral high resolution imagery, in Third Thematic Conference on Remote Sensing for Marine and Coastal Environments Seattle, USA, 2000. O'Reilly, J.E., S. Maritorena, B.G. Mitchell, D.A. Siegel, K.L. Carder, S.A. Garver, M. Kahru, and C. McClain, Ocean colo r chlorophyll algorithms for SeaWiFS, Journal of Geophysical Research 103 (C11), 24,937-24,953, 1998. Platt, T., and S. Sathyendranath, Oceanic primary production: estimation by remote sensing at local and regional scales, Science 241 1613-1620, 1988. Pope, R., and E. Fry, Absorption spectrum (380-700nm) of pure wa ters: II. Integrating cavity measurements, Applied Optics 36, 8710-8723, 1997. Roesler, C.S., M.J. Perry, and K.L. Carder, Modeling in situ phytoplankton absorption from total absorption spectra in productive inland marine waters, Limnology and Oceanography 34 (8), 1510-1523, 1989. Siegel, D.A., M. Wang, S. Maritorena, a nd W. Robinson, Atmospheric correction of satellite ocean color imagery: the black pixel assumption, Applied Optics, 39 (21), 3582-3591, 2000. Signorini, S.R., C.R. McClain, A. Mannino, and S. Bailey, Report on ocean color and carbon study for the South A tlantic Bight and Chesapeake Bay regions, pp. 45, Goddard Space Flight Center, Greenbelt, MD, 2005. Steward, R.G., and K.L. Carder, Compre ssion of autonomous hyperspectral data, in Ocean Optics XVI Office of Naval Research, Santa Fe, NM, 2002.

PAGE 68

59 Stumpf, R.P., and J.R. Pennock, Remote estima tion of the diffuse attenuation coefficient in a moderately turbid estuary, Remote Sensing of Environment 38, 183-191, 1991. Yentsch, C.S., Measurement of visible light absorption by particulate matter in the ocean, Limnology and Oceanography, 7 207-217, 1962.