Hyperspectral Remote Sensing of Suspended Minerals, Chlorophyll and Coloured Dissolved Organic Matter in Coastal and Inland Waters, British Columbia, Canada
Laurie C. Gallagher
B.Sc., University of Victoria,
2002
A Thesis Submitted in Partial Fulfillment of the Requirements for the Degree of
MASTER OF SCIENCE
in the School of Earth and Ocean Sciences
We accept this thesis as conforming to the required standard
O Laurie C. Gallagher. 2004 University of Victoria
All rights reserved. This thesis may not be reproduced in whole or in part,
by
photocopy or other means, without the permission of the author.Supervisor: Dr. Kevin Telmer
Abstract
Monitoring and maintaining water and ecological quality of coastal zones is a growing global concern. It is estimated that over 60% of the human population lives in the coastal zone. Human activities in the coastal zone and catchment areas result in modified runoff and delivery patterns of nutrients and sediments to lakes and coastal waters (i.e. land clearing, agriculture, mining, and urban and/or industrial
development). Coastal development can also lead to modification of foreshores, loss of key habitats, changes to flushing rates, re-suspension of sediments, and direct inputs of nutrients and pollutants. Due to the many impacts and uses of the coastal zone, coastal zone management has become a high priority for many nations.
Remote sensing applications for water quality in coastal and inland waters offer spatial and temporal coverage that is unattainable by other means and is ideally suited to cover the broad range of space and time scales associated with coastal applications. In order for remote sensing data to support water quality monitoring efforts and coastal decision making, standard and reliable algorithms are required. For instance, estimating phytoplankton abundance is critical for critical for improved understanding and management of pelagic and benthic ecosystems, wild fisheries and aquaculture, and impacts of added nutrients. Sediment concentrations can also be estimated from remote sensing data. Knowledge of sediment distribution is important as many contaminants (organics and heavy metals) adsorb strongly to fine sediments, which act as a carrier and serve to remove contaminants from the water column. The assessment of water quality using remote sensing techniques provides reliable, consistent and affordable data for coastal zone managers and researchers. Data collected from satellite, airborne or moored sensors can provide important time series data sets for biogeochemical studies and climate change research.
The goal of this thesis was to develop and calibrate a semi-analytical bio- optical model at high spatial and spectral resolution to characterize the optical and geochemical properties of inland and coastal waters in southern Vancouver Island, British Columbia. The calibrated bio-optical model was used as a predictive tool to
estimate the effect of varying concentrations of water quality parameters in order to develop robust algorithms for detecting suspended minerals and chlorophyll-a from radiometric data. In August 2003, optical data was acquired using hyperspectral radiometers, at 3 nm spectral resolution, in both Lake Cowichan and the coastal waters of Cowichan Bay and Satellite Channel. Water samples for geochemical analysis were collected simultaneously with the radiometric data. At that time, the inland waters of Lake Cowichan were oligotrophic, with concentrations of
chlorophyll-a < 1 pg L-' and suspended minerals < 1 mg L-'. Coastal waters were mesotrophic with concentrations of chlorophyll from 1 - 3 pg L-' and suspended
minerals from 1 - 5 mg L-'.
The semi-analytical bio-optical model predicts remote sensing reflectance as a function of the concentrations of chlorophyll and suspended minerals, and the
absorption by CDOM by relating the inherent optical properties of absorption and backscattering to measured values of upwelling radiance and downwelling irradiance. Results of the modelling indicated that total absorption was higher in the coastal waters than the inland waters due to higher concentrations of suspended minerals and chlorophyll, and higher coloured dissolved organic matter (CDOM) absorption. As well, CDOM absorption had a higher mean slope in coastal waters (SCDOM = 0.024 versus 0.017 nm-' for inland waters) indicating that the marine CDOM was likely more refractory (degraded) than the inland water CDOM. Total backscattering was also higher in the coastal waters, due to higher concentrations of suspended minerals and chlorophyll. Results of the modeling estimate average backscattering efficiency for coastal and lake waters was 0.7% and 2.6% respectively. The average n for coastal and lake waters was 1.07 and 1.17 respectively, suggesting that the suspended minerals may be feldspars, quartz or calcite. The lower n values in the coastal waters are those expected for a mix of phytoplankton and suspended minerals and clearly show the difference between the two environments.
Remote sensing reflectance, measured at 3 nm resolution over the range 400 -
700 nm, was used for model calibration. The model predicted remote sensing
reflectance to f 10% for the region 400 - 650 nm and therefore accurately reproduced this spectral region. Additionally, remote sensing reflectance measured from above-
iv water and in-water platforms agreed to within 10%. This strong agreement and the reliability of the model, suggested that remote sensing reflectance measured from other above-water hyperspectral platforms, such as moored, airborne or satellite sensors have great potential to retrieve the geochemical constituents of water in this region.
The next step was to develop a method to retrieve the geochemical data from the radiometric data. This was accomplished by evaluating the sensitivity of above- water remote sensing reflectance to changing concentrations of suspended minerals, chlorophyll, and CDOM absorption, using simulations of the bio-optical model calibrated for the coastal waters. The optical data that was acquired at 3 nm resolution was degraded to 11 nm resolution in order to mimic the resolution of the airborne sensor, the Compact Airborne Spectrographic Imager (CASI), which was flown over the study site at the time of optical and geochemical data collection.
Model simulations show an increase in Rrs(O+) values with increasing suspended mineral concentrations. Correlations between Rr,(O+) band ratios and the concentration of suspended minerals indicated that the spectral region 475 - 675 nm
was sensitive to changing concentrations (R
2
0.80). The band ratios that best estimated suspended mineral concentrations were around the ratio 560 / 660 nm. More than 90 Rrs(O+) band ratios at 3 nm resolution were used to predict theconcentration of suspended minerals (p < 0.005). The possible combinations of band ratios were less for the degraded 11 nm data, a result of the reduced dimensionality of the data set.
For chlorophyll, model simulations and correlation statistics indicated that the region 500 - 600 nm was the most sensitive to varying chlorophyll concentrations. The chlorophyll sensitivity across 500 - 600 nm was not expected, as the maximum absorption of chlorophyll occurs at
-
440 nm, and a secondary absorption feature occurs at-
680 nm. However, absorption by CDOM overwhelmed the absorption due to chlorophyll from 400 - 500 nm, and the secondary absorption signal at 680 nm was masked by absorption due to suspended minerals and waters itself. At 500 - 590 nm,chlorophyll was estimated using band ratios in this region. Band ratios around 520 1 560 produced the highest correlations with predicted values (p< 0.005).
Simulations using CDOM absorption effected the spectral region 400 - 550
nm. However, absorption by suspended minerals and chlorophyll also effect this region. As a result, CDOM absorption must be significantly greater than absorption due to suspended minerals and chlorophyll in order to accurately estimate CDOM absorption. Alternatively, CDOM can be determined by solving the equations of the bio-optical model, provided the residual value for CDOM is larger than cumulative sum of the measurement errors.
The capability of simulating varying optical water parameters is important for the development of algorithms beyond the scope offered using a traditional empirical approach. The model presented can potentially serve another important role, by enhancing regional algorithms to facilitate the use of existing satellites such as MODIS (Moderate Resolution Imaging Spectroradiometer), MERIS (Medium Resolution Imaging Spectrometer Instrument), and SeaWiFS (Sea-viewing Wide Field-of-view Sensor).
Table of Contents
ABSTRACT
...
I1 TABLE OF CONTENTS...
VII ACKNOWLEDGMENTS...
IX DEDICATION...
XI...
LIST OF FIGURES XI1
LIST OF TABLES
...
XI11...
CHAPTER 1 1
...
1.1 A BRIEF HISTORY 1
...
1.2 CONCEPTS AND DEFINITIONS 3
...
1.2.1 Radiance 4
...
1.2.2 Irradiance 5
...
1.2.3 Remote Sensing Reflectance 6
...
1.2.4 Inherent and Apparent Optical Properties 6
... 1.3 MAJOR LIGHT ABSORBING COMPONENTS OF THE AQUATIC SYSTEM 7 ...
1.3.1 Absorption due to Water 7
... ...
1.3.2 Absorption due to Coloured Dissolved Organic Matter ... 7
...
1.3.3 Absorption due to Suspended Minerals 9
...
1.3.4 Absorption due to Phytoplankton 9
...
1.3.5 The Package Effect 10
...
1.3.6 Total Absorption 13
... 1.4 MAJOR LIGHT SCATTERING COMPONENTS OF THE AQUATIC SYSTEM 14
...
1.4.1 Backscattering due to Water I5
...
1.4.2 Backscattering due to Suspended Minerals 16
...
1.4.3 Backscattering due to Phytoplankton 1 7
... ...
1.5 ATTENUATION IN THE AQUATIC MEDIUM .- 19
...
1.6 RATIONALE FOR STUDY 19
...
1.7 PURPOSE OF STUDY 20
...
CHAPTER 2 22
HYPERSPECTRAL REMOTE SENSING OF SUSPENDED MINERALS. CHLOROPHYLL AND CDOM IN COASTAL AND INLAND WATERS.
...
BRITISH COLUMBIA. CANADA 22
... 2.1 INTRODUCTION 22 ... 2.2 MODEL APPROACH 23 ... 2.2.1 Absorption Model 27 ... 2.2.2 Backscattering Model 29 ... 2.3 STUDY AREA 31 ... 2.4 METHODS 33 ... 2.4.1 Water Collection 33 ...
2.4.2 Analysis of Chlorophyll-a. TSS. DOC. CDOM 34
...
2.4.3 Optical Data Collection 36
...
2.5 REMOTE SENSING REFLECTANCE DETERMINATIONS 37
...
2.5.1 In- Water Remote Sensing Reflectance 38
...
2.5.2 Above- Water Remote Sensing Reflectance 39
...
2.6 RESULTS AND DISCUSSION 40
...
2.6.1 Water Geochemistry 40
...
2.6.2 Vertical Attenuation of Light 43
...
...
V l l l2.6.4 Closure between hyperspectral platforms ... 45
2.6.5 Above-water Remote Sensing Reflectance, R, (O+) ... 47
... 2.6.6 Model Calibration 51 2.7 CONCLUSIONS ... 56
...
CHAPTER 3 58 ESTIMATING SUSPENDED MINERALS AND CHLOROPHYLL FROM HYPERSPECTRAL SENSORS IN COASTAL WATERS OF BRITISH COLUMBIA. CANADA...
58 ... 3.1 INTRODUCTION 58 3.1.1 Theory ... 59 3.2 STUDY AREA ... 59 3.3 METHODS ... 61 3.3.1 Framework ... ... 613.3.2 Water Collection and Analysis ... 62
3.3.3 Optical Data Collection ... 63
3.4 MODELLING ... 64
3.4.1 Approach ... 64
3.4.2 Hyperspectral Bio-Optical Model ... 65
3.4.3 Absorption Model ... 66
... ... 3.4.4 Backscattering Model ..67
3.4.5 Measured Rrs(O+) Data Sets ... 69
... 3.5 SENSITIVITY ANALYSIS 71 ... 3.5.1 Simulations 71 3.6 RESULTS AND DISCUSSION
...
723.6.1 Suspended Minerals ... 72
3.6.2 Chlorophyll ... ... 73
3.6.3 CDOM ... 75
3.7 ALGORITHM DEVELOPMENT AND EVALUATION ... 76
... 3 . 7.1 Suspended Minerals 76 ... 3.7.2 Chlorophyll 80 3.7.3 Steps for Estimating Suspended Mineral and Chlorophyllj?om Above- water ... 83
3.8 CONCLUSIONS ... 83
LITERATURE CITED
...
85APPENDICES
...
91...
UNIVERSITY OF VICTORIA PARTIAL COPYRIGHT LICENSE 94
...
Acknowledgments
It is a pleasure to thank the many people who have been a part of this graduate degree; friends, teachers, and colleagues. First and foremost, I would like to thank my supervisor, Dr. Kevin Telmer for his tireless support throughout the course of my graduate degree. Kevin has been a wonderful advisor and teacher - Thank you Kevin!
I would also like to thank Dr. Maycira Costa, who has given so generously of her time and expertise that I will always be grateful. I am extremely grateful to Dr. Rob Macdonald for showing me what a career in science looks like, and for his guidance, inspiration and enthusiasm.
I must also thank the crazies in the Geochemistry and Spectral Labs -those who have made the experience fun and wonderful, including: Christiaan Piller (SBJ), Andrew Hamilton, Michelle DesJardins, Mike Sanborn, Jody Spence, Jen Dyck, Sarah Minnery, Dr. Richard Cox, Eduardo Loos, and Karolyn Jones. I must also mention my friends who have cheered me on: Colleen Delaney, Cathy Channing, Kathleen Milward, Leah Fraser and Larisa Hutchison.
Thank you also to my friends and colleagues at the Institute of Ocean Sciences, including the inspirational Sophie Johannessen, and her wonderful husband Duncan, Luanne Chew and Cindy Wright. You helped me keep it all in perspective and reminded me that it's all worth it!
I would also like to say a huge thank you to my Mom, who has always been my biggest supporter! Thank you Mom for creating an environment in which following this path seemed so natural. Thanks also to Chuck (one of the few who are likely to read this thesis!), my sister Lynda, my brother Jody and sister-in-law Eileen, and their families. My in-laws, Rosalind and Bryan Bisley, have also been a big support for me. Thank you so much! I wish to thank my entire family for providing such a loving environment for me.
And most importantly I would like to thank my husband Ian. Thank you for your love, kindness, patience and generosity.
I thank all the support staff at the School of Earth and Ocean Sciences, in particular, Sussi Arason, Claire Tugwell and Terry Russell.
This work was financially supported by GEOIDE, the GEOmatics for Informed DEcisions Network. I would like to thank GEOIDE for their interest and support in this project.
Dedication
xii
List of Figures
Figure 1 . I: Radiant intensity from a point source ... 4
Figure 1.2: Radiance (L(1. 8. p. sr-I)) ... 5
Figure 1.3: Absorption spectrum for water ... 7
Figure 1.4: Absorption spectrum for CDOM ... 8
... Figure 1.5. Absorption spectrum for suspended minerals 9 ... ... Figure 1.6. Absorption spectrum for phytoplankton .. I0 ... Figure 1.7. The Package Effect I I ... Figure 1.8. Absorption of light by a particle in suspension I I Figure 1.9. Total absorption spectrum for natural waters ... 13
... Figure I . 10: The volume scattering&nction, P(8) 14 ... Figure 1.11. Backscattering spectrum due to water 15 Figure 1.12. Backscattering spectrum due to suspended minerals ... 17
Figure 1.13. Backscattering spectrum due to suspended minerals ... 18
Figure 2.1: Schematic diagram of the semi-analytical optical model . ... 24
Figure 2.2. Absorption and backscattering spectra for productive coastal waters ... 27
Figure 2.3. Location of the sites on Southern Vancouver Island, British Columbia, Canada ... 33
Figure 2.4: a) The in-water optical instrument conjguration; b) the above-water optical instrument ... conzguration 37 Figure 2.5. Conditions in Lake Cowichan for August 2003 ... 40
Figure 2.6. Conditions in Cowichan Bay for August 2003 ... 42
Figure 2.7. Upwelling radiance (Lu) values for representative sites ... 43
Figure 2.8. The average vertical attenuation coeflcient (Kd) for downward irradiance ... 44
Figure 2.9: Comparison of above-water remote sensing reflectance (Rrs(O+)) from in- and above- ... water hyperspectral platforms 46 Figure 2.10: Above-water remote sensing rejlectance (Rrs(O+)) curves determinedfrom the in-water . . ~ ... hyperspectral platform for the seven the sites in the study area 48 Figure 2.11. Results of the absorption and backscattering models for two representative sites ... 49
Figure 2.12. The modeled Rrs(0+) vs . measured Rrs(O+) from the in-water platform ... 55
Figure 3.1: The study area. southeastern Vancouver Island, British Columbia. Canada ... 61
Figure 3.2. Schematic illustrating the data used in algorithm development and evaluation ... 62
Figure 3.3: Illustration of interrelationships between the apparent optical properties and the inherent ... optical properties 64 ... Figure 3.4. Absorption and backscattering spectra 67 ... Figure 3.5. Measured and modeled Rrs(0+) spectra data 69 ... Figure 3.6: 2 - 0 ternary scatterplot 70 Figure 3.7: Change in total absorption and backscattering due to increasing concentrations of ... suspended minerals 72 Figure 3.8. Change in Rrs(O+) due to increasing concentration of suspended minerals ... 73
Figure 3.9: Change in absorption, backscattering and Rrs(O+) due to increasing concentrations of chlorophyll ... 74
Figure 3.10: Changes in absorption and Rrs(0+) spectra due to varying CDOM440 ... 76 ... Figure 3.11: Estimation of the concentration of suspended minerals 77
...
...
X l l l
List of Tables
Table 2.1: Sample site information and illumination conditions at the time of sampling ... ... 26
Table 2.2: Data quality ... ... ... ... .. ... ... .... ... . . ... . . . . . . . . . . . . . . . . 35
Table 2.3: Calibration coeficients used in the tuning of the absorption and backscattering models. .. 52
Table 3.1: Data quality. . . . .. . .
.
.. . . 63Table 3.2: Sample site information and illumination conditions at the time of sampling. ... 65
Table 3.4: Above-water remote sensing reflectance (Rrs(O+)) I I nm band ratios used to estimate the concentration of suspended minerals between 0.5 and I0 mg L-'. . . . 79 Table 3.5: Above-water remote sensing reflectance (R,(O+)) 3 nm band ratios used to estimate the concentration of chlorophyll between I and 5 ,ug
c'.
... ... . . ... ... ... .... . . . . . . , . . , . . . . . . 81Table 3.6: Above-water remote sensing reflectance, R,,(O+) I I nm band ratios used to estimate the concentration of chlorophyll between I and 5 ,ug
c'.
... . . . .... . . . .... . . .. . ... . ... . . . . . . . . . . . . ... . . 82Chapter
I
The use of remote sensing models in coastal and inland waters offers an unprecedented and non-invasive method to study biological, chemical and physical processes at a variety of spatial and temporal scales. Below follows a brief description of the history of remote sensing in the open ocean as well as coastal and inland environments. Definitions and concepts used in aquatic remote sensing are introduced to help to set this work into a broader context, and two separate papers are presented. A list of symbols used in this thesis is presented in Appendix 1.
1.1 A Brief History
In 1978, the American National Aeronautics and Space Administration (NASA) launched the Coastal Zone Color Scanner (CZCS), the first satellite sensor designed to monitor ocean colour. The goal at the time was to measure water-leaving radiance for a limited number of wavebands in the visible spectrum, in order to retrieve the concentrations of phytoplankton pigments in the near-surface layers of the ocean. Despite the name, the CZCS's main focus was to determine the spatial distribution of chlorophyll, the main photosynthetic pigment in phytoplankton, in the open ocean, not in coastal or other optically-complex environments, where the presence of optically active components other than chlorophyll may mask or modify the water-leaving signal (IOCCG, 2000). The data collected from the CZCS over its lifespan, from 1978 to 1986, is still in use and has contributed greatly to the
knowledge of how chlorophyll is distributed in the ocean (Strombeck, 2001). Today, sensors such as SeaWiFS (Sea-viewing Wide Field-of-view Sensor), MODIS
(Moderate Resolution Imaging Spectroradiometer), and MERIS (Medium Resolution Imaging Spectrometer Instrument) have replaced the CZCS as sources of satellite remote sensing data for the ocean.
Remote sensing of coastal and inland waters was somewhat slower to develop than ocean studies as there were no specially-designed satellite sensors for these waters (Dekker, 1993). Early studies were performed with the Multispectral Scanner (MSS) and Thematic Mapper (TM) sensors onboard Landsat satellites, although they
were primarily designed for terrestrial applications. By the mid-1 990s, airborne multispectral sensors became commercially available (Strombeck, 2001). At about the same time, optical field equipment for oceanographic and limnological use also became available, and researchers such as Bukata et al. (1979), Kirk (1984) and Dekker (1993) established that the ocean optical models developed by Gordon et al. (1 988), Morel and Prieur (1 977), and Prieur and Sathyendranath (1 98 I), among others, were also applicable to coastal and fresh waters.
The success of the CZCS, SeaWiFs, MODIS, and MERIS sensors can be attributed to the use of traditional remote sensing techniques in combination with theoretical aquatic optical models developed by researchers such as Gordon (Gordon et al., 1975; Gordon and McCluney, 1975), Bukata et al. (1 978), Morel (Morel and Gordon, 1980), and Bricaud et al. (1 98 1). Prior to its use in remote sensing
applications of ocean waters, aquatic optics had been an independent study from the middle of the 2 0 ~ Century, with early work by Jerlov (1 976), Preisendorfer
(summarized in Preisendorfer, 1976) and Mobley (i.e. 1994).
Recently, the availability of hyperspectral in-situ instruments has resulted in a 'quiet revolution' in remote sensing of natural waters (IOCCG, 2000). The increased spectral and radiometric information available from these sensors can be used to develop and calibrate very high resolution spectral and spatial bio-optical models in optically-complex coastal and inland waters (Dickey, 2004). Advances in computer technology in the last decade have enabled more rapid processing of hyperspectral data and greatly improved the storing and archiving capability of the large, and often difficult to manage data sets (Chang et al., 2004). These high spectral and spatial resolution bio-optical models can be used for regional scale studies using data
acquired from new hyperspectral satellites, such as NASA's Hyperion sensor onboard the Earth Observing-1 (EO-1) spacecraft launched in 2001, as well as proposed satellites, such as the Canadian Space Agency's Hyperspectral Environment and Resource Observer (HERO).
Remotely-sensed products, used in combination with regionally calibrated in- situ bio-optical models and algorithms, are ideally suited to cover the broad range of space and time scales associated with coastal applications. These products can
3
provide the ability to track turbid river plumes, or study the re-suspension of bottom sediments in shallow waters due to tidal action, wind and waves (Warrick et al., 2004). They can also be used to measure the location and timing of phytoplankton blooms (Millie et al., 1997), determine the source of CDOM and its distribution (Cobel et al., 2004) and help to understand regional biogeochemical systems.
1.2 Concepts and Definitions
The following section provides an introduction to the basic concepts and terminology encountered in remote sensing of the aquatic environment. The topics covered include an introduction to the concepts and definitions used in aquatic optics as well as the components of absorption and scattering in the water column.
In this thesis, remote sensing is defined as the passive measurement of
electromagnetic radiation of wavelengths between 400 - 700 nm. This corresponds to the visible spectrum, ranging from blue (400 nm) over green and yellow, to red (700 nm). The constraint on the lower end of the spectrum is due to the fact that the intensity of solar radiation below 400 nm is quite low, and the signal measured by the remote sensor at wavelengths less than 400 nm is often overwhelmed by the noise of the instrument. Beyond 700 nm, the absorption of water, which is very strong in red and near infra-red light, reduces the signal reaching the remote sensor and the signal is also overwhelmed by the noise of the instrument.
When direct sun light or scattered sky light penetrates the airlwater interface, it is either absorbed or scattered by the water itself and the organic and inorganic water constituents. The additive result of these absorption and scattering interactions is the removal of energy from the beam of light as it propagates through the water column. This is referred to as attenuation. In particular, the optical properties of natural waters are influenced by water itself and the distribution and abundance of: phytoplankton and its co-varying detritus; coloured dissolved organic matter (CDOM); and, inorganic suspended matter. Reflection of light from the bottom and scattered sky light due to the atmosphere may also be measured by the remote sensor. In this study, the waters are all optically deep and bottom reflectance did not influence the water- leaving signal. Measurements made above-water are corrected for the small parcel of
4 atmosphere between the sensor, situated
-
2.5 metres above the water surface, and the water surface.1.2.1 Radiance
Light that originates from a point source, such as a fixed star, is observed to propagate radially outward in all directions. A measure of the radiant flux of this light source in a particular direction (d@) per unit solid angle (dR) is the radiant intensity (I), as illustrated in Figure 1.1 (Bukata et al, 1995). This radiant intensity is described by :
I = d@ / dR (Watts sr-') (1.1)
Radiant intensity can also describe incoming radiation on the earth's surface. In this case, the solid angle is used to define the containment of incoming rather than outgoing photons.
Figure 1.1: Radiant intensity from a point source, where zenith angle (Q), azimuthal angle
(@), and unit solid angle (dn) are identified (Bukata et al., 1995).
Radiance (L (A, 0,q, sr-I)) is defined as the radiant power in a beam per unit solid angle per unit area perpendicular to the beam per unit wavelength interval:
L (A, 0,q, sr-') = d2@ / dA di2 (Watts m-2 sr-' nm-') (1.2) where dA is the area perpendicular to specified direction of photon propagation. The measure of radiance in a surface areas that are not perpendicular to the selected direction of photon propagation is illustrated in Figure 1.2.
azimuthal angle (cp), that is not perpendicular to the direction of photon propagation. The radiance emitted from a light source along direction D is expressed by:
L (h,0,cp, sr-') = d2 (D / dS cos0 dil ;where, dS cos0 is equal to dA when dS is perpendicular to direction of photon propagation (Bukata et al., 1995).
1.2.2 Irradiance
Irradiance (E(h)) is the radiant flux per unit area per unit wavelength interval (Bukata et al, 1995):
E(h) = d @ / dS (Watts m-2 nm-') (1.3)
Irradiance is comprised of all the radiant flux impinging upon a selected point in the radiative field. As a result, the directional aspects attributable to radiance are not attributable to irradiance.
Downwelling irradiance (Ed@)) is the irradiance in the hemisphere above the horizontal plane containing the point of interest. Ed@) is the downwelling light recorded by an upward looking radiometer, equipped with a cosine response.
Upwelling irradiance (E,(h)) is the irradiance in the hemisphere below the horizontal plane containing the point of interest. Eu(h) is the upwelling light recorded by a downward looking radiometer, equipped with a cosine response.
Scalar irradiance (E,(h)) is described by the sum of upwelling and downwelling irradiance:
E&) = Ed@) + ELI@) (1.4)
When considering the total radiant intensity at a point in space independent of its arrival direction, one is regarding radiance from each direction equally. This is equivalent to equating the infinitesimal area dS, the actual area upon which the
6 directional radiance is impinging, to the infinitesimal area dA, the area of dS that would be projected to a radiant flux passing perpendicular through it. This equal treatment of radiances results in irradiance that is termed scalar irradiance (Bukata et al., 1995).
1.2.3 Remote Sensing Reflectance
Remote Sensing Reflectance (Rrs(h, sr-I)) is the ratio of upwelling radiance at a point to downwelling irradiance at that point:
Rrs(h, sr-') = L (h, 8, cp, sr-') / Ed@) (1 -5) The remote sensing reflectance ratio is free of the magnitude changes associated with changing atmospheric conditions (i.e. passing clouds, time of year) and can be easily compared among studies.
1.2.4 Inherent and Apparent Optical Properties
There are two types of optical properties of water: the inherent optical
properties (IOPs) and the apparent optical properties (AOPs). IOPs depend solely on the type and the concentration of constituents present in the water and are
independent of the ambient light field within the water (Kirk, 1994). Absorption (a@)), scattering (b(h)), and attenuation (c(h)) are IOPs. The AOPs depend on the type and the concentration of constituents present in the water as well as the
geometric structure of the ambient light field. Remote sensing applications rely upon the relationships that link the apparent optical properties of a water body, such as remote sensing reflectance, with its inherent optical properties. Radiative transfer theory provides the connection between the AOPs and the IOPs. It describes the rate of change with distance of the radiance in a collimated beam at a specified location, direction, and wavelength. The radiative transfer equations account for all losses ( e g , due to absorption and scattering out of the beam) and gains (e.g., by emission or scattering into the beam) (McCormick and Mobley, 2001) in the collimated beam.
1.3 Major Light Absorbing Components of the Aquatic System
Total absorption in the water column is the result of the additive contributions due to water itself (a,), coloured dissolved organic matter (k,,,), photosyntheticaIIy active pigments (a,& and suspended minerals (asM) (Jerlov, 1976):
1.3.1 Absorption due to Water
Pure water appears as a blue liquid because it absorbs very weakly in the blue (400 - 500 nm) and green (500 - 600 nm) regions of the spectrum. At wavelengths
greater than 550 nm the absorption increases. The harmonics of 0 - H stretch of the water molecule result in absorption peaks at 5 14 nm and 604 nm (Kirk, 1986), as illustrated in Figure 1.3.
400 450 500 550 600 650 700 Wavelength (nm)
Figure 1.3: Absorption spectrum for water, data from Pope and Fry (1997). 1.3.2 Absorption due to Coloured Dissolved Organic Matter
The decomposition of plant tissue by microbial action results in a complex group of compounds, referred to by many names, including; humic substances, yellow substance, gelbstoffe, gelbstuff, dissolved yellow color, and coloured
dissolved organic material (Kirk, 1994). In this thesis, these substances are referred to as coloured dissolved organic material (CDOM). The majority of CDOM in coastal and inland waters is derived from rivers containing organic materials leached from soils (Boss et al., 2001; Coble et al., 2004). This organic material is added to natural waters from rainfall that drains through the soil, into rivers and lakes, extracting humic substances that have a yellow andlor brown colour (Kirk, 1994). In addition to
8
rivers, CDOM is also produced in the ocean by release of organic molecules from organisms during microbial degradation, excretion and grazing (Coble et al., 2004). The absorption spectrum of CDOM decreases with increasing wavelength within the 350 - 700 nm region (Jerlov, 1 W6), as illustrated in Figure 1.4.
350 400 450 500 550 600 650 700
Wavelength (nm)
Figure 1.4: Absorption spectrum for CDOM, where the mean slope is equal to 0.024 nm-' and absorption at 440 nm = 0.25 m-'(Bricaud et al., 1981).
Bircaud et al. (1981) describe the shape of the CDOM absorption spectrum as:
a@> = a(&) e -SCDOM (A - LO) (1 -7) where a(h) is the absorption coefficient at any wavelength, a@,) is the absorption coefficient at a reference wavelength, and SCDoM is the mean slope of the absorption spectrum. The relative constancy of shape of the absorption spectrum means we can characterize the CDOM of natural waters in terms of the absorption coefficient at a single wavelength. The 440 nm wavelength is routinely used as absorption is reasonably high and it corresponds approximately with the peak of photosynthetic pigment absorption band in the blue spectrum, allowing for comparison amongst different waters (Kirk, 1994). A lower mean S C D ~ ~ suggests freshly produced material and S C ~ o ~ increases as the material is chemically degraded and CDOM becomes increasingly refractory (Schofield et al., 2004).
Absorption due to suspended minerals is determined from the specific absorption coefficient for suspended minerals (a*,, (A)) (m-' per mg L-'), and the concentration of suspended minerals.
as,@) = a*,, (1) [SM] (1.8)
The coefficient is dependent on particle shape, particle size distribution and refractive index of the suspended particles of a watershed (Bukata et al., 1995). In general, a*&) decreases with increasing wavelength to a minimum at 580nm at which point begins to increase to 700 nm (Gallie and Murtha, 1992; Bukata et al., 1995), as illustrated in Figure 1.5.
400 450 500 550 600 650 700
Wavelength (nm)
Figure 1.5: Absorption spectrum for suspended minerals. The concentration of suspended minerals = 1 mg L-1 and the coefficient a*sM
(A)
is derived from Gallie and Murtha (1992).1.3.4 Absorption due to Phytoplankton
Absorption due to phytoplankton (aph(h)) is determined from the absorption coefficient of the main photosynthetic pigment, chlorophyll a. According to Bricaud et al. (1995), aPh(A) is calculated as the product of the chlorophyll a specific
phytoplankton absorption coefficient (axph@)) (m*' per ug L-l) and the concentration of chlorophyll-a and pheopigments (Q L-'), as illustrated in Figure 1.6:
Wavelength (nm)
Figure 1.6: Absorption spectrum for phytoplankton. The concentration of chlorophyll is equal to 1 Q L-' and the chlorophyll a specific phytoplankton absorption coefficient a*&) from Bricaud et al. (1995).
1.3.5 The Package Effect
The absorption of light due to phytoplankton is complicated by the fact that chlorophyll a and other accessory pigments are contained within the phytoplankton, in discrete 'packages'. These discrete pigment packages are contained within
chloroplasts, within cells, within cell colonies. As a result, phytoplankton absorption is influenced by the total amount of the photosynthetic pigments present, as well as the size and shape of algal cells or colonies within which the pigments are located (Kirk, 1994). Taken together, these factors result in lower specific absorption, because the pigment's ability to collect light from the prevailing field is lessened. Kirk (1994) describes this as the 'package effect'. The package effect describes the difference seen between a spectrum of intact phytoplankton cells in a volume of water versus the same corresponding volume of water where the cells have been disrupted, or broken open, and the pigments are free in solution (Figure 1.7). The package effect is greatest when absorption is strongest. As a result, the absorption peaks at
-
440 nm and 680 nm are reduced much more than the region between 550 to 650 nm.-
.
-
Intact Cells400 450 500 550 600 650 700 Wavelength (nm)
Figure 1.7: The Package Effect. The absorbance spectrum of intact cells is higher than that of disrupted cells (ultrasonicated cells) (Kirk, 1994).
In the following discussion, details of the package effect are described. The 'suspension' refers to the medium that contains intact cells and is representative of the natural environment. The 'solution' is the medium in which the pigments are no longer contained within the cell, as the cells have been disrupted via sonication in the laboratory. To illustrate the package effect, consider the following Figure (1.8):
N particles 1 m3
Monochromatic light
jth particle
The suspension
+-- 1 meter pathlength +
jth particle contams p~grnents withm
Figure 1.8: Absorption of light by a particle in suspension with photosynthetic pigments. In the text, the jth particle absorbs the portion, sjAj, of incident light (schematic adapted from Kirk, 1994).
12 In suspension, the jth particle, would, in the absence of other particles or membranes, absorb a proportion (sjAj) of the light that is incident on the suspension. Where sj is the projected are of the particle (m2) in the direction of the light beam and Aj is the fraction of light incident which is absorbed, known as particle absorptance. Therefore, sjAj (m2) has the dimensions of area and is the absorption cross-section of the particle (Kirk, 1994). The absorption coefficient for the suspension is expressed as (Kirk, 1994):
a = -1 I r x Ln(1-A) (1.10)
where a is the absorption coefficient (m-'), r is depth in the medium (m) and A is the absorptance. Absorptance is described as a ratio of the radiant flux absorbed (CDabS) to the radiant flux incident on a system (ainc):
A = Qabs Qinc (1.11)
The absorption coefficient of the suspension due to the jth particle is described by:
a = Ln 1 I (1 - sjAj) (1.12)
Since sjAj is small, we can approximate a
-
sjAj (Kirk, 1994). Assuming the absorbance, or absorption coefficient, of the suspension is equal to the sum of the absorbance, or absorption coefficients, due to all the individual particles then the absorption coefficient of the suspension (a,,,) is (Kirk, 1994):N
a,,, =
C
sjAj = NsAJ = 1
where sA is the mean value of the product of the projected area and the particle absorptance for all the particles in suspension, (i.e. the average absorption cross- section).
To measure pigments in the laboratory, the particles are disrupted via sonication and the pigment is no longer 'packaged' but is uniformly distributed among the solution. The absorption coefficient due to this dispersed pigment (a,,,) is (Kirk, 1994):
where C is the pigment concentration (mg m-3 or Q L-I), v is the volume of the particles (m3), and y is the specific absorbance coefficient due to pigment at a specific wavelength (m-' per unit concentration). The package effect becomes more marked as absorption by individual particles increases. For instance, if the pigments begin to absorb more strongly in the solution both C and y can be increased in value
indefinitely by increasing the pigment concentration and changing to a more intensely absorbing wavelength. As C and y increase, so too does A. However, since A is a fraction of the incident light that is absorbed it can never be greater than one and therefore cannot increase in proportion to the increasing C and y. It is this
phenomenon that explains why the peaks are more effected (in the spectrum) than the troughs (Figure 1.7). It is the dependence of the absorption spectrum of a suspension on the fraction absorption per particle which accounts for a flattening of the spectrum and the lowered specific absorption per unit pigment concentration (Kirk, 1994).
1.3.6 Total Absorption
The total absorption coefficient is the sum of the individual absorption
coefficients of the light absorbing components. The major components into which the light-absorbing material of the system has been divided; water itself, CDOM,
suspended minerals, and phytoplankton can all be substantial light absorbers (Kirk, 1994). Figure 1.9 illustrates the total absorption spectrum for natural waters.
1.0 T o t a l Absorption Water -. - -. . CDOM
-
- - - - Phytoplankton Suspended Minerals 400 450 500 550 600 650 700 Wavelength (nm)Figure 1.9: Total absorption spectrum for natural waters (chlorophyll = 1 Eg L-' ; suspended minerals = 1 mg L-'; absorption by CDOM at 440 nm = 0.20 nm) with the spectra of
14
1.4 Major Light Scattering Components of the Aquatic System
Absorption of light in the aquatic environment is considered a linear loss of radiant energy. Scattering, on the other hand, has a directional dependence. Scattering processes are specified in terms of the scattering coefficient (b(h)) (m-I), and the normalized volume scattering function @(0)), which characterizes the angular distribution of single-event scattering (Kirk, 1984). The scattering coefficient
describes the fraction of radiant energy that is scattered from a beam per unit distance (m-I). The volume scattering function depends on the light scattering behaviour of particles relative to impinging wavelength (Dekker, 1993). For particles larger than the wavelength of light, Mie theory predicts that most of the scattering is in the forward direction within small angles of the initial direction of the light, as illustrated in Figure I. 10 (Dekker, 1993; Bukata et al., 1995). In natural waters, scattering is predominantly due to particles larger than the wavelength and as a result, the shape of the volume scattering function is such that there is much more scattering in a forward direction than in a backward direction (Kirk, 1984; Dekker, 1993). The volume scattering function has the same shape for most natural waters and therefore does not need to be determined routinely. A data set for P(0) obtained by Petzold (1 972) for coastal seawater in San Diego Harbour is regarded as applicable to all except very clear oceanic waters.
Incident direction of light on particle
particle
Total scattering in the water column is a sum of forward scattering (bf) and backward scattering (bb) light. For remote sensing applications we are concerned with the backward scattering component as it defines the scattering of light into the
hemisphere trailing the downward incident flux, i.e. upwards towards the water surface (Dekker 1993). Total backscattering in the water column is a consequence of backscattering due to pure water (bbw), phytoplankton (bbph), and suspended minerals (bbSM). CDOM is assumed to be a true absorber and does not contribute to scattering. Total backscattering is described by:
1.4.1 Backscattering due to Water
The backscattering due to water is determined from the scattering coefficients of pure water (b,) according to Buiteveld (1994). Assuming that the backscattering efficiency of pure water is 50% (Bukata et al., 1995; Sathyendranath et al., 1989), bbw is expressed as (Figure 1.1 I):
bbw(h)= 0.5 x bw(h)
400 450 500 550 600 650 700
Wavelength (nm)
1.4.2 Backscaftering due to Suspended Minerals
Scattering by inorganic suspended minerals (bs~(h)) is described as the product of scattering at a reference wavelength and the spectral shape of particle scattering (Babin 2003b):
~ S M (h) = ~ S M (ho) ( 1 / ho) - 0'4 (1.17)
where h, is a reference wavelength (= 555 nm). The following assumptions are made about the nature of the suspended particles:
i) particles are homogeneous spheres subject to Mie scattering;
ii) the imaginary part of the index of refraction is non-absorbing (n' = 0); and iii) the particle size distribution follows a Junge distribution power law where
j = 3.4, j determines the shape of the distribution and a value of 3.4 produces a particle size distribution typical of coastal waters (Babin et al., 2003b).
Scattering at 555 nm is described as the product of the mass specific scattering coefficient [ ~ * s M (555) m2 g-l] and the concentration of suspended minerals ([SM] g m'3 or mg L-I) in the water:
bSM (555) = b * s ~ (555) [SM] (1.18)
The average b * s ~ (555) value reported by Babin et al. (2003b) for coastal waters is 0.5 1 m2 g-'. Scattering by inorganic suspended mineral particles is therefore described as:
bSM (1) = 0.51 X [SM] x (h / 555) -0.4 (1.19)
The backscattering efficiency of suspended minerals ranges from 0.2% to 4% (Twardowski et al., 2001; Boss et al., 2004) and is estimated to be 0.7% and 2.6% for the waters under consideration. Backscattering by suspended minerals is illustrated in Figure 1.12.
1--
- - -
-Freshwater - - - - --.Coastal Waters0.01 6
400 450 500 550 600 650 700
Wavelength (nm)
Figure 1.12: Backscattering spectrum due to suspended minerals (Babin et al., 2003 b).
1.4.3 Backscattering due to Phytoplankton
Total scattering due to phytoplankton is described as a function of the
concentration of chlorophyll ([chl]) and the spectral shape of phytoplankton scattering (Gordon and Morel, 1983; Sathyendranath et al., 1989):
where
h,
= 550 nm. Assuming the backscattering efficiency of phytoplankton is 0.5% (Sathyendranath et al., 1989) the backscattering due to phytoplankton is (Figure 1.13):400 450 500 550 600 650 700 Wavelength (nm)
Figure 1.13: Backscattering spectrum due to phytoplankton (Gordon and Morel, 1983; Sathyendranath et al., 1989).
The total backscattering coefficient is the sum of the backscattering coefficients of the individual scattering components:
bb(h> = 0.5 b,
+
[ o . o o ~ ~ ( o . 12[~h1]~'~~)(550/h)]+[bbs~ X 0.51 [SM] ( ~ 5 5 5 ) - O ' ~ ] (1 2 2 )The major components into which the scattering material of the system has been divided; water itself, phytoplankton and suspended minerals are illustrated in the total backscattering spectrum for natural waters (Figure 1.14).
0.008 -Total Backscattering Water h . - . . - . . Suspended Minerals .; 0.006 E Phytoplankton Y 400 450 500 550 600 650 700 Wavelength (nm)
Figure 1.14: Total backscattering spectrum for natural waters (chlorophyll = 1 Q L";
1.5
The beam attenuation coefficient describes the fraction of radiant energy removed from a beam of light per unit distance as it traverses an infinitesimal distance. Attenuation is due to both scattering and absorption. Apart from the small amount of light scattered back out of the water, attenuation of the photosynthetically available radiation (PAR, 400 - 700 nm) in water bodies is due to absorption, although the extent of this absorption within a given depth may be greatly amplified by scattering, which increases the average pathlength of the photons within that depth (Kirk, 1 994).
The radiant flux, expressed as irradiance or radiance, diminishes with depth in an approximately exponential manner, due to absorption and scattering processes. The attenuation of downwelling irradiance with depth is described by:
Ed@, z) = Ed@, 0) e -Kdz (1.23)
Ln Ed(h, z) = (-Kd x z )+ In Ed@, Om) (1.24) where Ed(h, z) and Ed@, Om) are the downwelling irradiance values at depth z and just below the surface, respectively. Kd is the vertical attenuation coefficient (m-') for
downwelling irradiance. Values of Kd are attributed to the waterbody itself and are used to compare one waterbody at a given time with another, or as a guide to changes in the optical character of a given waterbody with time. For example,
&
values remain representative of water bodies, independent of the time of day or the weather conditions, as long as the composition, and therefore the inherent optical properties, of the water remain unchanged (Kirk, 1986).1.6 Rationale for Study
Monitoring and maintaining water and ecological quality of coastal zones is a growing global concern. It is estimated that over 60% of the human population lives in the coastal zone (Pernetta and Milliman, 1995). Human activities in the coastal zone and catchment areas result in modified runoff and delivery patterns of nutrients and sediments to lakes and coastal waters (i.e. land clearing, agriculture, mining, and urban and/or industrial development). Coastal development can also lead to
20 suspension of sediments, and direct inputs of nutrients and pollutants (IOCCG, 2000). Due to the many impacts and uses of the coastal zone, coastal zone management has become a high priority for many nations (IOCCG, 2000).
Remote sensing applications for water quality in coastal and inland waters offer spatial and temporal coverage that is unattainable by other means and is ideally suited to cover the broad range of space and time scales associated with coastal applications. In order for remote sensing data to support water quality monitoring efforts and coastal decision making, standard and reliable algorithms are required. For example, estimating phytoplankton abundance is critical for critical for improved understanding and management of pelagic and benthic ecosystems, wild fisheries and aquaculture, and impacts of added nutrients (IOCCG, 2000). Sediment concentrations can also be estimated from remote sensing data. Knowledge of sediment distribution is important as many contaminants (organics and heavy metals) adsorb strongly to fine sediments, which act as a carrier and serve to remove contaminants from the water column (Kersten et al., 1994).
The assessment of water quality using remote sensing techniques provides reliable, consistent and affordable data for coastal zone managers and researchers. Data collected from satellite, airborne or moored sensors can provide important time series data sets for biogeochemical studies and climate change research.
1.7 Purpose of Study
The motivation for this study was to develop and calibrate a bio-optical model with hyperspectral resolution, 3 nm band widths over 400 - 700 nm range, in order to characterize the optical and geochemical properties of inland and coastal waters in southern Vancouver Island, British Columbia. This hyperspectral model was used to develop algorithms to retrieve the concentrations of suspended minerals and
chlorophyll from remote sensing reflectance spectra in the coastal waters. This is accomplished using a stepped approach:
Step 1 : In August 2003, radiometric data was acquired with hyperspectral
radiometers, both in- and above-water, at 3 nm band widths, contiguous across the visible spectrum, 400 - 700 nm, in Lake Cowichan and the coastal waters of
Cowichan Bay and Satellite Channel. Differences between remote sensing reflectance determined from the in- and above-water platforms were evaluated. This initial step was important in terms of quantifying the bias in deriving remote sensing reflectance from different hyperspectral platforms.
Step 2: The semi-analytical optical model predicts remote sensing reflectance as a function of the concentrations of chlorophyll and suspended minerals and the absorption of CDOM. The model was are calibrated for the coastal and inland waters. The model relates the inherent optical properties of absorption and backscattering to the apparent optical properties of upwelling radiance and
downwelling irradiance. In this approach, the water constituents were expressed in their specific absorption and scattering coefficients. The resolution of the optical models calibrated for the coastal and freshwater environments is 3 nm over 400 - 700
nm.
Step 3: To assess the result of changing concentrations of suspended minerals, chlorophyll and absorption due to CDOM on remote sensing reflectance spectra, sensitivity analyses were performed using the bio-optical model. In the analysis, one variable (e.g.. concentration of suspended minerals) was changed incrementally and the other variables were held constant. The output of the sensitivity analyses were remote sensing reflectance spectra that illustrate the spectral region where the greatest change occurred.
Step 4: Algorithms were developed to estimate the concentration of suspended minerals and chlorophyll using remote sensing data sets at 3 and 11 nm resolution. The development of the algorithms was informed by the results of the sensitivity analysis and correlation coefficients between the variable of interest, and remote sensing reflectance band ratios.
Chapter
2
Hyperspectral Remote Sensing of Suspended Minerals, Chlorophyll and CDOM in Coastal and Inland Waters, British Columbia, Canada 2.1 Introduction
The use of semi-analytical models in remote sensing of water quality offers an unprecedented and non-invasive method to study biological, chemical and physical processes at a variety of spatial and temporal scales. Semi-analytical models, calibrated with recently available hyperspectral radiometers with less than 5 nm resolution, are a significant improvement over models calibrated with multispectral instruments, where the resolution is often 20 nm or worse. In coastal and inland waters, these hyperspectral models can be used in regression analyses to successfully estimate the spatial and temporal distribution of suspended minerals, phytoplankton and coloured dissolved organic matter, and can be applied to regional data collected from airborne or satellite hyperspectral platforms. There are a number of advantages to using hyperspectral regional models as a tool in the assessment of water quality for resource management. For instance, once the model is developed and calibrated to the region of interest, it is not dependent on extensive on-going field measurements, the processing time is fast, and the models are relatively easy to use.
In coastal and inland waters, the colour of the water is influenced by the distribution and abundance of phytoplankton and its co-varying detritus as well as varying amounts of coloured dissolved organic matter (CDOM), and suspended minerals (Kirk, 1994). Solar radiation that penetrates the airlwater interface is either absorbed or scattered by these organic and inorganic water constituents, and the additive result of these absorption and scattering interactions is the removal of energy from the beam of light as it propagates through the water column. This is referred to as attenuation. Remote sensing reflectance is an apparent optical property that depends on the inherent optical properties of absorption and scattering as well as the ambient light field, and can be used to retrieve the concentrations of the water constituents. Remote sensing reflectance retrieval algorithms can be simple band ratios, such as the bluelgreen ratio for chlorophyll in open oceans, or more complex
relationships based on theoretically or statistically determined spectral band combinations.
In this study, the optical properties and geochemistry of inland and coastal waters of Vancouver Island, British Columbia, Canada are measured and modeled at very high spatial and spectral resolution in order to develop regional models for monitoring capabilities. The inherent optical properties of absorption and
backscattering are used to model subsurface remote sensing reflectance using a semi- analytical approach. In this approach, the water constituents and their specific
absorption and scattering coefficients are used to model the subsurface remote
sensing reflectance (&,(O-)). Rrs(O-) is then propagated through the airlwater interface to derive above-water remote sensing reflectance (Rrs(O+)), which is the important parameter for remote sensing applications that utilize airborne and satellite sensors. The difference in the modeled remote sensing reflectance (R& derived as a function of the water constituents, and the measured Rrs spectra is quantified.
For the retrieval of the water constituents to be successful, any biases in above-water remote sensing reflectance must be understood and incorporated into models. We accomplish this by comparing in-water and above-water measurements to guide the appropriate selection of parameters used to project remote sensing reflectance through the water-air interface. This is done by determining water leaving radiance (L,) and above-water remote sensing reflectance independently from separate in- and above-water platforms.
2.2 Model Approach
A schematic of the semi-analytical optical model is illustrated in Figure 2.1. Similar to studies by Sathyendranath et al. (1989,2004) and Pierson and Strombeck (200 I), R,,(O-) is used to relate the inherent optical properties, absorption and
backscattering, to the apparent optical properties of upwelling radiance (L,(O-,z)), and downwelling irradiance (Ed(O-,z )), at depth z. Monte Carlo simulations by Jerome et al. (1 996) established that for sun elevation angles of 15 Cto 89 Wrs(O-) can be described as a function of the ratio of coefficients of backscattering (bb) and absorption (a) within the water column:
I
I
Measured EsI
!t
Measured LATE, R,,(O-) MODEL/
l31
O iI
I
Measured Ed (z)I
! ! [ ' Y i I $ ! ! ! ABSORPTION BACKSCATTERINGi
MODEL MODEL ! f ! !! Measured a,,,, (440) Measured [SM]
! 1 8 i
! I I
Figure 2.1: Schematic diagram of the semi-analytical optical model. Solution of the
absorption and backscattering models requires inputs of measured concentrations of chlorophyll [chl], suspended minerals [SM] and the absorption of coloured dissolved organic matter at 440nm acDoM(440) into the models to determine the total absorption and backscattering. The subsurface remote sensing reflectance &(O-) is a function of the total absorption and backscattering. Rr,(O-) is also determined by measuring the upwelling radiance L, and downwelling irradiance Ed in the water column. Rrs(O-) is propagated through the aidwater interface to derive above-water Rrs(O+).
through the aidwater interface according to Jerome et al. (1996):
Rrs(O+, 8) = Rrs(0-, 8') [(I - p') (1 - p(el,@)) 1 q21 (2
4
where p' is a proportionality factor that describes the reflectivity of the water surface, it is a function of wind speed, viewing angle, sun zenith angle (see Table 2.1 and discussion for above-water determination of water-leaving radiance (L,); p(8',8) is the internal surface reflectivity for an incident angle 8' refracted into radiance at the observing zenith angle 8 above the water surface (the in-water angle 8' is related to in-air solar zenith angle 8 through Snell's law where 8' is calculated as
[Or = sine1(sin0 /1.333)]; and q is the fresnel refractive index of water and is 1.345 and 1.333 for seawater and freshwater respectively. Following Jerome et al. (1988):
p(8',8) = 0.271
+
(0.249 / cos 8') (2.3) In the absorption and backscattering models, the water constituents are expressed in their specific (per unit measure) absorption and scattering coefficients. Solution of the absorption and backscattering models requires inputs of measured concentrations of suspended minerals, [SM], chlorophyll, [chl], and absorption due to CDOM into the models to determine the total absorption and backscattering. The models are calibrated by tuning the models to Rrs(O-) spectra measured in-water.Table 2.1: Sample site information and illumination conditions at the time of sampling. The numbers for the sample sites in the coastal environment refer to the sample numbers on Figure 2.3. Location Illumination Geometry Site Name Sun Date Latitude (N) Longitude (W) Time Wind Speed elevation Sampled m i1 P' 0 West Lake Cowichan (inland) 07/27/03 48E53' 16" 124018' 30" 1 1:30 <1.0 44 0 0.023 East Lake Cowichan (inland) 07/27/03 48 M 1 ' 44" 124 01 0' 3 1 " 14: 15 2.7 40 0 0.028 1 - Cowichan Bay (coastal) 07/28/03 48l35' 08" 123 iD7' 07" 14:56 3.5 44 0 0.033 2 - Separation Pt. (coastal) 07/28/03 48044' 13'' 123 a34' 13" 12:55 1.3 390 0.027 3 - Satellite Channel (coastal) 07/28/03 48E43' 40" 123 U32' 25" 12:lO 1.8 41 17 0.041 4 - Cape Kappel (coastal) 07/28/03 48E42' 46" 123 LQ9' 54" 1 1:05 1 .O 47 0 0.035 5 - Isabella Lt. (coastal) 07/28/03 48Dl2' 09" 123 E7' 27" 0950 < 1.0 57 0 0.022
Jerlov (1 976) describes total absorption in the water column as the result of the additive contributions due to pure water (a,), coloured dissolved organic matter (acDoM), photosynthetically active pigments (aph), and suspended minerals (a,,). The absorption model incorporates these individual components of total absorption:
TOTAL = aw + ~ C D O M + aph + ~ S M Qm4)
This model differs from the absorption model of Sathyendranath et al. (2004) for the coastal waters of British Columbia as it treats the absorption due to CDOM and suspended minerals separately. In our absorption model, the values for the absorption due to pure water are from Pope and Fry (1997).
Absorption by CDOM is highest in the shorter wavelengths and decreases exponentially with increasing wavelength. Bricaud et al. (1981) describe CDOM
where k0, (&) is the absorption of CDOM at a reference wavelength
(&=
440 nm) and So,, (nm-') is the spectral slope of the acDoM(l) versus wavelength. The presence of CDOM in the water results in reduced values of Rrs in the spectral region 400 -550 nm, illustrated in Figure 2.2. 1.0 T o t a l Absorption Water CDOM Suspended Minerals -Total Backscattering W a t e r . . . Suspended Minerals - - - - Phytoplankton 400 450 500 550 600 650 700 400 450 500 550 600 650 700 Wavelength (nm) Wavelength (nm)
Figure 2.2: Absorption and backscattering spectra for productive coastal waters. The concentration of chlorophyll = 1.0 Qg L-1; suspended minerals = 1.0 mg L-1 ; absorption due to CDOM at 440 nrn = 0.14 m-1; and the mean slope of CDOM absorption = 0.024 nm-'.
28 Absorption due to phytoplankton is determined from the chlorophyll-a
specific phytoplankton absorption coefficient (m-' per pg L-I), and the concentration of chlorophyll-a and pheopigments ([chl]) (@ L-'). It is calculated according to Bricaud et al. (1995):
aph(V = a*,h(V [chll (2.6)
This model makes use of the data set presented in Bricaud et al. (1 995) for a*,h(h), which includes 8 15 spectra and covers concentrations of [chl] from 0.02 to 25 Qj L-'.
The parameterization of a*,h(h) is:
a*,h(h) = A(h) x [chl] -B@) (2.7)
where A and B are positive wavelength dependent parameters, based on empirical relationships between a*,h(h) and concentration of [chl], that were derived by least squares fitting to power functions. Babin et al. (2003a) evaluated a,h spectra collected from 350 coastal sites against Bricaud et al.'s (1995) a*,h parameterization. They found that overall most of their aph values were contained within the confidence intervals of a*,h versus [chl] statistics established by Bricaud et al. (1995). However, departure from the a*,h parameterization is expected as phytoplankton absorption is influenced by changes in cell size and in the composition and degree of packaging of pigments, termed the package effect (Kirk, 1994). These influences result in
decreasing a*ph(h) values for increasing concentrations of [chl]. Stramski et al. (2001) and Ciotti et al. (2002) describe further departures from the a*,&) spectra due to different size classes and species composition of the planktonic community. In general, when phytoplankton abundance increases, larger size classes are added to a background of smaller cells. As a result, more than 80% of the spectral variability in
can be explained by different size class specification. Despite these limitations we are using the a*,h(h) parameterization to approximate the contribution of
phytoplankton absorption in the model. The presence of phytoplankton in the water results in reduced values of Rrs in the spectral regions where absorption due to chlorophyll is at a maximum, 400 - 500 nm and 650 - 700 nm, illustrated in Figure
Absorption due to suspended minerals is determined from the specific absorption coefficient for suspended minerals (m-' per mg L"), derived from data included in Gallie and Murtha (1992) and the concentration of suspended minerals. It is calculated as:
asM(h) = a*sM (1) [SM] (2.8)
The coefficient a*,,
(A)
is dependent on particle shape, particle size distribution and refractive index of the suspended particles of a watershed, as discussed in Bukata et al., (1995). The presence of suspended minerals in the water results in a minor decrease in a*,,(h) with increasing wavelength to a minimum at 580 nm at which point a*,,@) begins a slight increase to 700 nm, illustrated in Figure 2.2.The absorption model is represented by substitution of equations (2.5) to (2.8) into equation (4):
TOTAL = aw + ~ C D O M (440) e [-ScD0M(h-440)1
+
a*ph [ ~ h l ]+
a*sM [SM] (2.9) 2.2.2 Backscattering ModelTotal scattering (b) (m-I), in the water column is a sum of forward scattering (bf) and backward scattering (bb) light. For remote sensing applications we are
concerned with the backward scattering component as it defines the scattering of light into the hemisphere trailing the downward incident flux, i.e. upwards towards the water surface. Total bb in the water column is a consequence of backscattering due to pure water (bbw), phytoplankton (bbph), and suspended minerals (bbSM). CDOM is assumed to be a true absorber and is not considered in the backscattering model. The three component backscattering model is:
bb = bbw + bbph + b b s ~ (2.10)
The backscattering due to water was determined from the scattering
coefficients of pure water according to Buiteveld (1994). According to Bukata et al. (1995), the backscattering efficiency of pure water is 50%. Therefore, bbw is
3 0 Total scattering due to phytoplankton is described as a function of the
concentration of chlorophyll ([chl]) and the spectral shape of phytoplankton scattering (Gordon and Morel, 1983; Sathyendranath et al., 1989):
bph = (0.12 [ c ~ ~ I o . ~ ~ ) x ( L I A,) (2.12) where
h,
= 550 nm. Assuming the backscattering efficiency of phytoplankton is 0.5% (Sathyendranath et al., 1989) the backscattering due to phytoplankton is:Scattering by inorganic suspended minerals (bsM(h)) is described as the product of scattering at a reference wavelength and the spectral shape of particle scattering (Babin 2003b):
where h, is a reference wavelength (= 555 nm). The following assumptions are made about the nature of the suspended particles:
iv) particles are homogeneous spheres subject to Mie scattering;
v) the imaginary part of the index of refraction is non-absorbing (n' = 0); and vi) the particle size distribution follows a Junge distribution power law where
j = 3.4. j determines the shape of the distribution and a value of 3.4 produces a particle size distribution typical of coastal waters (Babin et al., 2003b).
Scattering at 555 nm is described as the product of the mass specific scattering coefficient [b*SM (555) m2 g-l] and the concentration of suspended minerals ([SM] g m-3 or mg L") in the water:
