View Foreoptic Lens

ABSTRACT

JUDICE, TAYLOR J., M.S. DECEMBER 2019 GEOLOGY

DETECTING COLOR-PRODUCING PIGMENTS IN THE INDIAN RIVER LAGOON

BY REMOTE SENSING

(131 pp.)

Thesis Advisor: Joseph D. Ortiz

The Indian River Lagoon (IRL) is a coastal marine estuary which encompasses over

250 km of the eastern Florida coastline. While providing refuge for diverse groups of marine and bird life, this sanctuary also generates an estimated $7.6 billion annually to

Florida’s economy through fisheries, tourism, and other industries. Over the last several decades, various types of harmful algal blooms, such as that caused by Brown Tide species

Aureoumbra lagunensis, have become more frequent across the region due to eutrophication of estuary waters from fertilizer and septic runoff, as well as from increased major storm events. Emerging satellite-based remote sensing techniques, such as the varimax-rotated, principal component analysis (VPCA) method, decomposes the integrated spectral signature from optically complex water into independent component spectra, which are identified with a library of known spectral constituents. Coupling in- situ cells counts, water-quality monitoring systems, and hyperspectral spectroradiometer reflectance measurements from June 29-30, 2018 for validation, this research addresses whether the VPCA technique applied to the Sentinel-3A Ocean Land Colour Imager

(OLCI) imagery can detect A. lagunensis constituents in optically complex waters.

Following the component validation, we conclude this investigation by discerning the detection limit of Brown Tide VPCA constituents with respect to the spatial frequency of

Chlorophyll-a and Ochrophyta concentrations. Next, we produced a time-series of 10 images for the IRL from August 1, 2017, to November 21, 2018, to determine how constituents of Brown Tide related spectra vary with seasonal fluctuations in these water- quality parameters. This study has shown the detection limits of Brown Tide constituents using the VPCA spectral decomposition method to be less than 80 μg/L of Chlorophyll-a.

Furthermore, our time-series observations of VPCA spatial variability suggests a primary potential source of nutrient pollution causing A. lagunensis growth to be centered within the Banana River region of the IRL. In conclusion, our findings indicate the VPCA satellite technique to be a transferable method for characterizing optically complex waters with harmful algal blooms.

DETECTING COLOR-PRODUCING PIGMENTS IN THE INDIAN RIVER LAGOON

BY REMOTE SENSING

A thesis submitted

to Kent State University in partial

fulfillment of the requirements for the

degree of Master of Science

Taylor Joseph Judice

December 2019

Except for previously published materials

Thesis written by

Taylor Joseph Judice

B.S., Louisiana State University, USA, 2015

M.S., Kent State University, USA, 2019

Approved by

______, Advisor, Master Thesis Committee Dr. Joseph D. Ortiz

______, Chair, Department of Computer Science Dr. Daniel K. Holm

______, Dean, College of Arts and Sciences Dr. James L. Blank

TABLE OF CONTENTS

TABLE OF CONTENTS ...... V

LIST OF FIGURES ...... VIII

LIST OF TABLES ...... XIV

DEDICATION...... XVI

ACKNOWLEDGEMENTS ...... XVII

INTRODUCTION ...... 1

1.1 Background ...... 1

1.2 Research Objectives ...... 6

1.3 Research Plan/Methods ...... 7

1.3.1 Field Sampling ...... 7

1.3.2 Remote Sensing ...... 8

1.3.3 VPCA Spectral Decomposition ...... 10

1.4 Initial Landsat-8 and Field Results ...... 12

1.5 Discussion and Future Work ...... 15

1.6 References ...... 18

1.7 Figures ...... 24

1.8 Tables ...... 32

FIELD-VALIDATED DETECTION OF AUREOUMBRA

LAGUNENSIS BLOOM IN INDIAN RIVER LAGOON, FLORIDA USING

SENTINEL-3A OLCI AND GROUND-BASED HYPERSPECTRAL

SPECTRORADIOMETERS ...... 35

2.1 Abstract ...... 35

2.2 Introduction ...... 36

2.3 Methods ...... 40

2.3.1 Field Data ...... 40

2.3.2 Lab Data ...... 42

2.3.3 Remote Sensing Image Analysis and VPCA Spectral Decomposition ...... 44

2.3.4 Validation and Spectral Identification...... 46

2.4 Results ...... 47

2.4.1 VPCA Spectral Identifications ...... 47

2.4.2 Field Validation ...... 49

2.5 Discussion ...... 50

2.6 Conclusion ...... 54

2.7 References ...... 56

2.8 Figures ...... 62

2.9 Table ...... 67

VARIABILITY OF THE BROWN TIDE IN THE INDIAN RIVER

LAGOON FROM 2017 TO 2019 BASED ON SENTINEL-3A OLCI VPCA

SPECTRAL DECOMPOSITION ...... 68

3.1 Introduction ...... 68

3.2 Methods ...... 74

3.3 Results ...... 76

3.4 Discussion ...... 78

3.5 Conclusion ...... 79

3.6 References ...... 83

3.7 Figure ...... 90

3.8 Tables ...... 100

BIBLIOGRAPHY ...... 103

vii

LIST OF FIGURES

Figure 1.1: True color image of the (a) nine locations of Kilroy water monitoring

instruments along the east coast of Southern Florida in the Indian River Lagoon, (b)

zoomed-in images of the five Kilroy systems in the northern portion of the IRL ...... 24

Figure 1.2: VPCA component loadings from in field ASD hyperspectral reflectance

measurements after standardizing and computing forward stepwise linear regression

with the spectral library...... 25

Figure 1.3: Component loadings from Landsat-8 swath 016/040 of the IRL after

standardizing and computing a forward stepwise linear regression with the spectral

library...... 26

Figure 1.4: Results of cell count analysis, defining the division of microorganisms by the

percentage of total bulk volume. (a) Turnbull Creek, (b) Haulover Canal, (c) Barge

Canal, (d) Dragon’s Point, (e) Syke’s Creek...... 27

Figure 1.5: Results of cell count analysis, defining the division of microorganisms by the

percentage of total density. (a) Turnbull Creek, (b) Haulover Canal, (c) Barge

Canal, (d) Dragon’s Point, (e) Syke’s Creek...... 28

Figure 1.6: Graphs showing the average measurement from Kilroy water monitoring

system for July 28, 2017 (“Full series” represents the average values that are

available at the Kilroys from 8:00 am to 3:00 pm)...... 28

Figure 1.7: Landsat-8 VPCA imagery of Northern IRL (July 28, 2017). VPCA (a)

component 1 has positive correlation with peridinin, (b) component 2 has a positive viii

correlation with phaeophorbide-b and negative with chlorophyll-a + cyanophyta, (c)

component 3 has a negative correlation bacillariophyceae and positive with illite,

(d) component 4 has a positive correlation with dinophyta and negative correlation

with fucoxanthin. Warmer colors indicate higher correlation with reference spectra;

colder colors represent lower correlation with reference spectra...... 29

Figure 1.8: Landsat-8 VPCA imagery of Northern IRL (July 28, 2017). VPCA (a)

component 1 has positive correlation with myxoxanthophyll, (b) component 2 has a

negative correlation with bacillariophyceae, (c) component 3 has a positive with

phycocyanin, (d) component 4 has a positive correlation with cyanobacteria.

Warmer colors indicate higher correlation with reference spectra; colder colors

represent lower correlation with reference spectra...... 31

Figure 2.1: Location map of the Northern Indian River Lagoon. (a) True color composite

image from Landsat-8 OLI acquired on November 17, 2017. (b) Location of in-situ

measurements obtained for both cell count analysis, hyperspectral ASD

measurements, and Secchi depths on June 29th and 30th of 2018. Sites labeled as

Turnbull, Haulover, Barge, Sykes, and Dragon Point are locations of piling-mounted

Kilroy instruments for water measurements...... 62

Figure 2.2: Indian River Lagoon spectral (a-d) and spatial distribution maps (e-h) of

component loadings from June 28, 2018 swath from Sentinel-3A OLCI. In all images

warm colors indicate higher association of the pixels with the spectral constituent

mixture represented by this pattern, while cool colors indicate a lower association.

...... 63

Figure 2.3: Validation graphs showing spectral signature of the A. lagunensis culture

compared with VPCA component loadings for both satellite and field measured

datasets at the 11-band resolution of Sentinel-3A OLCI (a). Comparison between the

component scores of field spectra and Sentinel-3A that represent constituents of A.

lagunensis (b)...... 64

Figure 2.4: Validation graphs showing the comparison between Ochrophyta phylum from

cell count analysis with VPCA field spectra component scores (a) and Sentinel-3A

OLCI component scores (b)...... 64

Figure 2.5: SEM and EDS mapping images of sediment samples obtained from (a) Sykes

Creek and (b) NIRL 2 sample sites. Images (red) show the detection of Fe on grains

sizes < 30 μm. These SEM images indicate that iron-bearing or iron-coated minerals

or sediment grains are present to within deposits at the IRL and Banana River...... 65

Figure 2.6: Sykes Creek Kilroy measurements for Chlorophyll-a and blue/green algae

from May 18, 2017 to May 22, 2019 (http://api.kilroydata.org/public/). Analysis of

Sentinel-3A OLCI was obtained on June 28, 2018 (red arrow) when higher seasonal

counts of Chlorophyll-a and blue/green algae are measured...... 66

Figure 3.1: Location map of the Northern Indian River Lagoon. (a) True color composite

image from Landsat-8 OLI acquired on November 17, 2017. (b) Location of in-situ

measurements obtained for both cell count analysis and Kilroy hydrological

parameters for VPCA validation in Chapter 2...... 90

Figure 3.2: (a) Image of one Kilroy instrument suite (Thosteson et al., 2009), (b) mounted

on submersed piers along five along the northern IRL...... 91

Figure 3.3: Sykes Creek Kilroy measurements for Chlorophyll-a and blue/green algae

from August 31, 2017 to January 1, 2019 (http://api.kilroydata.org/public/). Analysis

of Sentinel-3A OLCI imagery obtained for all 10 images are indicated with red

points along the seasonal counts of Chlorophyll-a and blue/green algae measured

for the two-year span...... 92

Figure 3.4: VPC Pattern averages with the combined constituent identified spectra,

produced from the VPCA spectral decomposition of all 10 OLCI images in the time

series. (a) Pattern A on average represents spectral constituents of + illite and α-

PEC. (b) Pattern B on average represents spectral constituents of + Aureoumbra

lagunensis, + goethite, and - Cryptophyta. (c) Pattern C on average represents

spectral constituents of + Aureoumbra lagunensis, + illite, and + chlorophyll-b. (d)

Pattern D on average represents spectral constituents of + allophycocyanin, and -

chlorophyll-b...... 93

Figure 3.5: Time series from August 1, 2017 to November 21, 2018 of 10 spatial

distribution maps for VPC loading Pattern A. VPC Pattern A, on average represents

spectral constituents of + illite and α-PEC. In all images warm colors indicate

higher association of the pixels with the spectral constituent mixture represented by

this pattern, while cool colors indicate a lower association...... 94

Figure 3.6: Time series from August 1, 2017 to November 21, 2018 of 10 spatial

distribution maps for VPC loading Pattern B. VPC Pattern B, on average represents

spectral constituents of + Aureoumbra lagunensis, + goethite, and - Cryptophyta. In

all images warm colors indicate higher association of the pixels with the spectral

constituent mixture represented by this pattern, while cool colors indicate a lower

association...... 95

Figure 3.7: Time series from August 1, 2017 to November 21, 2018 of 10 spatial

distribution maps for VPC loading Pattern C. VPC Pattern C, on average represents

spectral constituents of + Aureoumbra lagunensis, + illite, and + chlorophyll-b. In

all images warm colors indicate higher association of the pixels with the spectral

constituent mixture represented by this pattern, while cool colors indicate a lower

association...... 96

Figure 3.8: Time series from August 1, 2017 to November 21, 2018 of 6 spatial

distribution maps for VPC loading Pattern D. VPC Pattern D, on average represents

spectral constituents of + allophycocyanin, and – chlorophyll-b. In all images warm

colors indicate higher association of the pixels with the spectral constituent mixture

represented by this pattern, while cool colors indicate a lower association...... 97

Figure 3.9: Time series from August 1, 2017 to November 21, 2018 of 3 spatial

distribution maps for VPC loading Pattern E. VPC Pattern E, on average represents

xii

spectral constituents of + phycocyanin, and – chlorophyll-b and Cyanophyta. In all

images warm colors indicate higher association of the pixels with the spectral

constituent mixture represented by this pattern, while cool colors indicate a lower

association...... 98

Figure 3.10: Time series from August 1, 2017 to November 21, 2018 of 1 spatial

distribution map for VPC loading Pattern F. VPC Pattern F, on average represents

spectral constituents of + α-PEC, and – Phaeophytin-b. In all images warm colors

indicate higher association of the pixels with the spectral constituent mixture

represented by this pattern, while cool colors indicate a lower association...... 99

xiii

LIST OF TABLES

Table 1.1: Table showing all the band length measurements that are recorded by

Landsat-8 (https://landsat.usgs.gov/what-are-band-designations-landsat-satellites).

...... 32

Table 1.2: This table displays the linear regression results of the standardized component

loadings from ASD measurements along the IRL, compared to the spectral library

standards at 10 nm resolution...... 33

Table 1.3: This table displays the linear regression results of the standardized Landsat-8

spectral analysis loadings from 07/28/17 compared to the spectral library standards

but at the 4 band Landsat-8 spectral resolution...... 34

Table 2.1: Manually recorded Secchi depths and optical depth, along with Yellow Springs

InstrumentsTM (YSI) EXO2 Multiparameter Sonde probe measurements for salinity,

turbidity, chlorophyll, and blue/green algae and phycoerythrin (BGA-PE) obtained

at all 11 sample locations on June 29-30, 2018...... 67

Table 3.1: VPC loading constituent identifications from the spectral library for all 10

Sentinel-3A OLCI image acquisition dates...... 100

Table 3.2: Results from the forward stepwise multiple linear regression of the average

Sentinel-3A OLCI spectral signature patterns identified with the constituent library.

...... 101

xiv

Table 3.3: Changes in percent variance accounted for VPC spectral pattern from time

series of Sentinel-3A OLCI imagery along with corresponding Kilroy water-quality

measurements at Sykes Creek station...... 102

DEDICATION

To the people who threw me out of the house and into woods, nurtured my curiosity regardless of where it took me in life, and engendered a stubbornness to prevail against insurmountable odds, I dedicate this work to my mother, Barbara Tiffee-Judice, and my father, Ken Judice.

xvi

ACKNOWLEDGEMENTS

Funding for this research investigation was made possible by the H.W. Hoover

Foundation, with additional funding from the KSU Graduate Student Senate for dissemination of research findings at the American Geophysical Union 2018 Fall Meeting.

We thank Dr. Edith Widder, Dr. Beth Falls, Warren Falls, Mike Corbet, Morgan Marmitt,

Jerry Corsaut, Retta Rohm, and Jessica Espinosa of the Ocean Research and Conservation

Association in Fort Pierce, FL who coordinated boat transportation, sample shipments, provided algal cultures, lab facilities, and access to YSI probes and the ORCA Kilroy instrument suite. Field work would also not be possible without the assistance the Kent

State undergraduate students Dominic Cristiano and Charles Hendrix. Progress on sample analyses at the KSU EDS-SEM and XRD Laboratories (managed by Dr. Elizabeth M.

Herndon and Dr. David M. Singer) was also made possible in thanks to training from Dr.

Elizabeth M. Herndon, Dr. Kuldeep Singh Chaudhary, Dr. Anne J. Jefferson, and assistance from Merida Keatts, Raihan Chowdhury, and Bryan Ice.

Taylor Joseph Judice

August 15, 2019, Kent, OH

xvii

Introduction

1.1 Background

Over the last several decades, freshwaters flowing into areas of the Indian River

Lagoon (IRL), Florida have transported high concentrations of nitrogen and phosphorus runoff from agricultural and residential areas. In June 2016, heavy rainfall caused nutrient- rich waters to accumulate in Lake Okeechobee. By the following month, water levels along bordering levees in Lake Okeechobee rose to unsafe heights, requiring the surrounding dikes to discharge freshwater through the C44 canal, which flows east down the St. Lucie

River and into the Southern IRL. Water carrying high concentrations of nitrogen and phosphorus into the IRL drove eutrophication and growth of various types of harmful algal blooms (HABs) in the IRL. These toxic water conditions devastated the local aquatic ecosystem, causing health problems for citizens, and hindering tourism in the surrounding communities.

The increasing frequency and intensity of HABs from eutrophic waters in coastal areas such as the IRL, has caused growing concern for the surrounding communities and ecosystems. The recent Comprehensive Everglades Restoration Plan (CERP), has been implemented to reroute the drainage of nutrient-rich Okeechobee waters into the St. Lucie to the east and Caloosahatchee River to the west, or to a southern drainage path into the

Everglades (Mclean, Ogden, & Williams, 2002). However, the construction and rerouting of the Florida waterways is a 35+ year project that has yet to be completed and further research should be carried out. Prior research has indicated agricultural runoff, increased urbanization, and sewage-polluted waters as just a few suggested sources of high nutrient inputs driving HABs (Kang, Koch, & Gobler, 2015; Lapointe, Herren, Debortoli, & Vogel,

2015). Along with sedimentary transport of organic-rich soil (muck), sand-size particles, clay, and weathered limestone into the lagoon (Lane, 1987), there are also spatial and temporal changes in both the amount of nutrient loading and the growth of phytoplankton within the IRL (C. J. Gobler et al., 2013; Kang & Gobler, 2018; Li et al., 2017; Phlips et al., 2011, 2015). Over the years, several studies have attempted to identify the specific toxic or harmful algal species found along the Florida coastline (Steidinger, Landsberg,

Tomas, & Burns, 1999), and the effects their toxins have on the local marine wildlife

(Capper, Flewelling, & Arthur, 2013).

In response to the persistent eutrophic water conditions, several field research and water conservation organizations now monitor seasonal variations in water quality.

Organizations such as the Ocean Research and Conservation Association (ORCA) have developed pier-mounted instrument packages called Kilroys, which provide real-time water quality measurements (Thosteson et al., 2009). These Kilroys are set at 13 locations along the IRL and measure temperature, water level, turbidity, pH, salinity, fluorescent dissolved organic matter (FDOM), oxidation reduction potential (ORP), blue-green algae via phycocyanin fluorescence, chlorophyll-a fluorescence, and current flow and direction. 2

The Kilroys have been deployed in the hopes of detecting the sources of water pollution and nutrient loading into this estuary.

In addition to these field measurements to monitor water quality, the use of remote sensing has become ever more valuable to track seasonal variations of blooms over large spatial scales in the IRL. Previously, remote sensing research has focused on monitoring water quality by identifying the spectral characteristics of color-producing constituents associated with HABs (Avouris & Ortiz, 2019; Kamerosky, Cho, & Morris, 2015; Lekki et al., 2017). The ability to quickly and reliably distinguish constituents of HABs, color dissolved organic matter (CDOM), and terrigenous sediment over large areas with remote sensing methods has become critically important, to better monitor and manage regional water quality and nutrient loading. This research investigation aims to expand on the current remote sensing methodologies for identifying the constituents of HABs within the

Indian River Lagoon.

In the past, satellite instruments such as MODIS (Moderate Resolution Imaging

Spectroradiometer) have used ocean color chlorophyll-a algorithms (Kamerosky et al.,

2015; O’Reilly et al., 2000) to track changes in Red Tide (Karenia brevis) growth along the west coast of Florida over time (Carvalho, Minnett, Banzon, Baringer, & Heil, 2011;

Carvalho, Minnett, Fleming, Banzon, & Baringer, 2010; Tomlinson, Wynne, & Stumpf,

2009). Methods developed by (K. Adem Ali, Ortiz, Bonini, Shuman, & Sydow, 2016;

Khalid A. Ali, Witter, & Ortiz, 2014a; Avouris & Ortiz, 2019; Ortiz, 2011; Ortiz et al.,

2017a, 2013) have expanded beyond retrievals of chlorophyll-a and assist in correction of 3

common remote sensing interferences, such as atmospheric contamination. These statistical techniques or numerical transformations, such as the Kent State University spectral decomposition method, a type of Varimax-rotated, Principal Component Analysis

(VPCA) can be used to process visible reflectance spectra (400-700 nm) from multispectral and hyperspectral imaging systems. The method has been demonstrated to be successful for distinguishing various types of constituents in the eutrophic waters of Lake Erie (Lekki et al., 2017).

VPCA is a method which helps to determine the intercorrelation between dependent variables of large, multivariate or hyperspectral datasets. In the contexts of remote sensing,

VPCA is useful for resolving the heterogeneity of the spectral signal retrieved from constituents mixed in the water. Orthogonal eigenvectors describe the direction that variation occurs within a data set, while the eigenvalues describe the total percentage of variability explained by each eigenvector (Kaiser, 1958). The information provided by the

VPCA simplifies large data sets by determining the leading spectral components of the satellite image that contribute to the overall signal. The component loadings recovered from the VPCA decomposition can be used to compare against known spectral signatures of relevant materials to determine what mixture of constituents is related to each component. By identifying the leading spectral components, one can quantitatively assess which areas of the scene have relatively higher or lower proportions of sediment or pigment-related constituents over time.

To optimize the results of the spectral identification and assess HAB growth, we must distinguish which satellite imaging system is most appropriate for studying the IRL.

Selecting a specific sensor is determined based on temporal resolution, spatial resolution, and spectral resolution, with an adequate signal to noise ratio. MODIS Aqua/Terra, with

9 bands in the visible, would provide a high spectral resolution and high temporal coverage, with 1- to 2-day repeatability. However, while MODIS Aqua/Terra has also been successfully incorporated into the VPCA methodology for Lake Erie HAB studies (Avouris

& Ortiz, 2019), the 1 x 1 km pixel resolution presents a challenge when assessing the narrower channels of the IRL, which can range from 0.5 km to 8 km in width. The

European Space Agency’s Sentinel-3A Ocean and Land Colour Imager (OLCI) has 11 bands in the visible and uses a finer 300 x 300 m spatial resolution, with minimal loss of temporal resolution (4-day repeatability). Depending on the scale or how localized HABs occur within the IRL, this study might require small spatial resolution imagers to address bloom dynamics. Given this parameter trade space, Sentinel-3A/B will provide the best compromise for work in the Northern IRL.

While NASA’s Landsat-8 Operational Land Imager (OLI) lacks the spectral resolution (4 bands in the visible) and with a 14-day repeat, the temporal coverage of

Sentinel 3A/B or MODIS Aqua/Terra, it can provide a much finer spatial resolution of 30 m, which could be useful for more localized studies, particularly in the southern reaches of the IRL. Thus, in Chapter 1, prior to utilizing Sentinel-3A OLCI imagery for the remainder of this thesis, we will attempt to improve the spectral retrievals of the VPCA decomposition 5

for the lower spectral resolution Landsat-8 OLI imager and thus increase its signal-to-noise ratio. Overall for this research investigation, we employ both in-situ water data and satellite-based imagery to derive estimates for specific components that are contributing to the entire spectral signal in the IRL. Applying the KSU spectral decomposition method we can compare the component loadings to known constituents from the USGS spectral library (Kokaly et al., 2017) and other independent measurements (Avouris & Ortiz, 2019;

Ortiz et al., 2013) provided that the lower spectral resolution of Landsat-8 remains reliable for spectral identification. By comparing remote sensing to in-situ measurements of water quality, we aim to improve our detection of constituents associated with phytoplankton growth, which will provide additional reliable methods for tracking HABs over a large coastal region, with fine spatial resolution. Given the caveat of lower temporal resolution for Landsat-8 imagery, if temporal coverage of the IRL is insufficient for comparison with in-situ sample collections, the methods developed from this investigation may provide a useful reference for future remote sensing VPCA studies, which require imagery of finer spatial resolution.

1.2 Research Objectives

By applying the KSU spectral decomposition method we might improve the spectral retrieval quality of specific pigments found in HABs and potentially provide better resources for detecting blooms in earlier stages in the future. Regardless of whether we choose to implement Sentinel-3A/B or Landsat-8 imagery, in this research investigation,

our objectives are to accomplish the following: (1) apply the KSU VPCA spectral decomposition method for both field hyperspectral water reflectance measurements and satellite remote sensing imagery of the IRL to determine the leading spectral components in water, such as suspended sediment or pigmented constituents from HABs, other algae and cyanobacteria, (2) determine if the results from spectral analysis can be correlated to cell counts and water quality measurements at the Kilroy sites, and (3) produce a time series of satellite imagery for the IRL using the KSU spectral decomposition method to track the growth of HABs at each Kilroy site over time.

1.3 Research Plan/Methods

1.3.1 Field Sampling

To execute this study, we’ve worked in conjunction with ORCA to collect field samples and measurements along the IRL on July 30, 2017. ORCA currently has installed

Kilroys at 13 locations along the IRL (Figure 1.1), which measure temperature, turbidity, pH, salinity, FDOM, ORP, blue-green algae, and chlorophyll-a, approximately every 30 minutes to 1 hour. During our field deployment, we acquired water reflectance measurements using a Malvern Panalytical Analytical Spectral DevicesTM (hereafter ASD)

FieldSpec Handle-held 2 (HH2) hyperspectral spectroradiometer and acquired cell count data from water samples at 9 of the 13 Kilroy locations deployed in the Northern IRL. For the cell counts of samples acquired during the field deployments, each sample was identified to the genus and species level. These were also integrated to the Phylum level

for comparison with reflectance data. Both the total density (cells/L) and the cells by volume (μm3/L) were measured for each sample. The hyperspectral instruments recorded several measurements for water surface reflectance at 10 nm resolution in the visible/near- infrared wavelength range (400-900 nm) with a 10-degree field of view foreoptic lens.

Before the field hyperspectral spectroradiometer measured the surface water reflectance, the instrument was calibrated using a white SpectralonTM reference plate. The in-situ water reflectance spectra obtained from the ASD FieldSpec HH2 offers information that can be used to validate the satellite remote sensing data. Along with water reflectance measurements, two duplicate water samples at each of the 9 locations were filtered on 0.7

μm pore glass fiber filters (GF/F) and dried at 60° C for ~1 hour. The ASD FieldSpec HH2 was calibrated using a white reference plate before measuring the reflectance of particulates from these filtered samples. We measured the reflectance values of the particulates accumulated on these filters with the hyperspectral spectroradiometer, for comparison with chlorophyll-a concentrations recorded at the water monitoring stations.

1.3.2 Remote Sensing

In addition, we obtained satellite imagery from the multispectral instrument

Operational Land Imager (OLI) aboard Landsat-8 whose acquisition periods were within

±3 days of the acquired water sample dates. Landsat-8 imagery is available at no cost from the USGS Earth Explorer website. Landsat-8 OLI has 7 bands in the visible, infrared, and

SWIR spectrum (435-2294 nm) with 30 m spatial resolution. The OLI is a push-broom

sensor and provides imagery of the same location every 14-16 days. Because of the repeatability of the satellite orbit, and given cloud conditions, we expect one or two images of the IRL to be acquired each month. There are two levels of preprocessed images, Level

2 Landsat products are images that have already been atmospherically corrected and geo- referenced, while Level 1 has not been corrected. Currently, we use ENVI/IDL

(Environment for Visualizing Images-Interactive Data Language), a coupled image analysis software and programming environment package from Harris Geospatial, to process atmospherically corrected, Level 2 Landsat-8 products to Level 3 VPCA products.

ENVI provides the satellite imaging manipulation software, while IDL works in tandem with ENVI to provide additional coding algorithms for further image analysis.

Before importing images into ENVI/IDL, we choose the appropriate image swaths acquired for the IRL that are < 30% cloud covered to insure more reliable image quality for the final product. Image tiles are a global notation for referencing the location of

Landsat data, which are designated by path and row numbers. For Landsat-8, the path and row numbers available for the IRL are the following: (path/row) 016/040, 015/040, and

015/041. For the purposes of this research we will focus primarily on the tile 016/040 of the northern IRL and Banana River. Once the tiles for an acquisition date are downloaded, they are mosaiced together if necessary. Using data from the Landsat-8 aerosol imagery values we mask pixel values that represent land, clouds, or cloud shadows on the water to exclude them from further image analysis. To reduce the noise of the image, we apply a

3x3 kernel median filter. Median smoothing is a common low-pass filtering technique for 9

pre-processing in remote sensing, which compares the neighboring pixel values of an image to determine if a pixel is representative of the surrounding values. This is accomplished by moving through the image pixel by pixel and replacing the central pixel with the median value of a 3 x 3 pixel window.

1.3.3 VPCA Spectral Decomposition

Applying the VPCA spectral decomposition to Landsat-8 multispectral data and hyperspectral field data, we determine the leading spectral components of the signal using

SPSS (Statistical Package for the Social Sciences) computational software for the field spectrometer data or ENVI/IDL for image analysis. The multispectral Landsat-8 imagery can be thought of as a three-dimensional data cube, where there are two spatial dimensions

(i.e. the location of a given pixel in an image) and one spectral dimension (the wavelength bands). Landsat data is made up of several raster layers of the same image sample acquired at specific wavelength bands (Table 1.1). Using the four visible bands and the one near infrared band from Landsat-8 (430-450, 450-510, 530-590, and 640-670 nm), we compute the derivative spectra in IDL to separate out any peaks and troughs from overlapping bands and filter out low frequency noise from the signal. Additionally, to increase the spectral coverage, we calculate several normalized differences, which are also included in the

VPCA: Normalized Difference Vegetation Index (NDVI), Normalized Difference Water

Index (NDWI), modified Photochemical Reflectance Index (mPRI), modified Visible

Atmospherically Resistant Index green (mVARIgreen), and Normalized Difference of Red to Blue bands (NDRB).

NDVI and NDWI are typical indices used in remote sensing, NDVI is the normalization of the red and near-infrared (NIR) bands which is useful as a standardize method of measuring vegetation health and an indicator for drought (Carlson & Ripley, 1997), while

NDWI is the normalized difference of the NIR and the Shortwave Infrared (SWIR) bands.

Using both the SWIR and NIR in the NDWI removes the variation produced in remote sensing from dry leaf matter and the internal leaf structure such as the spongy mesophyll (Gao, 1996). mPRI is a normalized difference modified for the blue and green bands of Landsat (mPRI =

[Blue - Green] / [Blue + Green]), which was developed as an indicator of photosynthetic efficiency (Gamon, 2015; Gamon, Peñuelas, & Field, 1992; Garbulsky, Peñuelas, Gamon,

Inoue, & Filella, 2011). The VARI or VARIgreen algorithm was developed as a vegetation index that is affected minimally by atmospheric scattering (Gitelson, Kaufman, & Merzlyak,

1996; Gitelson, Kaufman, Stark, & Rundquist, 2002). This is accomplished by using the blue, green, and red bands, however to coordinate the increase of algal bloom growth with increasing

VARI values, we switch the green and red band orders in the equations and refer to it as mVARIgreen to avoid confusion (mVARIgreen = [Red - Green] / [Red + Green - Blue]). The

NDRB is a normalization of the red and blue bands (NDRB = [Red - Blue] / [Red + Blue]) used to produce a raster layer that is sensitive to detecting the absorption of constituents other than chlorophyll-a, such as accessory pigments and suspended sediment or CDOM.

The inclusion of the normalized differences helps in several ways. The normalized differences are used extensively and thus have a wide literature regarding their 11

interpretation. This helps to place constraints on the interpretation of the VPCA results.

The use of normalized differences also provides quantitative constraints to the VPCA that aids in the removal of detector offsets if present in the image. Finally, by adding various normalized differences to the VPCA, we can partition them into orthogonal signals, which aids in their interpretation.

From here the correlation structure of the seven reflectance bands, four derivative bands, and the five normalized difference indices are input into the PCA function to extract eigenvalues and eigenvectors by singular value decomposition. The component loadings

(eigenvectors) of the PCA are then varimax-rotated, which produces the final scores of the

VPCA loadings. There are typically 3 to 4 component scores retrieved from the decomposition that are projected back onto the derivative dataset in map view to display the component scores, the spatial distribution of these components throughout the IRL. The VPCA analysis is also applied in this manner to in-situ field hyperspectral spectroradiometer measurements and lab spectrophotometer measurements of the GF/F filtered water samples. Using a forward, stepwise, multiple linear regression, the z-scores of the component loadings are compared to the spectral library to identify the constituents that contribute to each component.

1.4 Initial Landsat-8 and Field Results

Figure 1.2 shows the standardized component loadings of the combined ASD field spectra correlated to previously measured spectral fits from the USGS spectral library at 10 nm wavelength resolution. The VPCA statistical method extracted four significant components which account for 36%, 30.3%, 19.1%, and 10.4% variance of the total combined variance of

95.9%. These four components made up most of the total spectral variability, which can be expressed on a band by band basis by the communality. The average communality equaled

0.97. The results from the linear regressions comparing known spectral standards with both

Landsat-8 and the ASD standardized component loadings is shown in Table 1.2 and Table 1.3.

Figure 1.3 shows the standardized component loadings of the extracted Landsat-8 imagery of the Northern Indian River Lagoon from July 28, 2017, correlated to previously measured spectra from the USGS spectral library. For this stepwise forward regression, the 10 nm resolution library spectra were averaged down to the Landsat-8 four band resolution (430-450, 450-510, 530-590, and 640-670 nm). The VPCA statistical method extracted four significant components, which account for 48%, 20%, 18%, and 6% variance of the total combined variance of 92%. These four components made up a large majority of the total spectral variability, which can be expressed from the average communality equaling 0.90.

Comparison of the component loadings to the known spectra for the hyperspectral

ASD measurements indicate the first component correlates with hematite, which could be suspended sediments (Figure 1.2a). The second component relates to bacillariophytes and kaolinite (Figure 1.2b). The third component correlates with illite and fucoxanthin (Figure

1.2c), while the fourth component relates to illite and chlorophyllide-b (Figure 1.2d).

Comparison of the component loadings to the known spectra for the Landsat-8 resolution indicates the first component relates to myxoxanthophyll, an accessory pigment associated with cyanobacteria (Figure 1.3a), while the second component relates to bacillariophyceae

(Figure 1.3b). The third component relates to phycocyanin (Figure 1.3c), while component four represents cyanobacteria (Figure 1.3d).

Both the total density (cells/L) and the cells by volume (μm3/L) were measured for each sample. The percentages of these were calculated and are shown in Figure 1.4 and

Figure 1.5 for the distribution of cell divisions at each site in the Northern Indian River

Lagoon. The top two total by volume percentages at Turnbull Creek is approximately 66%

Haptophyta and 20% Ochrophyta (Figure 1.4a). Haulover Canal shows 55% Ochrophyta and 24% cyanobacteria (Figure 1.4b). For Kilroy water monitoring locations at Sykes and

Barge, Miozoa (dinoflagellates) make up the major percentage (~93% of the total volume) of the sample volumes during these days (Figure 1.4c, Figure 1.4e). At the Dragon Point location, 88% of the sample volume was identified as Bacillariophyta (diatoms) and 11% of the total volume was identified as cyanobacteria (Figure 1.4d). For all Kilroy sites in the Northern IRL except for Dragon’s Point and Turnbull Creek, percentage by density from the cell count indicates the remaining samples contain ≥ 93% cyanobacteria (Figure

1.5). Turnbull Creek has 58% cyanobacteria, 37% Haptophyta, 4% Chlorophyta, 1%

Bacillariophyta, and < 1% Ochrophyta by total density (Figure 1.5a). Dragon’s Point has

89% cyanobacteria, 9% Bacillariophyta, and 2% Chlorophyta (Figure 1.5d).

The Kilroy Chlorophyll-a, blue-green algae, and turbidity measurements for July

28, 2017 were averaged for any measurement recorded between 9:00 am and 3:00 pm, as well as ± 2 hours and ± 1 hour from the acquisition time of the Landsat-8 imagery of the day. Average chlorophyll-a concentrations are recorded between approximately 5-20 14

μg/mL at all locations, except for Northern IRL site Turnbull Creek at 110μg/mL (Figure

1.6). This along with concentrations of blue/green algae tend to be higher at this site.

Turbidity in all these locations tends to fluctuate quite a bit in all locations.

Figure 1.7 shows the final July 28, 2017 Landsat-8 imagery of all four VPCA component loadings extracted by only using the derivative spectra. This was a preliminary image for testing the VPCA decomposition and does not have a 3-kernel median smoothing applied. Figure 1.8 shows the spectral decomposition for the same image from Landsat-8, but with a 3-kernel median smoothing applied and the addition of the seven reflectance bands and five indices incorporated into the VPCA with derivative spectra. Before the inclusion of smoothing and additional indices, component images 3 and 4 produce a large amount of noise and striping in the image (Figure 1.7).

1.5 Discussion and Future Work

By including the additional layers to the VPCA spectral decomposition and adding all seven Landsat-8 reflectance bands and normalized differences (NDWI, NDVI, mPRI, mVARI, and NDRB), to the four derivative bands, the degrees of freedom increase in the analysis. This increase in the degrees of freedom has decreased the noise and striping of the final VPCA loading images. The preliminary results suggest that by adding additional indices to the spectral decomposition analysis we can account for more of the variability in the signal and remove calibration offsets between bands. Furthermore, the component loadings of these additional variables should provide insight for interpreting the loadings

of derivative spectra rather than using the spectral decomposition on the derivative spectra alone. This process for spectral analysis, as illustrated from this one 07/28/2017 image, would be repeated for all low cloud cover images obtained of the IRL from June 2017 to

January 2019, however the lower temporal and spectral resolution of OLI will produce difficulties when coordinating with cell count collections for validation and characterizing the component loadings’ spectral identification with the constituent library. The lower temporal resolution increases the likelihood obtaining cloud covered images, which would affect in-situ samples collected for comparison within ± 3 days of the image acquisition.

Our study will thus produce more effective results for validation, spectral identification, and time series of images by utilizing the Sentinel-3A OLCI for the future.

Despite removing Landsat-8 imagery from our future analyses in our investigation, these results from the varimax-rotated principal component (VPC) scores (Figure 1.8) yield insightful questions for coordinating sample collections in 2018 and new areas of interest within the Northern IRL. High score counts appear to be concentrated in the Banana River

Lagoon, east of the Northern IRL (Figure 1.8). Whether these component loadings are correctly being identified with the constituent library still requires validation, but these higher score values within the Banana River do warrant the need to collect field samples in this area for the future. Additionally, preliminary data from ORCA colleagues has observed high counts of Brown Tide (Aureoumbra lagunensis) blooms, within the Banana

River region of the IRL. Aureoumbra lagunensis contains the pigment fucoxanthin, and thus has a spectral shape similar to that of diatoms, particularly at coarse spectral 16

resolution. Moving forward, the 2018 summer field deployment will collect cell counts and spectra from additional sites along with the Kilroy sample locations. These new sites will provide samples for both IRL and especially the Banana River, which has fewer Kilroy water monitoring systems. With this additional data we hope to characterize the bloom dynamics or disparities between the Northern IRL and the Banana River. Finally, if cultured samples of Aureoumbra lagunensis are available, we will obtain spectra that can be incorporated in the growing constituent spectral library and improve VPCA spectral identifications.

1.6 References

Ali, K. A., Witter, D. L., & Ortiz, J. D. (2014). Multivariate approach to estimate colour

producing agents in Case 2 waters using first-derivative spectrophotometer data.

Geocarto International, 29(2), 102–127.

https://doi.org/10.1080/10106049.2012.743601.

Capper, A., Flewelling, L. J., & Arthur, K. (2013). Dietary exposure to harmful algal bloom

(HAB) toxins in the endangered manatee (Trichechus manatus latirostris) and green

sea turtle (Chelonia mydas) in Florida, USA. Harmful Algae, 28, 1-9.

Carlson, T. C., & Ripley, D. a. (1997). On the relationship between NDVI, fractional

vegetation cover, and leaf area index. Remote Sensing of Environment, 62, 241–252.

https://doi.org/10.1016/S0034-4257(97)00104-1.

Carvalho, G. A., Minnett, P. J., Fleming, L. E., Banzon, V. F., & Baringer, W. (2010).

Satellite remote sensing of harmful algal blooms: A new multi-algorithm method for

detecting the Florida Red Tide (Karenia brevis). Harmful Algae, 9(5), 440–448.

https://doi.org/10.1016/j.hal.2010.02.002.

Carvalho, G. A., Minnett, P. J., Banzon, V. F., Baringer, W., & Heil, C. A. (2011). Long-

term evaluation of three satellite ocean color algorithms for identifying harmful algal

blooms (Karenia brevis) along the west coast of Florida: A matchup assessment.

Remote Sensing of Environment, 115(1), 1–18.

https://doi.org/10.1016/j.rse.2010.07.007.

Gamon, J. A., Peñuelas, J., & Field, C. B. (1992). A narrow-waveband spectral index that

tracks diurnal changes in photosynthetic efficiency. Remote Sensing of Environment,

41(1), 35–44. https://doi.org/10.1016/0034-4257(92)90059-S.

Gamon, J. A. (2015). Reviews and Syntheses: Optical sampling of the flux tower footprint.

Biogeosciences, 12(14), 4509–4523. https://doi.org/10.5194/bg-12-4509-2015.

Gao, B. C. (1996). NDWI - A normalized difference water index for remote sensing of

vegetation liquid water from space. Remote Sensing of Environment, 58(3), 257–266.

https://doi.org/10.1016/S0034-4257(96)00067-3.

Garbulsky, M. F., Peñuelas, J., Gamon, J., Inoue, Y., & Filella, I. (2011). The

photochemical reflectance index (PRI) and the remote sensing of leaf, canopy and

ecosystem radiation use efficiencies. A review and meta-analysis. Remote Sensing of

Environment, 115(2), 281–297. https://doi.org/10.1016/j.rse.2010.08.023.

Gitelson, A. A., Kaufman, Y. J., Stark, R., & Rundquist, D. (2002). Novel algorithms for

remote estimation of vegetation fraction. Remote Sensing of Environment, 80(1), 76–

87. https://doi.org/10.1016/S0034-4257(01)00289-9.

Gitelson, A. A., Kaufman, Y. J., & Merzlyak, M. N. (1996). Use of a green channel in

remote sensing of global vegetation from EOS-MODIS. Remote Sensing of

Environment, 58(3), 289–298. https://doi.org/10.1016/S0034-4257(96)00072-7.

Jeffrey, S. W. (1997). Introduction to marine phytoplankton and their pigment

signatures. Phytoplankton pigment in oceanography, 37-84.

Kang, Y., Koch, F., & Gobler, C. J. (2015). The interactive roles of nutrient loading and

zooplankton grazing in facilitating the expansion of harmful algal blooms caused by

the pelagophyte, Aureoumbra lagunensis, to the Indian River Lagoon, FL,

USA. Harmful Algae, 49, 162-173.

Kokaly, R.F., Clark, R.N., Swayze, G.A., Livo, K.E., Hoefen, T.M., Pearson, N.C., Wise,

R.A., Benzel, W.M., Lowers, H.A., Driscoll, R.L., and Klein, A.J., (2017). USGS

Spectral Library Version 7: U.S. Geological Survey Data Series 1035, 61 p.,

https://doi.org/10.3133/ds1035.

Lane, E. (1987). Guide to rocks and minerals of Florida (FGS: Special Publication 8 rev.).

Lapointe, B. E., Herren, L. W., Debortoli, D. D., & Vogel, M. A. (2015). Evidence of

sewage-driven eutrophication and harmful algal blooms in Florida's Indian River

Lagoon. Harmful Algae, 43, 82-102.

Lekki, J., Anderson, R., Avouris, D., Becker, R., Churnside, J., Cline, M., ... & Ortiz, J.

(2017). Airborne Hyperspectral Sensing of Monitoring Harmful Algal Blooms in the

Great Lakes Region: System Calibration and Validation.

Li, L., He, Z., Li, Z., Li, S., Wan, Y., & Stoffella, P. J. (2017). Spatiotemporal change of

phosphorous speciation and concentration in stormwater in the St. Lucie Estuary

watershed, South Florida. Chemosphere, 172, 488–495.

https://doi.org/10.1016/j.chemosphere.2017.01.020.

McLean, A. R., Ogden, J. C., & Williams, E. E. (2002). Comprehensive Everglades

Restoration Plan. West Palm Beach, Florida.

O’Reilly, J. E., Maritorena, S., Siegel, D., O’Brien, M. C., Toole, D., Mitchell, B. G., ... &

Hooker, S. B. (2001). Ocean color chlorophyll a algorithms for SeaWiFS, OC2, and

OC4: Version 4. SeaWiFS postlaunch technical report series, 11, 9-23.

Ortiz, J. D., Witter, D. L., Ali, K. A., Fela, N., Duff, M., & Mills, L. (2013). Evaluating

multiple colour-producing agents in Case II waters from Lake Erie. International

Journal of Remote Sensing, 34(24), 8854–8880.

https://doi.org/10.1080/01431161.2013.853892.

Ortiz, J. D., Avouris, D., Schiller, S., Luvall, J. C., Lekki, J. D., Tokars, R. P., … Becker,

R. (2017). Intercomparison of Approaches to the Empirical Line Method for Vicarious

Hyperspectral Reflectance Calibration. Frontiers in Marine Science, 4(September).

https://doi.org/10.3389/fmars.2017.00296.

Phlips, E. J., Badylak, S., Christman, M., Wolny, J., Brame, J., Garland, J., ... & Lockwood,

J. (2011). Scales of temporal and spatial variability in the distribution of harmful algae

species in the Indian River Lagoon, Florida, USA. Harmful Algae, 10(3), 277-290.

Schulte, T., Johanning, S., & Hofmann, E. (2010). Structure and function of native and

refolded peridinin-chlorophyll-proteins from dinoflagellates. European journal of cell

biology, 89(12), 990-997.

Steidinger, K.A., Landsberg, J.H., Tomas, C.R., and Burns, J.W. (1999). Harmful algal

blooms in Florida, Unpublished technical report submitted to the Florida Harmful

Algal Bloom Task Force, Florida Marine Research Institute, 63pp.

Thosteson, E. D., Widder, E. A., Cimaglia, C. A., Taylor, J. W., Burns, B. C., & Paglen,

K. J. (2009). New technology for ecosystem-based management: Marine monitoring

with the ORCA Kilroy Network. OCEANS ’09 IEEE Bremen: Balancing Technology

with Future Needs. https://doi.org/10.1109/OCEANSE.2009.5278229.

Tomlinson, M. C., Wynne, T. T., & Stumpf, R. P. (2009). An evaluation of remote sensing

techniques for enhanced detection of the toxic dinoflagellate, Karenia brevis. Remote

Sensing of Environment, 113(3), 598–609. https://doi.org/10.1016/j.rse.2008.11.003.

1.7 Figures

Figure 1.1: True color image of the (a) nine locations of Kilroy water monitoring instruments along the east coast of Southern Florida in the Indian River Lagoon, (b) zoomed-in images of the five Kilroy systems in the northern portion of the IRL

Figure 1.2: VPCA component loadings from in field ASD hyperspectral reflectance measurements after standardizing and computing forward stepwise linear regression with the spectral library.

Figure 1.3: Component loadings from Landsat-8 swath 016/040 of the IRL after standardizing and computing a forward stepwise linear regression with the spectral library.

Figure 1.4: Results of cell count analysis, defining the division of microorganisms by the percentage of total bulk volume. (a) Turnbull Creek, (b) Haulover Canal, (c) Barge Canal, (d) Dragon’s Point, (e) Syke’s Creek.

Figure 1.5: Results of cell count analysis, defining the division of microorganisms by the percentage of total density. (a) Turnbull Creek, (b) Haulover Canal, (c) Barge Canal, (d) Dragon’s Point, (e) Syke’s Creek.

Figure 1.6: Graphs showing the average measurement from Kilroy water monitoring system for July 28, 2017 (“Full series” represents the average values that are available at the Kilroys from 8:00 am to 3:00 pm).

Figure 1.7: Landsat-8 VPCA imagery of Northern IRL (July 28, 2017). VPCA (a) component 1 has positive correlation with peridinin, (b) component 2 has a positive correlation with phaeophorbide-b and negative with chlorophyll-a + cyanophyta, (c) component 3 has a negative correlation bacillariophyceae and positive with illite, (d) component 4 has a positive correlation with dinophyta and negative correlation with

fucoxanthin. Warmer colors indicate higher correlation with reference spectra; colder colors represent lower correlation with reference spectra.

Figure 1.8: Landsat-8 VPCA imagery of Northern IRL (July 28, 2017). VPCA (a) component 1 has positive correlation with myxoxanthophyll, (b) component 2 has a negative correlation with bacillariophyceae, (c) component 3 has a positive with phycocyanin, (d) component 4 has a positive correlation with cyanobacteria. Warmer 31

colors indicate higher correlation with reference spectra; colder colors represent lower correlation with reference spectra.

1.8 Tables

Table 1.1: Table showing all the band length measurements that are recorded by Landsat- 8 (https://landsat.usgs.gov/what-are-band-designations-landsat-satellites).

Table 1.2: This table displays the linear regression results of the standardized component loadings from ASD measurements along the IRL, compared to the spectral library standards at 10 nm resolution.

Table 1.3: This table displays the linear regression results of the standardized Landsat-8 spectral analysis loadings from 07/28/17 compared to the spectral library standards but at the 4 band Landsat-8 spectral resolution.

Field-validated detection of Aureoumbra lagunensis bloom in Indian River Lagoon,

Florida using Sentinel-3A OLCI and ground-based hyperspectral

spectroradiometers

2.1 Abstract

Frequent Aureoumbra lagunensis blooms in the Indian River Lagoon (IRL) coastal estuary have devastated populations of seagrass and marine life, resulting in a demand for early reliable assessments of harmful algal blooms over this ~5,500 km2 body of water using emerging remote sensing techniques. A varimax-rotated principal component analysis (VPCA) method developed for remote sensing imagery, decomposes the integrated spectral signature from optically complex water into independent component spectra, which are identified with a library of known spectral constituents. Coupling in- situ cells counts, water-quality monitoring systems, and hyperspectral spectroradiometer reflectance measurements from June 29-30, 2018 for validation, this research addresses whether the VPCA technique for Sentinel-3A OLCI imagery can detect A. lagunensis- related constituents in optically complex waters. Here we show that VPCA components retrieved from satellite and a field-based spectrometer are consistent. Furthermore, the

VPCA components associated with A. lagunensis, for both spectral datasets, indicate high correlations to concentrations of Ochrophyta (Phylum) and Chlorophyll-a (R2 > 0.90, p <

0.05). Overall, our observations provide excellent validation for Sentinel-3A VPCA spectral identification and suggest A. lagunensis, or Brown Tide-related constituents, are highly concentration within the Banana River region of the IRL. Future work will need to discern the detection limit of A. lagunensis VPCA constituents with respect to spatial frequency of Chlorophyll-a and Ochrophyta concentrations.

2.2 Introduction

Located along the east coast of Florida, the Indian River Lagoon (IRL) is a shallow- water estuary, stretching along 250 km of coastline. The IRL encompasses three major lagoons: Mosquito Lagoon in the north, Banana River in the northeast, and the major body of water, Indian River, which extends from Volusia County to Palm Beach County (Figure

2.1a). In recent years, the growth of harmful algal blooms (HABs) within the waters of the

IRL have devastated populations of marine life and fauna and hindered the local economy.

The increasing frequency of cyanobacterial HABs has posed an environmental hazard of growing concern to the more than 1 million residents living along the IRL, because of growing evidence of the human health impacts of cyanotoxins including neurodegenerative and liver diseases (Metcalf et al., 2018). Historically, the IRL has also provided a sanctuary for migratory animals such as dolphins, manatees, and several species of birds, and was considered one of the most biodiverse regions in the United States. In addition to its

importance to wildlife and fauna (Sime, 2005), the IRL provides an important economic resource for local tourism and fishery industries. These compounded environmental issues facing the IRL region have gathered research efforts which combine the resources of field instrumentation and recently developed satellite remote sensing techniques to help improve best management practices for the IRL watershed and reduce nutrient-rich runoff that favors environments for HAB growth.

One recurring species of HABs that has grown dominant in portions of the northern

IRL is the pelagophyte Aureoumbra lagunensis, a member of the Phylum Ochrophyta, also known as “Brown Tide” (C. J. Gobler et al., 2013). Although A. lagunensis is not a toxin producer, its extreme density of growth blocks sunlight and has led to the extensive die off of seagrasses, which is vital estuarine habitat. Along with Karenia brevis (“Red Tide”), the propagation of A. lagunensis along the North American coastlines has been monitored as it has migrated to areas of Texas, Cuba, and Florida (Hall et al., 2018; Koch, Kang,

Villareal, Anderson, & Gobler, 2014). The persistence of this HAB forming species in coastal waters is due to its resiliency to grow in hypersaline conditions with limited P and sunlight due to shallow and moderately more turbid environments (DeYoe et al., 1997; Liu,

Laws, Villareal, & Buskey, 2001). In many cases, A. lagunensis is adapted to low

Phosphorus (P) conditions, which has indicated higher N:P ratios needed for production

(Kang et al., 2015). Field research has linked the potential sources of nutrient pollution causing the Brown Tides to agricultural fertilizer runoff (Zhang, Wang, & He, 2007), leaching of N from onsite sewage disposal systems (OSDS) into the permeable limestone 37

and sandstone bedrock of Florida (Barile, 2018; Lapointe et al., 2015; Lapointe, Herren, &

Paule, 2017), and the accumulating impact of legacy nutrients resuspended by natural and human-related actions (Dunne, Clark, Corstanje, & Reddy, 2011; Fox & Trefry, 2018;

Reddy, Newman, Osborne, White, & Fitz, 2011; Yang et al., 2013). Some research suggests that the limiting nutrient for Brown Tide growth is organic nitrogen derived from

OSDS that is ultimately reduced to ammonium or urea (C. J. Gobler & Sunda, 2012a; C.

Gobler & Sañudo-Wilhelmy, 2001; Kang et al., 2015; Lapointe et al., 2017). Whether the nutrient pollution is sourced from agricultural or septic runoff, the permeable sandy subsurface of Florida provides a soil that has poor retention of N and P, which results in export to water bodies (McNeal et al., 1995), leading to more frequent HAB occurrences during the wet seasons of the year (Chamberlain & Hayward, 1996). While field sampling research has provided crucial information on the distribution of phytoplankton across the

IRL (Phlips et al., 2011), the long-term cost of operation and analysis of intensive sampling to monitor water-quality rapidly becomes difficult.

Given the spatial extent of the HAB and nutrient pollution issue across the IRL, satellite-based remote sensing provides an effective way to monitor the impact of nutrient pollution over this larger body of water that would otherwise be difficult to characterize using traditional field sampling methods alone. Various methods have been developed to characterize HABs through remote sensing from optically complex waters. These methods have provided several standard metrics for Chlorophyll-a (Chl-a) estimation (Kamerosky et al., 2015; Witter, Ortiz, Palm, Heath, & Budd, 2009). Limitations of these standard 38

remote sensing methods are that many types of autotrophs contain Chl-a, and HABs are often not monospecific, but are often heterogenous in composition. The Kent State

University, varimax-rotated principal component analysis (KSU VPCA) method was developed for decomposing the integrated, spectral signature from optically complex water retrieved by remote sensing instruments into independent component spectra, which account for quantifiable percentages of the total signal variability (Avouris & Ortiz, 2019;

Ortiz et al., 2019, 2017a, 2013). Components represent mixtures of constituents that can be characterized by forward stepwise multiple linear regression against a library of known algae, pigments, and sediments (Avouris & Ortiz, 2019; Ortiz et al., 2019). By partitioning spectral signatures into orthogonal components and identifying them, satellite images faithfully account for constituents in the water without bias from intercorrelated variables.

The importance of these findings has shown how HAB spectral constituents mixed in the water column can be detected before they dominate the whole signal. By remotely detecting minor HAB constituents in water with the VPCA spectral decomposition methodology, remediation efforts can be implemented earlier, providing additional warning in areas such as the IRL.

What can we learn about the spectral signature of Brown Tide blooms in the IRL using the KSU VPCA method? We hypothesize that if the Sentinel-3A Ocean and

Land Colour Instrument (OLCI) VPCA spectral decomposition produces spectral components that are identified as algal constituents, then the spatial distribution of those components would have a high correlation to phylum-level cell count measurements 39

collected within ± 2 days of each other. Moreover, given the state of sewage treatment with OSDS in the IRL, we predict that components associated with Brown Tide algae will be concentrated in the Banana River region. The results from this study were part of a larger two year-long research project on the Indian River Lagoon designed to offer proof of concept for the generalization of the method following prior studies conducted in the nutrient-polluted waters of Lake Erie (K. Adem Ali et al., 2016; Khalid A. Ali & Ortiz,

2016; Khalid A. Ali et al., 2014a; Avouris & Ortiz, 2019; Ortiz et al., 2019, 2017a) and the meso- to oligotrophic waters of the U.S. Virgin Islands (Katz et al., 2018).

2.3 Methods

2.3.1 Field Data

For this study, we implemented field sampling within the IRL on June 29-30, 2018 from small boats with assistance of the Ocean Research and Conservation Association

(ORCA). We obtained water samples for cell count analyses by a commercial lab (BSA,

Beachwood, OH), along with suspended sediment samples, Secchi depths, monitoring systems for water-quality, and hyperspectral surface reflectance measurements from 11 locations in the IRL: 6 samples from the northern Indian River Lagoon and 5 samples from the Banana River region (Figure 2.1b). These field samples provided in-situ data for supplemental validation of the KSU VPCA spectral decomposition method on Sentinel-3A

Ocean Land Colour Imager (OLCI) imagery acquired on June 28, 2018. To ensure spatial consistency for comparisons with OLCI imagery, GPS coordinates were recorded at all

locations for each sample collection. For cell count analyses, two replicate 125 mL surface water samples were collected at each of the 11 locations. Once obtained, the cell count samples were preserved using 2 mL of 5% Lugols solution per 25 mL of sample. The suspended sediment samples were collected with a clamshell sampler in order to increase the amount of usable sample for future Scanning Electron Microscope (SEM-EDS) analyses, then stored in clear 125 mL containers. All water and wet sediment samples were kept cold on ice until returning to the lab for processing and shipment.

Using an Analytical Spectral DevicesTM (ASD) FieldSpec® Handheld 2 (HH2) hyperspectral spectroradiometer, we measured the absolute surface water reflectance at the

11 sample locations using a 10-degree field of view foreoptic attachment. The reflectance spectra from the hyperspectral ASD FieldSpec® HH2 were averaged to 10 nm resolution from 400-700 nm. This provides a more continuous 31 band spectra compared to the 11 multispectral bands of Sentinel-3A OLCI. Furthermore, the hyperspectral field instrument allows us to obtain surface water reflectance with much less atmospheric interference than retrievals from satellite imagery. For each FieldSpec® HH2 surface reflectance measurement, the optics were calibrated with an ASD standard white reference

SpectralonTM plate before orienting the optic over the water with a ~1.5 m boom to obtain surface reflectance measurements. To minimize noise and increase the signal-to-noise ratio (SNR) of the surface water reflectance measurements, each separate spectrum collected was an integration of 30 individual spectra. Multiple 30-spectra averages were then collected as 5 to 10 groups of 8 taken at each site under full ambient illumination, thus 41

yielding a grand average for each site based on up to 1200 to 2400 individual spectra.

Spectra that were saturated or outliers due to poor measurement geometry were discarded and not included in the grand average. The site-averaged reflectance values were then transformed to centered-weighted, first derivatives to remove the low-frequency part of the signal in preparation for varimax-rotated, principal component analysis.

To couple the results of our field sampling and spectra with continuous water- quality data, we utilized the piling-mounted water-quality instrumentation, Kilroys36, provided by colleagues at ORCA. The suite of in-situ instruments were located at 5 of the

11 field sampling sites in the IRL (Figure 2.1b). These devices recorded water-quality parameters like Chl-a and blue/green algae continuously in 15 to 30-minute increments year-round with the occasional interruption for regular maintenance. Additionally, Yellow

Springs InstrumentsTM (YSI) EXO2 Multiparameter Sonde probes were used at all locations to record pH, salinity, turbidity, chlorophyll, and blue-green algae based on phycoerythrin (BGA-PE).

2.3.2 Lab Data

Cell count samples were kept refrigerated until shipped to BSA Environmental

Services for analysis. Each sample was identified to the genus and species level, measuring both the total density (cells/L) and the cells by volume (μm3/L). To compare reflectance data with the cell count results, samples were integrated to the Phylum level. Results were averaged between the two replicate samples at each location.

In addition to collecting water samples for cell count analyses, we measured the reflectance of a filtered Aureoumbra lagunensis culture provided by ORCA laboratories.

Measuring these spectra expanded the catalog of reference pigments in the current spectral library. A. lagunensis was cultured in a 500 mL flask under standard temperature controls and sterile conditions. From the 500 mL of culture, 250 mL of A. lagunensis sample was filtered through 47 mm diameter glass microfiber filters (GF/F) with 0.7 μm pore size at <

15 psi to prevent cell lysing. These filters accumulate the particulate material within the water but do not capture any color dissolved organic matter. Following procedures similar to Ortiz et al. (2013), the two filtered water samples were dried at 60° C for ~1 hour to remove water. Once the filtered samples dried, they were measured using an ASD

FieldSpec® HH2 equipped with a High-Intensity contact probe that utilizes a light source of known optical properties. The device was calibrated against a white SpectralonTM plate

(9 cm diameter) before the sample was placed on a scissor jack and raised up to the device until sealed underneath the contact probe. With the filter in contact with the probe, ambient light was excluded from any measurements. Approximately 1200 individual reflectance values were measured for the particulates accumulated on the GF/Fs as well as for unfiltered blank reference GF/Fs. The sample reflectance spectra were averaged and blank- corrected for the absorption from the white GF/F background using the averaged blank filter measurements. As with the field reflectance measurements, all GF/F reflectance values were averaged to one reflectance spectra for each site on each day and then transformed to centered derivatives. The A. lagunensis spectra were included into a library 43

of known spectral constituents (Kokaly et al., 2017; Ortiz et al., 2019, 2017a), which was necessary for later component loading identification following the VPCA decomposition.

In order to validate the spectral component identifications that were related to less abundant minerals, such as hematite or goethite, SEM-EDS was used to determine the presence of iron within the dried suspended sediment samples. The sediment collected with the clamshell sampler were wet sieved to grain sizes < 63 μm to simulate suspended material that would, in turn, be observed in the initial integrated spectral signature from

Sentinel-3A imagery. After sieving, sediment was dried in an oven at 60° C, hand ground with a mortar and pestle, and adhered to aluminum tacks for SEM-EDS analysis. This analysis was aimed at providing a qualitative assessment as to whether there was a presence of iron-bearing minerals, and therefore mineral stoichiometry from the SEM-EDS was not conducted.

2.3.3 Remote Sensing Image Analysis and VPCA Spectral Decomposition

We obtained and evaluated Sentinel-3A OLCI level-2 products with 300 x 300 m water pixels within ±3 days of the in-situ IRL field samples from the EUMETSAT

Copernicus data repository. The June 28, 2018 image acquired 0-1 days before field sampling was selected for further analysis due to coincident timing and minimal cloud coverage. Using Harris Geospatial ENVI/IDL software, the center-weighted derivative for the 11-visible bands in the Sentinel-3A OLCI spectra were calculated.

We conducted two separate varimax-rotated, principal component analyses

(VPCA), on the ASD Fieldspec® HH2 and Sentinel-3A OLCI derivative spectra datasets.

VPCA reduces the dimensionality of multivariate datasets by decomposing the integrated spectral signatures into orthogonal axes that each account for a portion of the total signal variability (Ortiz et al., 2019). The derivative spectra from images and field samples were analyzed with a forward PCA rotation, which computes the eigenvalues and eigenvectors.

This was then further transformed by a varimax rotation (Kaiser, 1958), providing better separation of the component loading coefficients, while preserving the orthogonality of the eigenvectors. This improves the interpretably of the resulting components (Kaiser, 1958).

The eigenvectors describe the axis of the new system, while the eigenvalues describe how long each axis is, providing a measure of the percent variance described. The number of spectral components to extract from the VPCA was set to the number of eigenvalues greater than 1, plus one additional component. From the varimax rotation the results yield component loadings, which represent independent spectral signatures of constituents in the water column, and component scores, which represent the spatial distribution of these spectral signatures. For satellite images, the component scores were displayed as a distribution maps, which represent the fractional variance associated with each component at each pixel, while the component scores of the field spectra were represented as single location values at the field sampling site. These orthogonal VPCA component loadings and scores were calculated using code employed by (Avouris & Ortiz, 2019; Ortiz et al.,

2019, 2017a) for multispectral satellite imagery and hyperspectral spectroradiometer data. 45

2.3.4 Validation and Spectral Identification

The VPCA component loadings obtained for all datasets were identified by forward stepwise regression against a spectral constituent library (Avouris & Ortiz, 2019; Ortiz et al., 2019), with the inclusion of the additional spectra measured from the filtered A. lagunensis culture. In addition to standard regression statistics (R2, F-value, p-value), the level of multicollinearity was assessed using the variance inflation factor (VIF). The individual VPCA loadings were fit to as many matching spectral constituents in the library with a stopping criterion set to values of the VIF less than 2 to minimize the risk of over fitting (Ortiz et al., 2019). Once the spectral signatures of all component loadings were identified using the constituent library, the component loadings and scores for the two datasets were correlated with each other. Correlations between the two datasets were useful to validate consistency of the KSU VPCA spectral decomposition method across the hyper- and multispectral instruments. To validate spectral identifications with A. lagunensis and other potential constituents, we correlated the component scores from the field spectra and extracted component score pixel values from the VPCA distribution map of the Sentinel-3A image to the phylum cell count density results obtained for each sample site.

2.4 Results

2.4.1 VPCA Spectral Identifications

The results from the VPCA spectral decomposition of Sentinel-3A imagery and field measured spectra both yield four unique varimax-rotated, principal components

(VPCs). For each VPC, the loading patterns are regressed and matched with known spectral signatures from a spectral library. In total, the components from the Sentinel-3A

OLCI decomposition account for 92.9% of the total signal variability. Sentinel-3A VPC loading 1 has a positive correlation with illite and negative correlation with alpha- phycoerythrocyanin (α-PEC), a cyanobacterial pigment, accounting for 36.8% of the total signal variability (Figure 2.2). Component 2 and 3 have a positive correlation with A. lagunensis spectra, but account for its association with different types of suspended sediment. Component 2 accounts for 30.4% of variability with a positive correlation between A. lagunensis spectra, and goethite, but a negative correlation with Cyanophyta pigments. Component 3 correlates with A. lagunensis and illite + kaolinite, but accounts for a smaller 16% of variability. Component 4 has a positive correlation with allophycocyanin and a negative correlation to Chlorophyll-b, accounting for 9.7% of the signal variability. The individual components, which represent the partitioned variance from the combined in-water constituents, can be linked to their spatial distribution within the IRL as component scores (Figure 2.2). Based on the spatial distribution map, component 2 and 4 were concentrated highest in the Banana River region of the IRL.

Although measured with different instruments at different spatial scales (300 m vs.

~35 cm), and at only a subset of the locations within the IRL sampled by Sentinel-3A

OLCI, the four component loadings produced from field data collected with an ASD

FieldSpec® HH2 at near coinciding locations exhibit similar constituent identifications and spatial patterns to those found with the Sentinel-3A OLCI analysis. The HH2 VPC 1 loading accounts for 48.4% of total signal variability and has a positive correlation with A. lagunensis and kaolinite. HH2 VPC loading 2 accounts for 31% of variability across all sites and has a positive correlation with illite and allophycocyanin. HH2 VPC 3 loading accounts for 10.5% of variability and has a positive correlation with α-PEC and illite + kaolinite. HH2 VPC loading 4 accounts for 5% of variability and has a positive correlation with α-PEC and Chlorophyll-a + carotenoids.

Comparing the results of the two independent VPCA spectral decompositions, we find significant correlation between the spectral shapes of the component loadings collected with the hyperspectral HH2 and the multispectral Sentinel-3A OLCI imagery.

The spectral shapes of OLCI VPC 2 loading shows similar absorption peaks to lab measured A. lagunensis cultures and the VPC 1 loading of the HH2 spectra (Figure 2.3a).

When correlating the component scores (spatial patterns) of these two data sets to each other we find a positive correlation, with R2 = 0.85 and p < 0.05 (Figure 2.3b).

2.4.2 Field Validation

When the two spectroradiometers were sampled at their common locations (Figure

2.1b), the spatial pattern of the retrievals from the OLCI sensor and the HH2 were highly statistically correlated (Figure 2.3b). The sample locations that had the highest cell counts for Ochrophyta were highest in the sample sites from the Banana River. When comparing the component scores of the HH2 field spectra VPC 1 to the cell counts we find a positive correlation to Ochrophyta biovolume counts (Figure 2.4a) with R2 = 0.92 (df = 9, p <

0.001). Likewise, when comparing the component scores of the OLCI VPC 2 to the cell counts, we find a positive correlation of Ochrophyta with A. lagunensis component spectra

(Figure 2.4b), with R2 = 0.92 (df = 7, p < 0.001).

Sediment analyses by SEM-EDS provides useful information on the elemental chemistry and a qualitative assessment of minerals present in the IRL through visual inspection. In all sediments samples the bulk mineral composition contained primarily clay minerals with the inclusion of modern marine shell fragments and sand-sized grains of quartz with clear characteristics of conchoidal fracturing. From bulk elemental composition, all sediment samples contained minor grains of primarily iron-bearing material. While no definite mineral crystal structure could be observed, the size of these iron-bearing grains range on average between approximately 10 μm and 30 μm (Figure

2.5).

The Chl-a and blue/green algae phycocyanin concentrations obtained from moored,

Ocean Research and Conservation Association (ORCA) deployed Kilroy environmental 49

monitoring systems provided a time series of measurements that indicate the stage of bloom development during the span of two years. When observing the Kilroy time series measurements within the Banana River region where high Ochrophyta counts are detected, such as Sykes Creek, Chl-a and blue/green algae concentrations range between 160-340

μg/mL for blue/green algae and 140-260 μg/mL for Chl-a (Figure 2.6Figure 2.6). These concentrations occur during one of the four highest recorded concentration value periods over the two-year span from May 2017 to May 2019. From Secchi depths recordings, the lowest visibility appears to be concentrated in the areas of the Banana River, where OLCI

VPC 2 has a high positive correlation with the YSI probe measurements for Chlorophyll, blue/green algae phycoerythrin (BGA-PE), and pH but, a negative correlation with salinity

(Table 2.1).

2.5 Discussion

Our component loading identification results show that it is possible to isolate the

A. lagunensis signal, with the inclusion of some minor suspended sediment constituents, from the integrated spectral signature of water to identify the region of the IRL most affected by Brown Tide. This is possible even when the A. lagunensis signal represents only on the order of 30% of the total image variance. The presence of matching VPCA components obtained from both the ASD FieldSpec® HH2 and Sentinel-3A OLCI decomposition further validates these results. Even when the HH2 component loadings were downgraded to the Sentinel-3A OLCI’s 11-band spectral resolution, we still obtain

statistically significant correlations between similarly identified components, specifically,

HH2 VPC 1 loading compared with OLCI VPC 2 loading 2 for A. lagunensis (Figure 2.3).

Focusing on the OCLI VPC 2 loading spatial distribution map (Figure 2.2), we observe the

Brown Tide bloom distribution was concentrated within the Banana River. We provide validation for the distribution of these components through the statistically significant correlations to cell count results, which were linear across the full response range observed with R2 > 0.9. The highest Ochrophyta biovolume counts were concentrated in the Banana

River region of the IRL where several unrecorded outdated sewage treatment plants overflow during heavy rains and much of the nutrient pollution from septic runoff occurs

(Barile, 2018). However, the signal was not identified down to the species level of A. lagunensis. With this caveat in mind, it is important to note that we can only be certain that the A. lagunensis-related VPCs are delineating the spatial distribution of Brown Tide phylum-related constituents. According to the Chlorophyll-a and blue/green algae phycoerythrin pigment concentrations measured at the Sykes Creek Kilroy, during the period in which the high counts of Ochrophyta were measured in the Banana River, the seasonal Chlorophyll-a and blue/green algae phycoerythrin pigment concentrations were high. This indicates that the A. lagunensis constituent detected occurred during a well- developed stage of HAB, which was also associated with cyanobacteria in late June of

2018. Our results in the Banana River agree with previous field studies that observe high seasonal biovolume counts of phytoplankton in this region during warmer temperatures and greater rainfall (Chamberlain & Hayward, 1996; Phlips et al., 2011). 51

From the SEM-EDS analyses, the sediment samples contain some type of iron- bearing minerals or iron-coated sediment grains. The confirmation of clay minerals and iron presence in our field samples provides sufficient validation for the type of mineralogical constituents identified in the OLCI VPCA decomposition, although there is ambiguity as to whether these iron grains were hematite, goethite, or some other iron- bearing mineral on the basis of the backscatter data. In addition, these results are consistent with previous surveys of the IRL, which have found iron in water samples (Trocine &

Trefry, 1996). From the Secchi depth and turbidity measurements, it appears that the visibility was lower in the Banana River as opposed to the northern IRL (Table 2.1).

Ecologically, it is plausible for an alga such as A. lagunensis to occur within an environment more turbid due to self-shading or suspended sediment because it does not require as much sunlight as other algae (Liu et al., 2001). Aside from A. lagunensis’s adaptation to lower-light conditions compared to many cyanobacteria, the limiting nutrient necessary to facilitate their growth is nitrogen in the form of ammonium (Backer, Fleming,

Rowan, & Baden, 2003; Liu & Buskey, 2000; Liu et al., 2001). There have been well documented cases of ammonium release from OSDS as the likely source of the nitrogen that drives A. lagunensis blooms (Lapointe et al., 2015, 2017). Historically, A. lagunensis occurs in the Banana River where most septic waste is treated by residential OSDS (Barile,

2018). The presence of a highly permeable, sandy subsurface throughout the Florida coastline could harbor an environment that was susceptible to more frequent ammonium leaching into to the IRL. Our results validate previous findings from past field studies 52

through our remote sensing methods, that A. lagunensis blooms occur in the Banana River.

Legacy nutrient pollution from growing agricultural and other human inputs could also provide a viable source of nitrogen to fuel the Brown Tides (Dunne et al., 2011; Reddy et al., 2011; Yang et al., 2013). While the Secchi depth and turbidity variables are positively correlated with OLCI VPC 2 scores, without field measurements for suspended sediment,

N, and P, we cannot determine whether A. lagunensis blooms are also potentially linked to resuspension of legacy nutrients within the IRL. Furthermore, our field measurements for cell counts show high Ochrophyta in these Banana River areas of high associations with

Brown Tide components. This indicates that moderate resolution imagers can be used to locate and source nutrient polluted areas in large bodies of water. The excellent comparison between the ASD HH2 (~35 cm) and OLCI (300 m) despite differences in spatial resolution also indicates that the method developed here is independent of the spatial scale of the results, which opens tremendous opportunities. Stakeholders can compare results between sensors to make integrative policy decision. The small spot size of the handheld sensor allows application to lakes and ponds smaller than can be studied with orbital remote sensing using handheld sensors or sensor-equipped drones. The ability for data fusion also has implications for inter-comparison of results from proposed orbital hyperspectral missions such as the Plankton, Aerosol, Cloud, ocean Ecosystem (PACE) mission (1 km spot size), and a Surface Biology and Geology (SBG) mission concept as recommended in the NASA decadal survey, which will likely have resolution in coastal and inland water that is on the order of 30-60m. 53

2.6 Conclusion

The Indian River Lagoon is a large and optically complex estuary. By using the

KSU VPCA method to partition the integrated optical water signal into individual constituents, rather than employing traditional Chlorophyll-a algorithms, we can separate out spectral signatures that have similar reflectance peaks, and better elucidate the location and composition of the Brown Tide. Given correlations of the A. lagunensis-related OLCI

VPCA component loadings to Ochrophyta cell counts and relatively higher suspended sediment in brackish salinity conditions, our results indicate an environment suitable for abundant Brown Tide blooms (Liu & Buskey, 2000). While we hypothesize that the primary driver for Brown Tides could be from septic runoff, our analyses do not provide enough evidence to rule out other potential drivers such as fertilizer runoff or legacy nutrient sources. Future field sampling would require tests for fecal coliform and elemental analyses of water samples for N and P to discriminate between the multiple, potential sources. Regardless of which nutrient source within the IRL is dominant, we can conclude that the distributions of the A. lagunensis-related spectral components have a high association in the region of the Banana River, along with strong correlations between different remote sensing instrumentation and cell count information. Moving forward, because the A. lagunensis component loading from the OLCI VPCA decomposition were highly correlated with Chl-a, can determine the threshold of Brown Tide detection in the

IRL using the KSU VPCA method. By determining the minimum Chl-a concentration necessary for Brown Tide spectral constituents to still be retrieved from the VPCA 54

decomposition, we can discern the minimum quantitative detection limits for Brown Tide using this remote sensing method. Utilizing this method has shown that if spectral signatures of organisms representative to the region of interest are added to the spectral library, then VPCA components can be better characterized to known constituents and correlated to field cell count measurements. By expansion of the spectral library with additional spectra and implementing time series analysis of imagery that can be correlated with additional cell counts, we can expand the usefulness of the library for additional environments.

2.7 References

Ali, K. A., Witter, D. L., & Ortiz, J. D. (2014). Multivariate approach to estimate colour

producing agents in Case 2 waters using first-derivative spectrophotometer

data. Geocarto International, 29(2), 102-127.

Ali, K. A., Ortiz, J., Bonini, N., Shuman, M., & Sydow, C. (2016). Application of Aqua

MODIS sensor data for estimating chlorophyll a in the turbid Case 2 waters of Lake

Erie using bio-optical models. GIScience & Remote Sensing, 53(4), 483-505.

Avouris, D. M., & Ortiz, J. D. (2019). Validation of 2015 Lake Erie MODIS image spectral

decomposition using visible derivative spectroscopy and field campaign data. Journal

of Great Lakes Research.

Backer, L. C., Fleming, L. E., Rowan, A. D., & Baden, D. G. (2003). Epidemiology, public

health and human diseases associated with harmful marine algae. Manual on harmful

marine microalgae, 725-750.

Barile, P. J. (2018). Widespread sewage pollution of the Indian River Lagoon system,

Florida (USA) resolved by spatial analyses of macroalgal biogeochemistry. Marine

pollution bulletin, 128, 557-574.

Chamberlain, R., & Hayward, D. (1996). Evaluation of water quality and monitoring in the

St. Lucie Estuary, Florida. JAWRA Journal of the American Water Resources

Association, 32(4), 681-696.

DeYoe, H. R., Stockwell, D. A., Biolagare, R. R., Latasa, M., Johnson, P. W., Hargraves,

P. E., & Suttle, C. A. (1997). Description and characterization of the algal species 56

Aureoumbra lagunensis gen. et sp. nov. and referral of Aureoumbra and Aureococcus

to the Pelagophyceae. Journal of Phycology, 33(6), 1042-1048.

Dunne, E. J., Clark, M. W., Corstanje, R., & Reddy, K. R. (2011). Legacy phosphorus in

subtropical wetland soils: influence of dairy, improved and unimproved pasture land

use. Ecological engineering, 37(10), 1481-149 1.

Fox, A.L, Trefry, J.H. 2018. “Environmental Dredging to Remove Fine-Grained,

OrganicRich Sediments and Reduce Inputs of Nitrogen and Phosphorous to a

Subtropical Estuary.”, Marine Technology Society Journal, 52 (4).

Gobler, C. J., & Sañudo-Wilhelmy, S. A. (2001). Temporal variability of groundwater

seepage and brown tide blooms in a Long Island embayment. Marine Ecology Progress

Series, 217, 299-309.

Gobler, C. J., & Sunda, W. G. (2012). Ecosystem disruptive algal blooms of the brown tide

species, Aureococcus anophagefferens and Aureoumbra lagunensis. Harmful

Algae, 14, 36-45.

Gobler, C. J., Koch, F., Kang, Y., Berry, D. L., Tang, Y. Z., Lasi, M., ... & Miller, J. D.

(2013). Expansion of harmful brown tides caused by the pelagophyte, Aureoumbra

lagunensis DeYoe et Stockwell, to the US east coast. Harmful Algae, 27, 29-41.

Hall, N. S., Litaker, R. W., Kenworthy, W. J., Vandersea, M. W., Sunda, W. G., Reid, J.

P., ... & Butler, S. (2018). Consortial brown tide− picocyanobacteria blooms in

Guantánamo Bay, Cuba. Harmful algae, 73, 30-43.

Kaiser, H. F. (1958). The varimax criterion for analytic rotation in factor analysis.

Psychometrika, 23(3), 187-200.

Kamerosky, A., Cho, H., & Morris, L. (2015). Monitoring of the 2011 super algal bloom

in Indian River Lagoon, FL, USA, using MERIS. Remote Sensing, 7(2), 1441-1460.

Kang, Y., Koch, F., & Gobler, C. J. (2015). The interactive roles of nutrient loading and

zooplankton grazing in facilitating the expansion of harmful algal blooms caused by

the pelagophyte, Aureoumbra lagunensis, to the Indian River Lagoon, FL, USA.

Harmful Algae, 49, 162-173.

Koch, F., Kang, Y., Villareal, T. A., Anderson, D. M., & Gobler, C. J. (2014). A novel

immunofluorescence flow cytometry technique detects the expansion of brown tides

caused by Aureoumbra lagunensis to the Caribbean Sea. Appl. Environ. Microbiol.,

80(16), 4947-4957.

Kokaly, R. F., Clark, R. N., Swayze, G. A., Livo, K. E., Hoefen, T. M., Pearson, N. C., ...

& Klein, A. J. (2017). USGS spectral library version 7 (No. 1035). US Geological

Survey.

Lapointe, B. E., Herren, L. W., Debortoli, D. D., & Vogel, M. A. (2015). Evidence of

sewage-driven eutrophication and harmful algal blooms in Florida's Indian River

Lagoon. Harmful Algae, 43, 82-102.

Lapointe, B. E., Herren, L. W., & Paule, A. L. (2017). Septic systems contribute to nutrient

pollution and harmful algal blooms in the St. Lucie Estuary, Southeast Florida,

USA. Harmful algae, 70, 1-22. 58

Liu, H., & Buskey, E. J. (2000). Hypersalinity enhances the production of extracellular

polymeric substance (EPS) in the Texas brown tide alga, Aureoumbra lagunensis

(Pelagophyceae). Journal of Phycology, 36(1), 71-77.

Liu, H., Laws, E. A., Villareal, T. A., & Buskey, E. J. (2001). Nutrient‐limited growth of

Aureoumbra lagunensis (Pelagophyceae), with implications for its capability to

outgrow other phytoplankton species in phosphate‐limited environments. Journal of

Phycology, 37(4), 500-508.

McNeal, B. L., Stanley, C. D., Graham, W. D., Gilreath, P. R., Downey, D., & Creighton,

J. F. (1995). Nutrient-loss trends for vegetable and citrus fields in West-Central Florida:

I. Nitrate. Journal of environmental quality, 24(1), 95-100.

Metcalf, J.S., Banack, S.A., Powell, J.T., Tymm, F.J.M., Murch, S.J., Brand, L.E., & Cox,

P.A. (2018). Public health responses to toxic cyanobacterial blooms: perspectives from

the 2016 Florida event. Water Policy 20, 919–932.

Ortiz, J. D., Witter, D. L., Ali, K. A., Fela, N., Duff, M., & Mills, L. (2013). Evaluating

multiple colour-producing agents in Case II waters from Lake Erie. International

journal of remote sensing, 34(24), 8854-8880.

Ortiz, J., Avouris, D., Schiller, S., Luvall, J., Lekki, J., Tokars, R., ... & Becker, R. (2017).

Visible Derivative Spectroscopy of Multispectral and Hyperspectral Images: A New

Approach to Algal and Cyanobacterial Differentiation.

Ortiz, J. D., Avouris, D. M., Schiller, S. J., Luvall, J. C., Lekki, J. D., Tokars, R. P., ... &

Becker, R. (2019). Evaluating visible derivative spectroscopy by varimax-rotated, 59

principal component analysis of aerial hyperspectral images from the western basin of

Lake Erie. Journal of Great Lakes Research.

Phlips, E. J., Badylak, S., Christman, M., Wolny, J., Brame, J., Garland, J., ... & Lockwood,

J. (2011). Scales of temporal and spatial variability in the distribution of harmful algae

species in the Indian River Lagoon, Florida, USA. Harmful Algae, 10(3), 277-290.

Reddy, K. R., Newman, S., Osborne, T. Z., White, J. R., & Fitz, H. C. (2011). Phosphorous

cycling in the greater Everglades ecosystem: legacy phosphorous implications for

management and restoration. Critical Reviews in Environmental Science and

Technology, 41(S1), 149-186.

Sime, P. (2005). St. Lucie estuary and Indian River Lagoon conceptual ecological

model. Wetlands, 25(4), 898.

Thosteson, E. D., Widder, E. A., Cimaglia, C. A., Taylor, J. W., Burns, B. C., & Paglen,

K. J. (2009, May). New technology for ecosystem-based management: Marine

monitoring with the orca kilroy network. In Oceans 2009-Europe (pp. 1-7). IEEE.

Trocine, R. P., & Trefry, J. H. (1996). Metal concentrations in sediment, water and clams

from the Indian River Lagoon, Florida. Marine Pollution Bulletin, 32(10), 754-759.

Witter, D. L., Ortiz, J. D., Palm, S., Heath, R. T., & Budd, J. W. (2009). Assessing the

application of SeaWiFS ocean color algorithms to Lake Erie. Journal of Great Lakes

Research, 35(3), 361-370.

Zhang, M. K., Wang, L. P., & He, Z. L. (2007). Spatial and temporal variation of nitrogen

exported by runoff from sandy agricultural soils. Journal of Environmental

Sciences, 19(9), 1086-1092.

Yang, Y., He, Z., Wang, Y., Fan, J., Liang, Z., & Stoffella, P. J. (2013). Dissolved organic

matter in relation to nutrients (N and P) and heavy metals in surface runoff water as

affected by temporal variation and land uses–A case study from Indian River Area,

south Florida, USA. Agricultural water management, 118, 38-49.

2.8 Figures

Figure 2.1: Location map of the Northern Indian River Lagoon. (a) True color composite image from Landsat-8 OLI acquired on November 17, 2017. (b) Location of in-situ measurements obtained for both cell count analysis, hyperspectral ASD measurements, and Secchi depths on June 29th and 30th of 2018. Sites labeled as Turnbull, Haulover, Barge, Sykes, and Dragon Point are locations of piling-mounted Kilroy instruments for water measurements.

e pixels pixels e

onent loadings from

indicate a lower association. lower a indicate

h) h) of comp

rs indicate higher association of th of association higher indicate rs

spatial distribution maps spatial (e distribution maps

d) d) and

3A OLCI. In all images warm colo warm images all In OLCI. 3A

Indian River Indian River Lagoon (a spectral

: :

re re

Figu with the spectral constituent mixture represented represented this mixture colors pattern, by while cool constituent withspectral the 63 Sentinel from swath 2018 28, June

Figure 2.3: Validation graphs showing spectral signature of the A. lagunensis culture compared with VPCA component loadings for both satellite and field measured datasets at the 11-band resolution of Sentinel-3A OLCI (a). Comparison between the component scores of field spectra and Sentinel-3A that represent constituents of A. lagunensis (b).

Figure 2.4: Validation graphs showing the comparison between Ochrophyta phylum from cell count analysis with VPCA field spectra component scores (a) and Sentinel-3A OLCI component scores (b).

Figure 2.5: SEM and EDS mapping images of sediment samples obtained from (a) Sykes Creek and (b) NIRL 2 sample sites. Images (red) show the detection of Fe on grains sizes < 30 μm. These SEM images indicate that iron-bearing or iron-coated minerals or sediment grains are present to within deposits at the IRL and Banana River.

Figure 2.6: Sykes Creek Kilroy measurements for Chlorophyll-a and blue/green algae from May 18, 2017 to May 22, 2019 (http://api.kilroydata.org/public/). Analysis of Sentinel-3A OLCI was obtained on June 28, 2018 (red arrow) when higher seasonal counts of Chlorophyll-a and blue/green algae are measured.

2.9 Table

Table 2.1: Manually recorded Secchi depths and optical depth, along with Yellow Springs InstrumentsTM (YSI) EXO2 Multiparameter Sonde probe measurements for salinity, turbidity, chlorophyll, and blue/green algae and phycoerythrin (BGA-PE) obtained at all 11 sample locations on June 29-30, 2018.

Variability of the Brown Tide in the Indian River Lagoon from 2017 to 2019 based

on Sentinel-3A OLCI VPCA spectral decomposition

3.1 Introduction

The 5,600 km2 coastal estuary of the Indian River Lagoon (IRL), Florida encompasses three major lagoons, which have undergone various degrees of ecological stress: Mosquito Lagoon in the north, Banana River in the northeast, and the major body of water, Indian River, which extends from Volusia County to Palm Beach County (Figure

3.1). Over the last 100 years, growing agricultural industries, expansion of coastal populations, and major storm events have driven demands for flood control management of the Florida watersheds (Harvey, Loftus, Rehage, & Mazzotti, 2010; Sime, 2005). These flood control projects have altered the natural hydrology of this wetland estuary to a largely urbanized landscape and increased the amount of stormwater discharge into the IRL. As a result, freshwater flowing into areas of the IRL now carry eutrophic runoff concentrated in nitrogen and phosphorus from fertilized farmlands and septic systems, along with biogenic particles and the resuspension of legacy nutrients adsorbed onto sediment in the shallow lagoon waters during periods of increased turbidity (Barile, 2018; Dunne et al., 2011; Fox

& Trefry, 2018; Lapointe et al., 2015, 2017; McNeal et al., 1995; Ming-kui, Li-ping, &

Zhen-li, 2007; Reddy et al., 2011; Yang et al., 2013). Nutrient-rich waters flowing into the

IRL have caused eutrophication and growth of various types of harmful algal blooms

(HABs). These toxic water conditions devastate the local aquatic ecosystem, cause health problems for citizens from exposure to the cyanobacterial hepatotoxin, microcystin, and hinder tourism and fishing industries vital to Florida’s coastal economy (Backer et al.,

2003; Kirkpatrick et al., 2004; Metcalf et al., 2018; Steidinger et al., 1999).

In response to increasing HAB occurrences in both Florida and across the United

States, efforts have been made to characterize the types of toxic species (Steidinger et al.,

1999), the limiting conditions (e.g. nutrients, light, temperature, and salinity) necessary for growth, the locations of bloom reoccurrence, and their effects on local ecosystems (Capper et al., 2013) through field measurements and satellite imagery (Phlips et al., 2011, 2015).

In the past, nutrient-rich stormwater runoff from agricultural fertilizers has accumulated in

Lake Okeechobee and discharged into areas of the southern IRL (Lapointe, Herren, &

Bedford, 2012). These eutrophic fresh waters have provided habitable conditions for blue- green algae growth as seen in the St. Lucie blooms during 2013 and 2016 (Kramer et al.,

2018; Oehrle, Rodriguez-Matos, Cartamil, Zavala, & Rein, 2017). On the other hand,

Florida’s mostly permeable sandy soil (Lane, 1987; Zhang et al., 2007), low wind-driven water circulation (Smith, 1993), and ammonium leaching from onsite sewage disposal systems, primarily in the northern lagoons (Barile, 2018; Koch et al., 2014), has allowed suitable conditions for the Brown Tide organism, Aureoumbra lagunensis, to thrive

(DeYoe et al., 1997; C. J. Gobler & Sunda, 2012b, 2012a; Hall et al., 2018; Kang & Gobler,

2018; Kang et al., 2015; Liu & Buskey, 2000; Liu et al., 2001; Rhudy, Sharma, Lehman,

& McKee, 1999). 69

Following these events, the Comprehensive Everglades Restoration Plan (CERP), has since commenced implementation to reroute the drainage of Okeechobee waters that currently flow into the St. Lucie and Caloosahatchee River, to a southern drainage path into the Everglades (Mclean et al., 2002). However, the rerouting of Lake Okeechobee’s drainage path is a 35-year project that remains under construction. Regardless of whether this project proves to be effective, it still does not address nutrient pollution and wastewater leaching in the northern lagoons that do not receive input from the St. Lucie River drainage.

While prior field research has provided crucial insight on the locality of certain HAB species and nutrient pollution sources, the costs, region of interest, and time required to obtain and analyze field samples has limited the ability to inform local government and guide remediation efforts in a timely manner. With billions of dollars being invested into watershed restoration projects, such as CERP, automated in-situ water quality instruments and reliable satellite-based remote sensing imagery could become essential tools for quickly informing the public and implementing smart flood control policies that more effectively improve the long-term water quality across the IRL.

In addition to traditional field-based water quality monitoring and cell count analyses, satellite-based remote sensing becomes a vital resource to track the seasonal spatial variability of blooms across large water bodies such as the IRL where field monitoring can struggle to retrieve high spatial and temporal coverage. This is a challenging problem because high spatial, temporal and spectral resolution is required for optimal monitoring. High spatial resolution is needed to allow work in the narrow 70

waterways of the IRL. High temporal resolution is needed because conditions can change rapidly in response to environmental variability. High spectral resolution is needed to provide more diagnostic information to identify different types of algae and cyanobacteria using their spectral signatures. No sensor currently in orbit can meet all these requirements.

The Sentinel-2A/B and Landsat series sensors provide fine spatial resolution at the expense of high temporal and spectral resolution. The MODIS (Moderate Resolution Imaging

Spectroradiometer) Aqua and Terra and Sentinel-3A/B Ocean and Land Colour Imager

(OLCI) sensors provide finer spectral resolution and daily sampling under cloud free conditions but have spatial resolutions of 1000 m and 300 m respectively. In Chapter 1, we addressed the payload caveats for each satellite sensor and determined Sentinel-3A/B

OLCI to the most practical imagery in this study.

Previous remote sensing studies have applied finer spectrally resolved imagery from satellites such as MODIS to track the spatial distribution of HABs by employing ocean color chlorophyll-a algorithms (Kamerosky et al., 2015; O’Reilly et al., 2000) to track changes in blue-green algae within Lake Okeechobee and Red Tide (Karenia brevis) blooms along Florida and the Gulf Coast over time (Carvalho et al., 2011, 2010; Tomlinson et al., 2009). However, the narrow waterways of the IRL make the relatively coarse spatial resolution MODIS images less useful than the finer 300 m spatial resolution imagery acquired from the Sentinel-3A OLCI. Historically, identifying the optical properties of water has been difficult because the spectra retrieved from the water column provide an

integrated signal of multiple constituents. The constituents mixed within the water column could be pigments from HAB organisms, other algae and cyanobacteria, color dissolved organic matter (CDOM), and/or suspended sediment. The usefulness of chlorophyll-a algorithms is limited when the signal retrieved is dominated by either more than one type of algae or cyanobacteria, or otherwise contaminated by extraneous signals such as CDOM or suspended sediment. Due to this drawback, often HABs can be either misidentified or cannot be detected until the integrated spectral signature is dominated by the green or red end of the spectra after suspended sediment and HABs have grown beyond the point of early remediation. In order to limit nutrient pollution early on, it is critical to improve satellite spectral resolution and implement image processing methods that detect signal components at lower concentrations.

The Kent State University varimax-rotated, principal component analysis (KSU

VPCA) has previously been shown to account for variability within the mixed water column signal and partition this mixed signal into independent components that can be identified by comparison with a library of known spectral constituents, account for greater than 90% of the total signal variability, and increase the signal-to-noise ratio by removing atmospheric interference and signal cross-talk (K. Adem Ali et al., 2016; Khalid A. Ali,

Witter, & Ortiz, 2014b; Avouris & Ortiz, 2019; Lekki et al., 2017; Ortiz et al., 2019, 2017b,

2013; Witter et al., 2009). In Chapter 2, we used the KSU VPCA spectral decomposition on Sentinel-3A OLCI imagery to address the question of whether Brown Tide or A. lagunensis can be detected if we were to add the necessary spectra to the constituent library. 72

We also coupled these data collections with a suite of the Ocean Research and

Conservations Association (ORCA) Kilroy water-quality sensor packages (Figure 3.2). At the time of the original OLCI imagery decomposition Brown Tide detection on June 28,

2018, Ochrophyta Phylum counts from field sampling were as high as ~7.8 x 1010 μm3/L, while average Kilroy measurements for Chl-a and blue/green algae (BGA) were 175 μg/L and ~220 μg/L in areas such as the Sykes Creek station within the Banana River region of the IRL. These field measurements were observed in the same areas where high spatial distributions of OLCI and hyperspectral spectroradiometer VPC spectral signatures for A. lagunensis were detected in the Banana River. Three questions now arise following this investigation. First, can we determine the consistency of the spectral shapes that are retrieved from the VPCA spectral decomposition over the span of a two-years period.

Second, can we determine the minimum detection limit of Brown Tide in OLCI VPCA components relative to Chl-a and BGA concentrations. Third, how has the distribution of other components varied independently over the same time period? Given the high correlations of Chl-a and BGA to A. lagunensis-related components shown in Chapter 2, we hypothesize (1) that the VPCA for a suite of 10 images will produce the same reoccurring spectral shapes, which can be characterized into individual orthogonal groups that account for a different percentage of the total variance from day to day, and (2) for a suite of images processed using the VPCA decomposition we can determine a detection limit of Brown Tide by the level of signal variance it accounts for with respect to seasonal changes in Chl-a and BGA concentrations, provided the cloud cover is sufficiently low. 73

Furthermore, we expect that as these concentrations decrease, the variance associated with

A. lagunensis-related components will decrease given its association with these water- quality parameters. This study is meant to provide a synthesis of Sentinel-3A OLCI data of the Northern IRL and a conclusive chapter to previous finding from Chapter 2 discoveries.

3.2 Methods

For this study, we utilized the same VPCA spectral decomposition methods as previously outlined in Chapter 2. In order to observe the detection limits of Brown Tide within the Banana River region using the VPCA spectral decomposition method on OLCI imagery, we obtained and evaluated Sentinel-3A OLCI level-2 products with 300 x 300 m water pixels from the EUMETSAT Copernicus data repository for 9 additional images along with the June 28, 2018 image processed in Chapter 2. The total 10 images were selected for further analysis due to minimal cloud coverage and placement within the seasonal Chl-a and BGA concentrations observed at the Sykes Creek Kilroy (Figure 3.3).

The Sykes Creek Kilroy is the only water-quality system that was located within the

Banana River region of the IRL, where much of the A. lagunensis-related VPCA signal was concentrated from Chapter 2 observations. The OLCI images retrieved were from the following dates: 8/1/2017, 10/16/2017, 11/17/2017, 3/9/2018, 3/23/2018, 6/9/2018,

6/28/2018, 8/17/2018, 11/2/2018, and 11/21/2018.

Using Harris Geospatial ENVI/IDL software, the center-weighted derivative for the

11-visible bands in the Sentinel-3A OLCI spectra were calculated, with the derivative for the end bands approximated using a first difference derivative. The varimax-rotated principal component (VPC) loadings obtained for the OLCI datasets were identified by forward stepwise regression against a spectral constituent library (Avouris & Ortiz, 2019;

Kokaly et al., 2017; Ortiz et al., 2019), with the inclusion of the additional spectra measured from the filtered A. lagunensis culture obtained in Chapter 2. In addition to standard regression statistics (R2, F-value, p-value), the level of multicollinearity was assessed using the variance inflation factor (VIF). The individual VPCA loadings were fit to as many matching spectral constituents in the library with a stopping criterion set to VIF values less than 2 to minimize the risk of overfitting (Ortiz et al., 2019). Once the spectral signatures of all component loadings were identified using the constituent library, we compared the spectral loading patterns from each date with each other to qualitatively determine consistent patterns retrieved from the VPCA decomposition. VPC loadings with similar spectral shapes were averaged together to account for fluctuations in the approximately monthly changes in VPC loading derivative peaks. These averaged VPC loadings are assigned individual pattern names and regressed to the constituent spectral library once again to determine the overall spectral identification retrieved during this approximately two-year time span.

3.3 Results

The results from the VPCA spectral decomposition of Sentinel-3A OLCI imagery yield six unique VPCs for all image acquisition dates. For each VPC, the loading patterns which are regressed and matched with known spectral signatures from a spectral library are presented in Table 3.1. On average, the combined components from the Sentinel-3A

OLCI decomposition for each day account for more than 90% of the total signal variability.

There were four different reoccurring VPC loading patterns found, which identify with similar spectral constituents from the reference library that were retrieved from the VPCA spectral decomposition of the 10 OLCI images. The first spectral pattern, Pattern A, accounts for between 28% to 45% of the total signal variability (Figure 3.4a) and is identified as spectral constituents that are generally positively associated with illite and negatively associated with alpha-phycoerythrocyanin (α-PEC) (Table 3.2). Patterns B and

C have a positive correlation with A. lagunensis, but account for associations with different types of suspended sediment and pigments. Pattern B accounts for between 19% to 37% of the total signal variability (Figure 3.4b) and is identified as spectral constituents that are generally positively correlated with A. lagunensis, and goethite and negatively correlated with Cryptophyta (Table 3.2). Pattern C accounts for between 16% to 27% of the total signal variability (Figure 3.4c) and is identified as spectral constituents that are generally positively correlated between A. lagunensis, illite, and chlorophyll-b (Table 3.1). Pattern

D accounts for between 5% to 9% of the total signal variability (Figure 3.4d) and is identified as spectral constituents that are generally positively correlated with 76

allophycocyanin and negatively correlated with chlorophyll-a (Table 3.2). Pattern E, which is only retrieved 3 times from the 10 image decomposition, accounts for between

7% to 16% of the total signal variability and is identified as spectral constituents that are generally positively correlated with phycocyanin and negatively correlated with chlorophyll-b and Cyanophyta (Table 3.2). Pattern F, which is only retrieved once from the 10 image VPCA decomposition, accounts for 6% of the total signal variability and is identified as spectral constituents that are positively correlated with α-PEC and negatively correlated with phaeophytin-b (Table 3.2). The individual components, which represent the partitioned variance from the combined in-water constituents, can be linked to their spatial distribution within the IRL as component scores (Figure 3.5-Figure 3.10). Pattern

A, which associates with suspended sediment constituents, is consistently the first leading

VPC retrieved from the decomposition, given that much of the area has water depth between ~40-300 cm, allowing sediment to be easily resuspended (see CHAPTER 2).

Based on the spatial distribution map, Pattern B was concentrated highest in the Banana

River region of the IRL and appears to radiate from a source in the northern most portion of the Banana River over this two-year time span (Figure 3.6). Pattern C on the other hand, which represent a minor portion of the A. lagunensis spectral signature along with sediment and chlorophyll-b is distributed throughout the IRL, but is more highly concentrated in the

Banana River compared to the Northern IRL, during the lower Chl-a and BGA concentrations of the year such as 10/16/17 for example (Figure 3.7). Throughout the course of the time series Pattern B and C are not identified with the library as A. lagunensis 77

on three individual days: 10/16/17, 11/2/18, and 11/21/18, however these VPC loadings still maintain the same general spectral shape as Pattern B and C respectively. Therefore, we continue to classify Patterns B and C as components associated with Brown Tide.

3.4 Discussion

While the spectral shapes of VPCA loading Patterns B and C are consistently retrieved from the decomposition throughout the time series, when Chl-a is lower these patterns are not identified with A. lagunensis constituents from the library for 11/2/18,

11/21/18 or are minor components in the case of 8/1/17 or 10/16/17. This suggests that minor day to day variations in the spectral shape are starting to interfere with the identification at this level, which is Chl-a of < 80 μg/L. However, to remove that uncertainty, when identifying the lower Chl-a concentration day signals based on the interannual average for each pattern we see that the VPC loadings still correlate positively with their respective interannual spectral patterns. With this consistency in mind, Patterns

B and C are still clearly retrieved from the VPCA decomposition at even the lowest Chl-a and BGA concentration (~4 μg/L and ~7 μg/L respectively) on dates such 8/1/17 (Table

3.3). The interannual spatial variability of Pattern B Brown Tide signal is shown to have association in the northern Banana River. While Pattern C variability accounts a lower proportion of Brown Tide constituents and more sediment constituents, such as illite, the spatial distribution is not only highly associated with the Banana River, but extends in spatial distribution out from the Banana River and further into the Northern IRL and

Mosquito Lagoon during more developed bloom periods (i.e. higher Chl-a concentrations).

This could be due the portion of the Pattern C signal accounting for sediment constituents.

When trying to correlate the interannual changes in the amount of variance explained for each pattern with the daily Kilroy water-quality parameters, we find that there is no correlated with the changes in these variables. This could as be due to the lower temporal sample size of currently available Kilroy data which decreases the degrees of freedom for

Pearson correlations. Pattern A percentage of variance does not correlate with turbidity in day-to-day changes. This could indicate that we are detecting the bottom of the lagoon floor reflectance on some days when water visibility is better.

3.5 Conclusion

From our results we show that some type of Brown Tide species, possibly A. lagunensis or other types, is concentrated in the Banana River region of the IRL and appears to be sourcing from the northern most extent of the Banana River. From this information we predicted that the source of nutrient pollution allowing Brown Tide growth is located within the Banana River. We come to this conclusion due to the highest concentrations of A. lagunensis-related components when looking at the spatial distribution map for Pattern B and C for not only when the Chl-a and BGA concentrations of the Sykes

Creek Kilroy are as high as 380 μg/L, but also during the lower concentration of 4 μg/L.

Furthermore, the study has shown the extent of the VPCA spectral decomposition utility to retrieve algal constituents associated with Brown Tide at significantly lower Chl-a

concentrations. This finding illustrates the effectiveness of this method over the traditional satellite imagery processing techniques which rely more on computing Chlorophyll algorithms to detect the extent of HAB growth in nutrient polluted areas. These results are similar to the findings from Avouris and Ortiz (2019) and Ortiz et al. (2019) in that the

VPCA spectral decomposition retrieves consistent spectral shapes can be grouped into different patterns, in this case 4 patterns, which can describe the development and dynamics of HAB growth across the IRL. Future work will need to obtain a greater sample size of data from the Kilroys water-quality systems to look at seasonal correlations of Pattern A sediment constituents to turbidity, and Pattern B and C A. lagunensis-related constituents to Chl-a and BGA measurements. Additionally, more work needs to be conducted on computing the daily total precipitation across an entire watershed area of the IRL and determine the latency correlation of total daily rainfall to the arrival of HABs in a given watershed.

To summarize, in this research investigation, our objectives were to accomplish the following: (1) apply the KSU VPCA spectral decomposition method for both field hyperspectral water reflectance measurements and satellite remote sensing imagery of the

IRL to determine the leading spectral components in water, such as suspended sediment or pigmented constituents from HABs, other algae and cyanobacteria, (2) determine if the results from spectral analysis can be correlated to cell counts and water quality measurements at the Kilroy sites, and (3) produce a time series of satellite imagery using

the KSU VPCA spectral decomposition method to track the growth of HABs within the

IRL.

In Chapter 1, we developed new methods for utilizing the VPCA decomposition on lower spectral and temporal resolution sensors such as the Landsat-8 OLI and determined

Sentinel-3A OLCI as the most functional imager for this study. Along with improving the final product of VPCA spatial distributions maps for Landsat-8, these imagery results delineated additional sampling locations of interest in the Banana River region which directed subsequent field sampling for the summer of 2018. In Chapter 2, by using the

KSU VPCA method to partition the integrated optical water signal for field spectra and

OLCI datasets into individual constituents, we separated out spectral signatures that have similar reflectance peaks, and better elucidate the location and composition of the Brown

Tide. With the inclusion of A. lagunensis spectra to expand the constituent library and field sampling, we concluded the A. lagunensis-related OLCI VPCA component loadings are correlated not only the VPC retrievals of field spectra, but also with spatial distribution of Ochrophyta cell counts, Chl-a concentrations, and turbidity. The field-based validations for VPCA spectral identification enabled us to address the final objective of producing a time series of VPCA OLCI imagery. The final Chapter 3 synthesis revealed how the KSU

VPCA can consistently retrieve the same average spectral VPC loadings to describe the seasonal spatial variability of constituents such as sediment or Brown Tide over a two-year timespan. In conclusion, the culmination of these research investigations has illustrated the validity and effectiveness of incorporating the KSU VPCA methods across a suite of 81

remote sensing systems to facilitate more timely aerial assessments of bloom dynamics in the IRL and other eutrophic waterbodies.

3.6 References

Ali, K. A., Witter, D. L., & Ortiz, J. D. (2014). Multivariate approach to estimate colour

producing agents in Case 2 waters using first-derivative spectrophotometer

data. Geocarto International, 29(2), 102-127.

Ali, K. A., Ortiz, J., Bonini, N., Shuman, M., & Sydow, C. (2016). Application of Aqua

MODIS sensor data for estimating chlorophyll a in the turbid Case 2 waters of Lake

Erie using bio-optical models. GIScience & Remote Sensing, 53(4), 483-505.

Avouris, D. M., & Ortiz, J. D. (2019). Validation of 2015 Lake Erie MODIS image spectral

decomposition using visible derivative spectroscopy and field campaign data. Journal

of Great Lakes Research.

Backer, L. C., Fleming, L. E., Rowan, A. D., & Baden, D. G. (2003). Epidemiology, public

health and human diseases associated with harmful marine algae. Manual on harmful

marine microalgae, 725-750.

Barile, P. J. (2018). Widespread sewage pollution of the Indian River Lagoon system,

Florida (USA) resolved by spatial analyses of macroalgal biogeochemistry. Marine

pollution bulletin, 128, 557-574.

Capper, A., Flewelling, L. J., & Arthur, K. (2013). Dietary exposure to harmful algal

bloom (HAB) toxins in the endangered manatee (Trichechus manatus latirostris) and

green sea turtle (Chelonia mydas) in Florida, USA. Harmful Algae, 28, 1-9.

Carvalho, G. A., Minnett, P. J., Fleming, L. E., Banzon, V. F., & Baringer, W. (2010).

Satellite remote sensing of harmful algal blooms: A new multi-algorithm method for

detecting the Florida Red Tide (Karenia brevis). Harmful Algae, 9(5), 440–448.

https://doi.org/10.1016/j.hal.2010.02.002.

Carvalho, G. A., Minnett, P. J., Banzon, V. F., Baringer, W., & Heil, C. A. (2011). Long-

term evaluation of three satellite ocean color algorithms for identifying harmful algal

blooms (Karenia brevis) along the west coast of Florida: A matchup assessment.

Remote Sensing of Environment, 115(1), 1–18.

https://doi.org/10.1016/j.rse.2010.07.007.

DeYoe, H. R., Stockwell, D. A., Biolagare, R. R., Latasa, M., Johnson, P. W., Hargraves,

P. E., & Suttle, C. A. (1997). Description and characterization of the algal species

Aureoumbra lagunensis gen. et sp. nov. and referral of Aureoumbra and Aureococcus

to the Pelagophyceae. Journal of Phycology, 33(6), 1042-1048.

Dunne, E. J., Clark, M. W., Corstanje, R., & Reddy, K. R. (2011). Legacy phosphorus in

subtropical wetland soils: influence of dairy, improved and unimproved pasture land

use. Ecological engineering, 37(10), 1481-149 1.

Fox, A.L, Trefry, J.H. 2018. “Environmental Dredging to Remove Fine-Grained,

OrganicRich Sediments and Reduce Inputs of Nitrogen and Phosphorous to a

Subtropical Estuary.”, Marine Technology Society Journal, 52 (4).

Gobler, C. J., & Sunda, W. G. (2012). Ecosystem disruptive algal blooms of the brown tide

species, Aureococcus anophagefferens and Aureoumbra lagunensis. Harmful

Algae, 14, 36-45.

Gobler, C. J., Koch, F., Kang, Y., Berry, D. L., Tang, Y. Z., Lasi, M., ... & Miller, J. D.

(2013). Expansion of harmful brown tides caused by the pelagophyte, Aureoumbra

lagunensis DeYoe et Stockwell, to the US east coast. Harmful Algae, 27, 29-41.

Hall, N. S., Litaker, R. W., Kenworthy, W. J., Vandersea, M. W., Sunda, W. G., Reid, J.

P., ... & Butler, S. (2018). Consortial brown tide− picocyanobacteria blooms in

Guantánamo Bay, Cuba. Harmful algae, 73, 30-43.

Kaiser, H. F. (1958). The varimax criterion for analytic rotation in factor analysis.

Psychometrika, 23(3), 187-200.

Kamerosky, A., Cho, H., & Morris, L. (2015). Monitoring of the 2011 super algal bloom

in Indian River Lagoon, FL, USA, using MERIS. Remote Sensing, 7(2), 1441-1460.

Kang, Y., & Gobler, C. J. (2018). The brown tide algae, Aureococcus anophagefferens and

Aureoumbra lagunensis (Pelagophyceae), allelopathically inhibit the growth of

competing microalgae during harmful algal blooms. Limnology and Oceanography,

63(2), 985–1003. https://doi.org/10.1002/lno.10714

Kang, Y., Koch, F., & Gobler, C. J. (2015). The interactive roles of nutrient loading and

zooplankton grazing in facilitating the expansion of harmful algal blooms caused by

the pelagophyte, Aureoumbra lagunensis, to the Indian River Lagoon, FL, USA.

Harmful Algae, 49, 162-173.

Kirkpatrick, B., Fleming, L. E., Squicciarini, D., Backer, L. C., Clark, R., Abraham, W.,

… Baden, D. G. (2004). Literature review of Florida red tide: implications for human

health effects. Harmful Algae, 3(2), 99–115.

https://doi.org/10.1016/J.HAL.2003.08.005

Koch, F., Kang, Y., Villareal, T. A., Anderson, D. M., & Gobler, C. J. (2014). A novel

immunofluorescence flow cytometry technique detects the expansion of brown tides

caused by Aureoumbra lagunensis to the Caribbean Sea. Appl. Environ. Microbiol.,

80(16), 4947-4957.

Kokaly, R. F., Clark, R. N., Swayze, G. A., Livo, K. E., Hoefen, T. M., Pearson, N. C., ...

& Klein, A. J. (2017). USGS spectral library version 7 (No. 1035). US Geological

Survey.

Lane, E. (1987). Guide to rocks and minerals of Florida (FGS: Special Publication 8

rev.).

Lapointe, B. E., Herren, L. W., Debortoli, D. D., & Vogel, M. A. (2015). Evidence of

sewage-driven eutrophication and harmful algal blooms in Florida's Indian River

Lagoon. Harmful Algae, 43, 82-102.

Lapointe, B. E., Herren, L. W., & Paule, A. L. (2017). Septic systems contribute to nutrient

pollution and harmful algal blooms in the St. Lucie Estuary, Southeast Florida,

USA. Harmful algae, 70, 1-22.

Lekki, J., Anderson, R., Avouris, D., Becker, R., Churnside, J., Cline, M., ... & Ortiz, J.

(2017). Airborne Hyperspectral Sensing of Monitoring Harmful Algal Blooms in the

Great Lakes Region: System Calibration and Validation.

Liu, H., & Buskey, E. J. (2000). Hypersalinity enhances the production of extracellular

polymeric substance (EPS) in the Texas brown tide alga, Aureoumbra lagunensis

(Pelagophyceae). Journal of Phycology, 36(1), 71-77.

Liu, H., Laws, E. A., Villareal, T. A., & Buskey, E. J. (2001). Nutrient‐limited growth of

Aureoumbra lagunensis (Pelagophyceae), with implications for its capability to

outgrow other phytoplankton species in phosphate‐limited environments. Journal of

Phycology, 37(4), 500-508.

McNeal, B. L., Stanley, C. D., Graham, W. D., Gilreath, P. R., Downey, D., & Creighton,

J. F. (1995). Nutrient-loss trends for vegetable and citrus fields in West-Central Florida:

I. Nitrate. Journal of environmental quality, 24(1), 95-100.

McLean, A. R., Ogden, J. C., & Williams, E. E. (2002). Comprehensive Everglades

Restoration Plan. West Palm Beach, Florida.

Metcalf, J.S., Banack, S.A., Powell, J.T., Tymm, F.J.M., Murch, S.J., Brand, L.E., & Cox,

P.A. (2018). Public health responses to toxic cyanobacterial blooms: perspectives from

the 2016 Florida event. Water Policy 20, 919–932.

O’Reilly, J. E., Maritorena, S., Siegel, D., O’Brien, M. C., Toole, D., Mitchell, B. G., ...

& Hooker, S. B. (2001). Ocean color chlorophyll a algorithms for SeaWiFS, OC2,

and OC4: Version 4. SeaWiFS postlaunch technical report series, 11, 9-23.

Ortiz, J. D., Witter, D. L., Ali, K. A., Fela, N., Duff, M., & Mills, L. (2013). Evaluating

multiple colour-producing agents in Case II waters from Lake Erie. International

journal of remote sensing, 34(24), 8854-8880. 87

Ortiz, J., Avouris, D., Schiller, S., Luvall, J., Lekki, J., Tokars, R., ... & Becker, R. (2017).

Visible Derivative Spectroscopy of Multispectral and Hyperspectral Images: A New

Approach to Algal and Cyanobacterial Differentiation.

Ortiz, J. D., Avouris, D. M., Schiller, S. J., Luvall, J. C., Lekki, J. D., Tokars, R. P., ... &

Becker, R. (2019). Evaluating visible derivative spectroscopy by varimax-rotated,

principal component analysis of aerial hyperspectral images from the western basin of

Lake Erie. Journal of Great Lakes Research.

Phlips, E. J., Badylak, S., Christman, M., Wolny, J., Brame, J., Garland, J., ... & Lockwood,

J. (2011). Scales of temporal and spatial variability in the distribution of harmful algae

species in the Indian River Lagoon, Florida, USA. Harmful Algae, 10(3), 277-290.

Phlips, E. J., Badylak, S., Lasi, M. A., Chamberlain, R., Green, W. C., Hall, L. M., …

Steward, J. S. (2015). From Red Tides to Green and Brown Tides: Bloom Dynamics in

a Restricted Subtropical Lagoon Under Shifting Climatic Conditions. Estuaries and

Coasts, 38(3), 886–904. https://doi.org/10.1007/s12237-014-9874-6.

Reddy, K. R., Newman, S., Osborne, T. Z., White, J. R., & Fitz, H. C. (2011). Phosphorous

cycling in the greater Everglades ecosystem: legacy phosphorous implications for

management and restoration. Critical Reviews in Environmental Science and

Technology, 41(S1), 149-186.

Sime, P. (2005). St. Lucie estuary and Indian River Lagoon conceptual ecological

model. Wetlands, 25(4), 898.

Steidinger, K.A., Landsberg, J.H., Tomas, C.R., and Burns, J.W. (1999). Harmful algal

blooms in Florida, Unpublished technical report submitted to the Florida Harmful

Algal Bloom Task Force, Florida Marine Research Institute, 63pp.

Thosteson, E. D., Widder, E. A., Cimaglia, C. A., Taylor, J. W., Burns, B. C., & Paglen,

K. J. (2009). New technology for ecosystem-based management: Marine monitoring

with the ORCA Kilroy Network. OCEANS ’09 IEEE Bremen: Balancing

Technology with Future Needs. https://doi.org/10.1109/OCEANSE.2009.5278229.

Witter, D. L., Ortiz, J. D., Palm, S., Heath, R. T., & Budd, J. W. (2009). Assessing the

application of SeaWiFS ocean color algorithms to Lake Erie. Journal of Great Lakes

Research, 35(3), 361-370.

Zhang, M. K., Wang, L. P., & He, Z. L. (2007). Spatial and temporal variation of nitrogen

exported by runoff from sandy agricultural soils. Journal of Environmental

Sciences, 19(9), 1086-1092.

Yang, Y., He, Z., Wang, Y., Fan, J., Liang, Z., & Stoffella, P. J. (2013). Dissolved organic

matter in relation to nutrients (N and P) and heavy metals in surface runoff water as

affected by temporal variation and land uses–A case study from Indian River Area,

south Florida, USA. Agricultural water management, 118, 38-49.

3.7 Figure

Figure 3.1: Location map of the Northern Indian River Lagoon. (a) True color composite image from Landsat-8 OLI acquired on November 17, 2017. (b) Location of in-situ measurements obtained for both cell count analysis and Kilroy hydrological parameters for VPCA validation in Chapter 2.

Figure 3.2: (a) Image of one Kilroy instrument suite (Thosteson et al., 2009), (b) mounted on submersed piers along five along the northern IRL.

Figure 3.3: Sykes Creek Kilroy measurements for Chlorophyll-a and blue/green algae from August 31, 2017 to January 1, 2019 (http://api.kilroydata.org/public/). Analysis of Sentinel-3A OLCI imagery obtained for all 10 images are indicated with red points along the seasonal counts of Chlorophyll-a and blue/green algae measured for the two-year span.

Figure 3.4: VPC Pattern averages with the combined constituent identified spectra, produced from the VPCA spectral decomposition of all 10 OLCI images in the time series. (a) Pattern A on average represents spectral constituents of + illite and α-PEC. (b) Pattern B on average represents spectral constituents of + Aureoumbra lagunensis, + goethite, and - Cryptophyta. (c) Pattern C on average represents spectral constituents of + Aureoumbra lagunensis, + illite, and + chlorophyll-b. (d) Pattern D on average represents spectral constituents of + allophycocyanin, and - chlorophyll-b.

Figure 3.5: Time series from August 1, 2017 to November 21, 2018 of 10 spatial distribution maps for VPC loading Pattern A. VPC Pattern A, on average represents spectral constituents of + illite and α-PEC. In all images warm colors indicate higher association of the pixels with the spectral constituent mixture represented by this pattern, while cool colors indicate a lower association. 94

Figure 3.6: Time series from August 1, 2017 to November 21, 2018 of 10 spatial distribution maps for VPC loading Pattern B. VPC Pattern B, on average represents spectral constituents of + Aureoumbra lagunensis, + goethite, and - Cryptophyta. In all images warm colors indicate higher association of the pixels with the spectral constituent mixture represented by this pattern, while cool colors indicate a lower association. 95

Figure 3.7: Time series from August 1, 2017 to November 21, 2018 of 10 spatial distribution maps for VPC loading Pattern C. VPC Pattern C, on average represents spectral constituents of + Aureoumbra lagunensis, + illite, and + chlorophyll-b. In all images warm colors indicate higher association of the pixels with the spectral constituent mixture represented by this pattern, while cool colors indicate a lower association. 96

Figure 3.8: Time series from August 1, 2017 to November 21, 2018 of 6 spatial distribution maps for VPC loading Pattern D. VPC Pattern D, on average represents spectral constituents of + allophycocyanin, and – chlorophyll-b. In all images warm colors indicate higher association of the pixels with the spectral constituent mixture represented by this pattern, while cool colors indicate a lower association. 97

Figure 3.9: Time series from August 1, 2017 to November 21, 2018 of 3 spatial distribution maps for VPC loading Pattern E. VPC Pattern E, on average represents spectral constituents of + phycocyanin, and – chlorophyll-b and Cyanophyta. In all images warm colors indicate higher association of the pixels with the spectral constituent mixture represented by this pattern, while cool colors indicate a lower association.

Figure 3.10: Time series from August 1, 2017 to November 21, 2018 of 1 spatial distribution map for VPC loading Pattern F. VPC Pattern F, on average represents spectral constituents of + α-PEC, and – Phaeophytin-b. In all images warm colors indicate higher association of the pixels with the spectral constituent mixture represented by this pattern, while cool colors indicate a lower association.

3.8 Tables

Table 3.1: VPC loading constituent identifications from the spectral library for all 10 Sentinel-3A OLCI image acquisition dates.

OLCI Acquisition Pattern A Pattern B Pattern C Pattern D Pattern E Date + illite + kaolinite + chlorophyll-a Cyanophyta + A. lagunensis + allophycocyanin 8/1/2017 - α-PEC + fucoxanthin + illite - cyanobacteria + allophycocyanin + illite + chlorophyll-a Cyanophyta - phycocyanin + phycocyanin 10/16/2017 - carotenoids Cyanophyta + allophycocyanin - goethite + fucoxanthin + illite + kaolinite + hematite + A. lagunensis + allophycocyanin 11/17/2017 - α-PEC + A. lagunensis + illite - phycocyanin + allophycocyanin + illite + A. lagunensis + A. lagunensis + allophycocyanin 3/9/2018 - α-PEC + goethite + illite + kaolinite + phycocyanin - Chlorophyta + illite + A. lagunensis + chlorophyll-a carotenoids 3/23/2018 - carotenoids Cyanophyta + goethite + illite - goethite + illite + chlorophyll-a Cyanophyta + A. lagunensis + allophycocyanin 6/9/2018 - α-PEC + goethite + illite - chlorophyll-b - neoxanthin + chlorophyll-b + illite + A. lagunensis + A. lagunensis + allophycocyanin 6/28/2018 - α-PEC + goethite + illite + kaolinite - chlorophyll-b - Cyanophyta + illite + A. lagunensis + A. lagunensis + allophycocyanin 8/17/2018 - α-PEC + goethite + illite + kaolinite - chlorophyll-b - Cryptophyta + goethite + illite + chlorophyll-a Cyanophyta - phycocyanin + phycocyanin 11/2/2018 - α-PEC + chlorophyll-b - chlorophyll-b Cyanophyta - dinoxanthin - kaolinite + illite + chlorophyll-a Cyanophyta - phycocyanin + phycocyanin 11/21/2018 - α-PEC + fucoxanthin - chlorophyll-b Cyanophyta + phycocyanin + phycocyanin - kaolinite

100

Table 3.2: Results from the forward stepwise multiple linear regression of the average Sentinel-3A OLCI spectral signature patterns identified with the constituent library.

Pattern A - Average Sentinel-3A OLCI spectral signature R R-Squared Adj. R-Squared S F p-value 0.9745 0.9497 0.9371 0.2508 75.4887 6.4124E-6 Constituent Coeff. Standard Error Beta t p-value > t VIF Illite 0.9086 0.0793 0.9086 11.4516 3.0596E-6 1.0009 a-PEC -0.3258 0.0793 -0.3258 -4.1064 0.0034 1.0009

Pattern B - Average Sentinel-3A OLCI spectral signature R R-Squared Adj. R-Squared S F p-value 0.9745 0.9496 0.9280 0.2684 43.9309 6.5749E-5 Constituent Coeff. Standard Error Beta t p-value > t VIF A. lagunensis 0.8950 0.1153 0.8950 7.7638 0.0001 1.8443 Goethite 0.4520 0.0914 0.4520 4.9467 0.0017 1.1590 Cryptophyta -0.3345 0.1111 -0.3345 -3.0105 0.0196 1.7135

Pattern C - Average Sentinel-3A OLCI spectral signature R R-Squared Adj. R-Squared S F p-value 0.9335 0.8714 0.8164 0.4285 15.8176 0.0017 Constituent Coeff. Standard Error Beta t p-value > t VIF A. lagunensis 0.9338 0.1488 0.9338 6.2751 0.0004 1.2059 Illite 0.7603 0.1826 0.7603 4.1625 0.0042 1.8165 Chl-b 0.4312 0.1867 0.4312 2.3093 0.0542 1.8987

Pattern D - Average Sentinel-3A OLCI spectral signature R R-Squared Adj. R-Squared S F p-value 0.7935 0.6296 0.5885 0.6415 15.3006 0.0036 Constituent Coeff. Standard Error Beta t p-value > t VIF Allophycocyanin 0.7935 0.2029 0.7935 3.9116 0.0036 1.0000

Pattern E - Average Sentinel-3A OLCI spectral signature R R-Squared Adj. R-Squared S F p-value 0.8800 0.7744 0.7180 0.5310 13.7309 0.0026 Constituent Coeff. Standard Error Beta t p-value > t VIF phycocyanin 0.8085 0.1726 0.8085 4.6850 0.0016 1.0561 Chl-b + Cyanophyta -0.5806 0.1726 -0.5806 -3.3645 0.0099 1.0561

Pattern F - Average Sentinel-3A OLCI spectral signature R R-Squared Adj. R-Squared S F p-value 0.9242 0.8542 0.8177 0.4269 23.4318 0.0005 Constituent Coeff. Standard Error Beta t p-value > t VIF a-PEC 0.8159 0.1351 0.8159 6.0399 0.0003 1.0012 Phaeophytin-b -0.4073 0.1351 -0.4073 -3.0151 0.0167 1.0012

101

3.67

Avg. Avg.

88.60

Chl-a

(ug/L)

176.05

129.90 185.97

244.19

3A 3A OLCI -

6.72

Avg. Avg.

BGA BGA

(ug/L)

169.33

218.97

231.24

194.76 259.19

8.73

8.85

9.02

9.28

9.30

8.20 Avg. Avg.

series of Sentinel

Avg. Avg.

(mV)

ORP ORP

58.35

18.27

150.39

175.85 142.46 281.44

(m)

0.37

0.14

0.70

1.15

0.48

Avg. Avg. Depth Depth

Avg. Avg.

16.37

18.27

13.96

19.95

19.34

26.89 (PSU)

Salinity

Avg. Avg.

63.51

89.47

73.76 99.40

65.33

106.85 Per. Sat. Per.

Avg. Avg.

17.87

14.86

16.08

15.67

16.51

12.86

(RFU) FDOM

for VPC spectral pattern from time

3.81

Avg. Avg.

quality measurements at Creek station. Sykes measurements quality

14.75

14.22

18.45

15.30

13.36 (NTU)

- Turbidity

% %

9.8

9.7

8.1

9.1 Variance

D Pattern

% %

16.7

18.6 21.2

18.2

Variance Pattern C Pattern

% %

30.4

26.3

37.6 Variance

B Pattern

% %

41.3

38.6

42.6 28.5

45.9

Changes Changes in percent variance accounted

Variance

Pattern A Pattern

: : 3

Date

6/9/2018

3/9/2018

8/1/2017

8/17/2018

6/28/2018

3/23/2018

Acquisition

Table with along Kilroy water corresponding imagery

102

BIBLIOGRAPHY

Ali, K. Adem, Ortiz, J., Bonini, N., Shuman, M., & Sydow, C. (2016). Application of Aqua

MODIS sensor data for estimating chlorophyll a in the turbid Case 2 waters of Lake

Erie using bio-optical models. GIScience and Remote Sensing, 53(4), 483–505.

https://doi.org/10.1080/15481603.2016.1177248

Ali, Khalid A., & Ortiz, J. D. (2016). Multivariate approach for chlorophyll-a and

suspended matter retrievals in Case II type waters using hyperspectral data.

Hydrological Sciences Journal, 61(1), 200–213.

https://doi.org/10.1080/02626667.2014.964242

Ali, Khalid A., Witter, D. L., & Ortiz, J. D. (2014a). Multivariate approach to estimate

colour producing agents in Case 2 waters using first-derivative spectrophotometer

data. Geocarto International, 29(2), 102–127.

https://doi.org/10.1080/10106049.2012.743601

Ali, Khalid A., Witter, D. L., & Ortiz, J. D. (2014b). Multivariate approach to estimate

colour producing agents in Case 2 waters using first-derivative spectrophotometer

data. Geocarto International, 29(2), 102–127.

https://doi.org/10.1080/10106049.2012.743601

Avouris, D. M., & Ortiz, J. D. (2019). Validation of 2015 Lake Erie MODIS image spectral

decomposition using visible derivative spectroscopy and field campaign data. Journal

of Great Lakes Research, 45(3), 466–479. https://doi.org/10.1016/j.jglr.2019.02.005

Backer, L. C., Fleming, L. E., Rowan, A. D., & Baden, D. G. (2003). Epidemiology, public

103

health and human diseases associated with harmful marine algae. (G. M. Hallegraeff,

D. M. Anderson, & Cembella, Eds.), Manual on Harmful Marine Mircoalgae

(Second). UNESCO. Retrieved from

https://www.researchgate.net/publication/290125074

Barile, P. J. (2018). Widespread sewage pollution of the Indian River Lagoon system,

Florida (USA) resolved by spatial analyses of macroalgal biogeochemistry. Marine

Pollution Bulletin, 128(October 2017), 557–574.

https://doi.org/10.1016/j.marpolbul.2018.01.046

Capper, A., Flewelling, L. J., & Arthur, K. (2013). Dietary exposure to harmful algal bloom

(HAB) toxins in the endangered manatee (Trichechus manatus latirostris) and green

sea turtle (Chelonia mydas) in Florida, USA. Harmful Algae, 28, 1–9.

https://doi.org/10.1016/j.hal.2013.04.009

Carlson, T. C., & Ripley, D. a. (1997). On the relationship between NDVI, fractional

vegetation cover, and leaf area index. Remote Sensing of Environment, 62, 241–252.

https://doi.org/10.1016/S0034-4257(97)00104-1

Carvalho, G. A., Minnett, P. J., Banzon, V. F., Baringer, W., & Heil, C. A. (2011). Long-

term evaluation of three satellite ocean color algorithms for identifying harmful algal

blooms (Karenia brevis) along the west coast of Florida: A matchup assessment.

Remote Sensing of Environment, 115(1), 1–18.

https://doi.org/10.1016/j.rse.2010.07.007

Carvalho, G. A., Minnett, P. J., Fleming, L. E., Banzon, V. F., & Baringer, W. (2010).

104

Satellite remote sensing of harmful algal blooms: A new multi-algorithm method for

detecting the Florida Red Tide (Karenia brevis). Harmful Algae, 9(5), 440–448.

https://doi.org/10.1016/j.hal.2010.02.002

Chamberlain, R., & Hayward, D. (1996). Evaluation of water quality and monitoring in the

St. Lucie Estuary, Florida. Journal of the American Water Resources Association,

32(4), 681–696. https://doi.org/10.1111/j.1752-1688.1996.tb03466.x

DeYoe, H. R., Stockwell, D. A., Bidigare, R. R., Latasa, M., Johnson, P. W., Hargraves,

P. E., & Suttle, C. A. (1997). Description and characterization of the algal species

Aureoumbra lagunensis gen. et sp. nov. and referral of Aureoumbra and Aureococcus

to the Pelagophyceae. Journal of Phycology, 33(6), 1042–1048.

https://doi.org/10.1111/j.0022-3646.1997.01042.x

Dunne, E. J., Clark, M. W., Corstanje, R., & Reddy, K. R. (2011). Legacy phosphorus in

subtropical wetland soils: Influence of dairy, improved and unimproved pasture land

use. Ecological Engineering, 37(10), 1481–1491.

https://doi.org/10.1016/J.ECOLENG.2011.04.003

Fox, A. L., & Trefry, J. H. (2018). Environmental Dredging to Remove Fine-Grained,

Organic-Rich Sediments and Reduce Inputs of Nitrogen and Phosphorus to a

Subtropical Estuary. Marine Technology Society Journal, 52(4), 42–57.

https://doi.org/10.4031/MTSJ.52.4.3

Gamon, J. A. (2015). Reviews and Syntheses: Optical sampling of the flux tower footprint.

Biogeosciences, 12(14), 4509–4523. https://doi.org/10.5194/bg-12-4509-2015

105

Gamon, J. A., Peñuelas, J., & Field, C. B. (1992). A narrow-waveband spectral index that

tracks diurnal changes in photosynthetic efficiency. Remote Sensing of Environment,

41(1), 35–44. https://doi.org/10.1016/0034-4257(92)90059-S

Gao, B. C. (1996). NDWI - A normalized difference water index for remote sensing of

vegetation liquid water from space. Remote Sensing of Environment, 58(3), 257–266.

https://doi.org/10.1016/S0034-4257(96)00067-3

Garbulsky, M. F., Peñuelas, J., Gamon, J., Inoue, Y., & Filella, I. (2011). The

photochemical reflectance index (PRI) and the remote sensing of leaf, canopy and

ecosystem radiation use efficiencies. A review and meta-analysis. Remote Sensing of

Environment, 115(2), 281–297. https://doi.org/10.1016/j.rse.2010.08.023

Gitelson, A. A., Kaufman, Y. J., & Merzlyak, M. N. (1996). Use of a green channel in

remote sensing of global vegetation from EOS-MODIS. Remote Sensing of

Environment, 58(3), 289–298. https://doi.org/10.1016/S0034-4257(96)00072-7

Gitelson, A. A., Kaufman, Y. J., Stark, R., & Rundquist, D. (2002). Novel algorithms for

remote estimation of vegetation fraction. Remote Sensing of Environment, 80(1), 76–

87. https://doi.org/10.1016/S0034-4257(01)00289-9

Gobler, C. J., Koch, F., Kang, Y., Berry, D. L., Tang, Y. Z., Lasi, M., … Miller, J. D.

(2013). Expansion of harmful brown tides caused by the pelagophyte, Aureoumbra

lagunensis DeYoe et Stockwell, to the US east coast. Harmful Algae, 27, 29–41.

https://doi.org/10.1016/j.hal.2013.04.004

Gobler, C. J., & Sunda, W. G. (2012a). Ecosystem disruptive algal blooms of the brown

106

tide species, Aureococcus anophagefferens and Aureoumbra lagunensis. Harmful

Algae, 14, 36–45. https://doi.org/10.1016/j.hal.2011.10.013

Gobler, C. J., & Sunda, W. G. (2012b). Ecosystem disruptive algal blooms of the brown

tide species, Aureococcus anophagefferens and Aureoumbra lagunensis. Harmful

Algae, 14, 36–45. https://doi.org/10.1016/j.hal.2011.10.013

Gobler, C., & Sañudo-Wilhelmy, S. (2001). Temporal variability of groundwater seepage

and brown tide blooms in a Long Island embayment. Marine Ecology Progress Series,

217, 299–309. https://doi.org/10.3354/meps217299

Hall, N. S., Litaker, R. W., Kenworthy, W. J., Vandersea, M. W., Sunda, W. G., Reid, J.

P., … Butler, S. (2018). Consortial brown tide − picocyanobacteria blooms in

Guantánamo Bay, Cuba. Harmful Algae, 73, 30–43.

https://doi.org/10.1016/j.hal.2018.01.003

Harvey, R. G., Loftus, W. F., Rehage, J. S., & Mazzotti, F. J. (2010). Effects of Canals and

Levees on Everglades Ecosystems 1. Environmental Research, 1–21.

Kaiser, H. F. (1958). The varimax criterion for analytic rotation in factor analysis.

Psychometrika, 23(3), 187–200. https://doi.org/10.1007/BF02289233

Kamerosky, A., Cho, H. J., & Morris, L. (2015). Monitoring of the 2011 super algal bloom

in Indian River Lagoon, FL, USA, Using MERIS. Remote Sensing, 7(2), 1441–1460.

https://doi.org/10.3390/rs70201441

Kang, Y., & Gobler, C. J. (2018). The brown tide algae, Aureococcus anophagefferens and

Aureoumbra lagunensis (Pelagophyceae), allelopathically inhibit the growth of

107

competing microalgae during harmful algal blooms. Limnology and Oceanography,

63(2), 985–1003. https://doi.org/10.1002/lno.10714

Kang, Y., Koch, F., & Gobler, C. J. (2015). The interactive roles of nutrient loading and

zooplankton grazing in facilitating the expansion of harmful algal blooms caused by

the pelagophyte, Aureoumbra lagunensis, to the Indian River Lagoon, FL, USA.

Harmful Algae, 49, 162–173. https://doi.org/10.1016/j.hal.2015.09.005

Katz, N., Ali, K. A., Ortiz, J. D., Brandt, M., Smith, T. B., & Shahin, M. (2018). Estimating

Chlorophyll-a Concentrations in Coastal Waters of the U.S. Virgin Islands using

NASA’s New Generation Landsat 8 OLI Satellite Sensor. In American Geophysical

Union, Fall Meeting 2018, abstract #H33I-2214. Retrieved from

http://adsabs.harvard.edu/abs/2018AGUFM.H33I2214K

Kirkpatrick, B., Fleming, L. E., Squicciarini, D., Backer, L. C., Clark, R., Abraham, W.,

… Baden, D. G. (2004). Literature review of Florida red tide: implications for human

health effects. Harmful Algae, 3(2), 99–115.

https://doi.org/10.1016/J.HAL.2003.08.005

Koch, F., Kang, Y., Villareal, T. A., Anderson, D. M., & Gobler, C. J. (2014). A Novel

Immunofluorescence Flow Cytometry Technique Detects the Expansion of Brown

Tides Caused by Aureoumbra lagunensis to the Caribbean Sea. Applied and

Environmental Microbiology, 80(16), 4947–4957.

https://doi.org/10.1128/AEM.00888-14

Kokaly, R. F., Clark, R. N., Swayze, G. A., Livo, K. E., Hoefen, T. M., Pearson, N. C., …

108

Klein, A. J. (2017). USGS Spectral Library Version 7. United States Geological

Survey Data Series 1035. https://doi.org/10.3133/ds1035

Kramer, B. J., Davis, T. W., Meyer, K. A., Rosen, B. H., Goleski, J. A., Dick, G. J., …

Gobler, C. J. (2018). Nitrogen limitation, toxin synthesis potential, and toxicity of

cyanobacterial populations in Lake Okeechobee and the St. Lucie River Estuary,

Florida, during the 2016 state of emergency event. PLOS ONE, 13(5), e0196278.

https://doi.org/10.1371/journal.pone.0196278

Lane, E. (1987). Guide to Rocks and Minerals of Florida (8th ed.). Florida Dept. of Natural

Resources, Bureau of Geology, 1971-1986.

Lapointe, B. E., Herren, L. W., & Bedford, B. J. (2012). Effects of Hurricanes, Land Use,

and Water Management on Nutrient and Microbial Pollution: St. Lucie Estuary,

Southeast Florida. Journal of Coastal Research, 285(6), 1345–1361.

https://doi.org/10.2112/JCOASTRES-D-12-00070.1

Lapointe, B. E., Herren, L. W., Debortoli, D. D., & Vogel, M. A. (2015). Evidence of

sewage-driven eutrophication and harmful algal blooms in Florida’s Indian River

Lagoon. Harmful Algae, 43, 82–102. https://doi.org/10.1016/j.hal.2015.01.004

Lapointe, B. E., Herren, L. W., & Paule, A. L. (2017). Septic systems contribute to nutrient

pollution and harmful algal blooms in the St. Lucie Estuary, Southeast Florida, USA.

Harmful Algae, 70, 1–22. https://doi.org/10.1016/j.hal.2017.09.005

Lekki, J., Anderson, R., Avouris, D., Becker, Ri., Churnside, J., Cline, M., … Vander

Woude, A. (2017). Airborne Hyperspectral Sensing of Monitoring Harmful Algal

109

Blooms in the Great Lakes Region: System Calibration and Validation. Retrieved

from https://ntrs.nasa.gov/search.jsp?R=20170002298

Li, L., He, Z., Li, Z., Li, S., Wan, Y., & Stoffella, P. J. (2017). Spatiotemporal change of

phosphorous speciation and concentration in stormwater in the St. Lucie Estuary

watershed, South Florida. Chemosphere, 172, 488–495.

https://doi.org/10.1016/j.chemosphere.2017.01.020

Liu, H., & Buskey, E. J. (2000). Hypersalinity enhances the production of extracellular

polymeric substance (EPS) in the Texas brown tide alga, Aureoumbra lagunensis

(Pelagophyceae). Journal of Phycology, 36(1), 71–77. https://doi.org/10.1046/j.1529-

8817.2000.99076.x

Liu, H., Laws, E. A., Villareal, T. A., & Buskey, E. J. (2001). Nutrient‐limited growth of

Aureoumbra lagunensis (Pelagophyceae), with implications for its capability to

outgrow other phytoplankton species in phosphate‐limited environments. Journal of

Phycology, 37(4), 500–508. Retrieved from https://onlinelibrary-wiley-

com.proxy.library.kent.edu/doi/pdf/10.1046/j.1529-8817.2001.037004500.x

Mclean, A. R., Ogden, J. C., & Williams, E. E. (2002). Chapter 7: Comprehensive

Everglades Restoration Plan. Retrieved from

http://xportal.sfwmd.gov/portal/page/portal/pg_grp_sfwmd_sfer/portlet_prevreport/2

002_ecr/Chapters/Ch7.pdf

McNeal, B. L., Stanley, C. D., Graham, W. D., Gilreath, P. R., Downey, D., & Creighton,

J. F. (1995). Nutrient-Loss Trends for Vegetable and Citrus Fields in West-Central

110

Florida: I. Nitrate. Journal of Environment Quality, 24(1), 95.

https://doi.org/10.2134/jeq1995.00472425002400010013x

Metcalf, J. S., Banack, S. A., Powell, J. T., Tymm, F. J. M., Murch, S. J., Brand, L. E., &

Cox, P. A. (2018). Public health responses to toxic cyanobacterial blooms:

perspectives from the 2016 Florida event. Water Policy, 20(5), 919–932.

https://doi.org/10.2166/wp.2018.012

Ming-kui, Z., Li-ping, W., & Zhen-li, H. (2007). Spatial and temporal variation of nitrogen

exported by runoff from sandy agricultural soils. Journal of Environmental Sciences,

19(7), 1086–1192.

O’Reilly, J. E., Maritorena, S., Siegel, D. A., O’brien, M. C., Toole, D., Mitchell, B. G.,

… Culver, M. (2000). Ocean Color Chlorophyll a Algorithms for SeaWiFS, OC2 and

OC4: Version 4. Retrieved from

https://s3.amazonaws.com/academia.edu.documents/36421462/2000_OReilly_et_al

_OC4v4_11-Chapt2.pdf?response-content-disposition=inline%3B

filename%3DOcean_color_chlorophyll_a_algorithms_for.pdf&X-Amz-

Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIAIWOWYYGZ

Oehrle, S., Rodriguez-Matos, M., Cartamil, M., Zavala, C., & Rein, K. S. (2017). Toxin

composition of the 2016 Microcystis aeruginosa bloom in the St. Lucie Estuary,

Florida. https://doi.org/10.1016/j.toxicon.2017.09.005

Ortiz, J. D. (2011). Application of visible/near infrared derivative spectroscopy to Arctic

paleoceanography. In IOP Conference Series: Earth and Environmental Science (Vol.

111

14, p. 012011). https://doi.org/10.1088/1755-1315/14/1/012011

Ortiz, J. D., Avouris, D. M., Schiller, S. J., Luvall, J. C., Lekki, J. D., Tokars, R. P., …

Becker, R. (2019). Evaluating visible derivative spectroscopy by varimax-rotated,

principal component analysis of aerial hyperspectral images from the western basin

of Lake Erie. Journal of Great Lakes Research, 45(3), 522–535.

https://doi.org/10.1016/j.jglr.2019.03.005

Ortiz, J. D., Avouris, D., Schiller, S., Luvall, J. C., Lekki, J. D., Tokars, R. P., … Becker,

R. (2017a). Intercomparison of Approaches to the Empirical Line Method for

Vicarious Hyperspectral Reflectance Calibration. Frontiers in Marine Science, 4.

https://doi.org/10.3389/fmars.2017.00296

Ortiz, J. D., Avouris, D., Schiller, S., Luvall, J. C., Lekki, J. D., Tokars, R. P., … Becker,

R. (2017b). Intercomparison of Approaches to the Empirical Line Method for

Vicarious Hyperspectral Reflectance Calibration. Frontiers in Marine Science,

4(September). https://doi.org/10.3389/fmars.2017.00296

Ortiz, J. D., Witter, D. L., Ali, K. A., Fela, N., Duff, M., & Mills, L. (2013). Evaluating

multiple colour-producing agents in Case II waters from Lake Erie. International

Journal of Remote Sensing, 34(24), 8854–8880.

https://doi.org/10.1080/01431161.2013.853892

Phlips, E. J., Badylak, S., Christman, M., Wolny, J., Brame, J., Garland, J., … Steidinger,

K. (2011). Scales of temporal and spatial variability in the distribution of harmful

algae species in the Indian River Lagoon, Florida, USA. Harmful Algae, 10(3), 277–

112

290. https://doi.org/10.1016/j.hal.2010.11.001

Phlips, E. J., Badylak, S., Lasi, M. A., Chamberlain, R., Green, W. C., Hall, L. M., …

Steward, J. S. (2015). From Red Tides to Green and Brown Tides: Bloom Dynamics

in a Restricted Subtropical Lagoon Under Shifting Climatic Conditions. Estuaries and

Coasts, 38(3), 886–904. https://doi.org/10.1007/s12237-014-9874-6

Reddy, K. R., Newman, S., Osborne, T. Z., White, J. R., & Fitz, H. C. (2011). Phosphorous

Cycling in the Greater Everglades Ecosystem: Legacy Phosphorous Implications for

Management and Restoration. Critical Reviews in Environmental Science and

Technology, 41(sup1), 149–186. https://doi.org/10.1080/10643389.2010.530932

Rhudy, K. B., Sharma, V. K., Lehman, R. L., & McKee, D. A. (1999). Seasonal variability

of the Texas “brown tide” (Aureoumbra lagunensis) in relation to environmental

parameters. Estuarine, Coastal and Shelf Science, 48(5), 565–574.

https://doi.org/10.1006/ecss.1998.0460

Sime, P. (2005). St. Lucie Estuary and Indian River Lagoon conceptual ecological model.

Wetlands, 25(4), 898–907. https://doi.org/10.1672/0277-

5212(2005)025[0898:sleair]2.0.co;2

Smith, N. P. (1993). Tidal and Nontidal Flushing of Florida’s Indian River Lagoon.

Estuaries, 16(4), 739. https://doi.org/10.2307/1352432

Steidinger, K. A., Landsberg, J. H., Tomas, C. R., & Burns, J. W. (1999). Harmful Algal

Blooms in Florida.

Thosteson, E. D., Widder, E. A., Cimaglia, C. A., Taylor, J. W., Burns, B. C., & Paglen,

113

K. J. (2009). New technology for Ecosystem-Based Management: Marine monitoring

with the ORCA Kilroy Network. In OCEANS 2009-EUROPE (pp. 1–7). IEEE.

https://doi.org/10.1109/OCEANSE.2009.5278229

Tomlinson, M. C., Wynne, T. T., & Stumpf, R. P. (2009). An evaluation of remote sensing

techniques for enhanced detection of the toxic dinoflagellate, Karenia brevis. Remote

Sensing of Environment, 113(3), 598–609. https://doi.org/10.1016/j.rse.2008.11.003

Trocine, R. P., & Trefry, J. H. (1996). Metal concentrations in sediment, water and clams

from the Indian River Lagoon, Florida. Marine Pollution Bulletin, 32(10), 754–759.

https://doi.org/10.1016/0025-326x(96)00071-9

Witter, D. L., Ortiz, J. D., Palm, S., Heath, R. T., & Budd, J. W. (2009). Assessing the

application of SeaWiFS ocean color algorithms to Lake Erie. Journal of Great Lakes

Research, 35(3), 361–370. https://doi.org/10.1016/j.jglr.2009.03.002

Yang, Y., He, Z., Wang, Y., Fan, J., Liang, Z., & Stoffella, P. J. (2013). Dissolved organic

matter in relation to nutrients (N and P) and heavy metals in surface runoff water as

affected by temporal variation and land uses - A case study from Indian River Area,

south Florida, USA. Agricultural Water Management, 118, 38–49.

https://doi.org/10.1016/j.agwat.2012.12.001

Zhang, M., Wang, L., & He, Z. (2007). Spatial and temporal variation of nitrogen exported

by runoff from sandy agricultural soils. Journal of Environmental Sciences, 19(9),

1086–1092. https://doi.org/10.1016/S1001-0742(07)60177-6

114