Remote Sensing of Environment 157 (2015) 24–34
Contents lists available at ScienceDirect
Remote Sensing of Environment
journal homepage: www.elsevier.com/locate/rse
Evaluation of cyanobacteria cell count detection derived from MERIS
imagery across the eastern USA
Ross S. Lunetta a, Blake A. Schaeffer b,⁎, Richard P. Stumpf c, Darryl Keith d, Scott A. Jacobs e, Mark S. Murphy f
a
US EPA National Exposure Research Laboratory, 109 T. W. Alexander Drive, Mail Code: E243-05, Research Triangle Park, NC 27709, United States
US EPA National Health and Environmental Effects Research Laboratory, Gulf Ecology Division, 1 Sabine Island Drive, Gulf Breeze, FL 32561, United States
NOAA National Oceans Service, Center for Coastal Monitoring and Assessment, 1305 East West Highway, Silver Springs, MD 20910, United States
d
US EPA National Health and Environmental Effects Research Laboratory, Atlantic Ecology Division, 27 Tarzwell Drive, Narragansett, RI 02882, United States
e
US EPA National Risk Management Research Laboratory, Land Remediation and Pollution Control Division, 26 Martin Luther King Drive, Cincinnati, OH 45268, United States
f
Innovatel, Incorporated. 2800 Meridian Parkway, Durham, NC 27713, United States
b
c
a r t i c l e
i n f o
Article history:
Received 27 August 2013
Received in revised form 30 May 2014
Accepted 9 June 2014
Available online 7 July 2014
Keywords:
Cyanobacteria monitoring
Harmful algal blooms
Algorithm validation
MERIS
Sentinel-3
Water quality
a b s t r a c t
Inland waters across the United States (US) are at potential risk for increased outbreaks of toxic cyanobacteria
blooms events resulting from elevated water temperatures and extreme hydrologic events attributable to climate
change and increased nutrient loadings associated with intensive agricultural practices. Current monitoring efforts are limited in scope due to resource limitations, analytical complexity, and data integration efforts. The
goals of this study were to validate an algorithm for satellite imagery that could potentially be used to monitor
surface cyanobacteria events in near real-time to provide a compressive monitoring capability for freshwater
lakes (N 100 ha). The algorithm incorporated narrow spectral bands specific to the European Space Agency's
(ESA's) MEdium Resolution Imaging Spectrometer (MERIS) instrument that were optimally oriented at phytoplankton pigment absorption features including phycocyanin at 620 nm. A validation of derived cyanobacteria
cell counts was performed using available in situ data assembled from existing monitoring programs across
eight states in the eastern US over a 39-month period (2009–2012). Results indicated that MERIS provided robust
estimates for low (10,000–109,000 cells/mL) and very high (N 1,000,000 cells/mL) cell enumeration ranges (approximately 90% and 83%, respectively). However, the results for two intermediate ranges (110,000–299,000 and
300,000–1,000,000 cells/mL) were substandard, at approximately 28% and 40%, respectively. The confusion associated with intermediate cyanobacteria cell count ranges was largely attributed to the lack of available taxonomic
data and distinction of natural counting units for the in situ measurements that would have facilitated conversions between cell counts and cell volumes. The results of this study document the potential for using MERISderived cyanobacteria cell count estimates to monitor freshwater lakes (N100 ha) across the eastern US.
Published by Elsevier Inc.
1. Introduction
Cyanobacteria and their cyanotoxins are unregulated contaminants. However, cyanotoxins are included in the US Environmental
Protection Agency (USEPA) Safe Drinking Water Act “Contaminant
Candidate List” (USEPA, 2013a). Cyanobacteria blooms occur worldwide and are associated with human respiratory irritation, taste and
odor of potable water, and human illness as a result of ingestion or
skin exposure during recreational activities. The term bloom is defined here as anytime occurrence may result in negative environmental or health consequences (Smayda, 1997). Pets, domestic
animals, and wildlife are also affected by exposure to cyanotoxins,
with deaths reported annually (Backer, 2002). Cyanotoxins can be
⁎ Corresponding author. Tel.: +1 919 541 5571.
E-mail address: [email protected] (B.A. Schaeffer).
http://dx.doi.org/10.1016/j.rse.2014.06.008
0034-4257/Published by Elsevier Inc.
found in water bodies used for drinking, aquaculture, crop irrigation,
and recreation (Stewart, Webb, Schluter, & Shaw, 2006). It has been
hypothesized that cyanotoxins could even be transferred to crops
designated for human consumption via spray irrigation (Hunter,
Tyler, Carvalho, Codd, & Maberly, 2010). Cyanobacteria blooms can
be aesthetically unappealing, which is enhanced by their tendency
to concentrate along shorelines where they are encountered frequently by the public (Chorus, Salas, & Bartram, 2000). Increasing
frequency, duration and magnitude of blooms within some systems
has prompted management actions. Postulated causes of these
blooms include excessive nutrient loads, introduction of invasive
species (Budd, Drummer, Nalepa, & Fahnenstiel, 2001), and increasing temperatures from climate change and variability, where warmer surface waters favor cyanobacteria growth (Paerl & Huisman,
2008). Alterations in land-cover (e.g., urbanization) and changes in
land-use practices, such as intensive agricultural practices and biofuel mandates (e.g., 2007 Energy Independence and Security Act) can
R.S. Lunetta et al. / Remote Sensing of Environment 157 (2015) 24–34
result in watersheds experiencing additional sediment loading and nutrient delivery, which can influence cyanobacteria growth (Lunetta,
Shao, Ediriwickrema, & Lyon, 2010; Michalak et al., 2013; USEPA, 2011).
Harmful algal blooms exact a cost in freshwater degradation of approximately $2.2 billion annually in the United States (US) (Dodds
et al., 2009). Costs include loss of recreational water usage, declines in
waterfront real estate value, spending on recovery of biodiversity, and
drinking water treatment. The greatest economic losses were in the
decline of real estate value and income resulting from recreational
use. Accordingly, there is a critical need to determine the actual underlying causes of these blooms. Despite ecosystem, economic, and public
health concerns, cyanotoxins are typically assessed infrequently due
to the cost, required expertise, and time requirements for complex analysis. Likewise, measuring associated water quality parameters is time
and cost intensive, often requiring a field team to spend a majority of
the day on a boat collecting samples. Data from discrete stations typically provide incomplete spatial and temporal coverage of a water body.
While field sampling is certainly necessary and will continue to be so,
managers need other options to improve monitoring, reduce costs,
and minimize potential exposures associated with sample collection.
The incidence of cyanobacteria blooms has been increasing in the US
(Hudnell, 2010) and extends across varied habitats from smaller eutrophic prairie ponds to larger more oligotrophic regions of the Laurentian
Great Lakes (Fristachi & Sinclair, 2008). Microcystin concentrations are
moving northward along a gradient of increasing lake trophic status
across the Midwest (Graham, Jones, Jones, Downing, & Clevenger,
2004). Stumpf, Wynne, Baker, and Fahenstiel (2012) determined that
in 2011 Lake Erie experienced a cyanobacteria bloom event three
times greater than any previously observed event as a result of unusually high runoff. Michalak et al. (2013) further link the outbreak to increased phosphorous loading associated with long-term agricultural
land-use practices (expanded corn production) coupled with meteorological conditions in spring of 2011 that were consistent with climate
trends predicted for future conditions. Climate change is a potential
catalyst for further bloom development as cyanobacteria often out
compete other phytoplankton taxa at higher temperatures (Paerl &
Huisman, 2008). However, there are no comprehensive data sources
documenting the occurrences of cyanobacteria outbreaks across the
US (Hudnell, 2010).
1.1. Cyanobacteria monitoring
Cyanobacteria blooms have been documented across the US, and numerous states have issued health advisories or closed recreational areas
due to potential risks from exposure. Cyanobacteria and their toxins are
addressed differently by each state. A few states have toxin monitoring
programs in place, while others only conduct event-based responses,
and some provide public education focused on human and animal protection from toxin exposure (Graham, Loftin, & Kamman, 2009). Typically, monitoring programs include visual assessments and water
sampling. Many use the World Health Organization (WHO) guidelines,
with a three-level approach using chlorophyll-a (Chl-a), cell counts, and
microcystin-LR measurements during recreational activities (Chorus &
Bartram, 1999). It is important to emphasize that the cell count range
categories used by the WHO are based on “relative probabilities of
acute health effects” and do not directly correspond to the ranges
used in this study. In addition, taxonomy is typically not a standard
part of these measurements. This allows for a series of alert levels
where various data sources are the foundation of advisory postings or
water body closings. Other states, such as Oklahoma and Massachusetts,
have developed state-specific guidelines to establish protective levels.
Oklahoma has legislation limiting exposure to freshwater algae, where
warnings are issued to lake users if algae cell counts exceed 100,000
cells/mL and microcystin concentrations exceed 20 μg/L. Massachusetts
has established guidelines for issuing an advisory against contact with
25
water when counts exceed 70,000 cells/mL or microcystin concentrations exceed 14 μg/L.
States have experienced many challenges in the development of
monitoring programs for cyanobacteria toxins. Field based monitoring programs may be insufficient to provide timely warnings of
cyanobacteria bloom development across large geographic areas because estimating cyanobacteria concentrations is a labor intensive
and time consuming endeavor, involving water sample collections,
laboratory analysis, and the visual identification and enumeration
of phytoplankton species. Further, the recognition of cyanobacteria
blooms in the field are typically limited to visual assessments of
water discoloration or the identification of floating scums and
mats, which are patchy and transient in space and time. Finally,
there are difficulties in developing appropriate sampling designs
that address large areas using fewer resources or affordable detection methods; and development of guidance based on taxonomic identification and/or cell counts can be overly conservative in protecting
public health compared to states that identify and measure toxins. In
light of these challenges, new rapid tools are needed to assist states in
development of efficient and effective cyanobacteria monitoring programs. Satellite remote sensing provides an opportunity for such a
tool. Although this technology is not capable of detecting cyanobacteria
toxins based on optical properties of the water column, it can be used to
derive cyanobacteria concentration information.
1.2. Remote sensing approaches
Studies have shown that satellite data can detect and quantify
cyanobacteria blooms in lakes and estuaries (Kahru, Savchuk, &
Elmgren, 2007; Simis, Peters, & Gons, 2005; Alikas, Kangro, & Reinart,
2010; Wynne, Stumpf, Tomlinson, & Dybleb, 2010; Gómez, Alonso, &
García, 2011; Matthews, Bernard, & Robertson, 2012; Duan, Ma, & Hu,
2012; see also review by Kutser, 2009). These include a variety of
methods and approaches including identifying blooms and scums
(Kahru et al., 2007), estimates of phycocyanin (Ruiz-Verdú, Simis,
Hoyos, Gons, & Peña-Martínez, 2008) or the presence of phycocyanin
(Matthews et al., 2012; Simis et al., 2005), biomass as chlorophyll-a
(Matthews et al., 2012), and cell count concentrations (Hunter et al.,
2010; Wynne et al., 2010). Various satellites have been examined such
as the OCM (Dash et al., 2011), Landsat (Vincent, Qin, Mckay, & Miner,
2004), AVHRR (Kahru et al., 2007), MODIS (Becker, Sultan, Boyer,
Twiss, & Konopko, 2009; Wynne et al., 2013), and MERIS (Binding,
Greenberg, Jerome, Bukata, & Letourneau, 2011; Matthews, Bernard, &
Winter, 2010; Ruiz-Verdú et al., 2008; Wynne et al., 2008). In the US,
the longest running monitoring program for cyanobacteria is for Lake
Erie. The standard algorithm for that program is the “cyanobacterial
index” (CI) (Wynne et al., 2008), which uses a second derivative centered on the MERIS fluorescence band at 681 nm or the MODIS
678 nm band (Wynne et al., 2010; Wynne et al., 2013; Wynne et al.,
2008). The second derivative functions with minimal or poor atmospheric correction (Philpot, 1991), eliminating a significant challenge
to near daily application in highly turbid water. In fact, several applications of second derivative functions do not use any atmospheric correction (Gower, King, & Goncalves, 2008; Hu, 2009). The resulting “Lake
Erie HAB Bulletin” product is used by hundreds of people including
many state and city managers (Wynne et al., 2013). With the demonstrated application of this analysis, and need for monitoring in many
lakes in the US, an evaluation of its ability to detect cyanobacteria
blooms in other lakes is warranted. It is acknowledged there are issues
in the comparison of two measured quantities with no direct overlap;
optical signals of light absorption and scattering by pigments in cells,
and cell counts without biovolume or biomass information. The ratio
of pigmentation and biomass is not constant, and neither is the ratio
of cell number to biomass. The study objective was to evaluate whether
a cyanobacteria algorithm based on Wynne et al. (2008) can be applied
26
R.S. Lunetta et al. / Remote Sensing of Environment 157 (2015) 24–34
to MERIS data to identify surface cyanobacteria bloom events in freshwater lake systems for applications throughout the eastern US.
Potential benefits associated with remote sensing tools include enabling policymakers and environmental managers to assess the sustainability of watershed ecosystems, and the services they provide, under
current and future land-use practices. The most direct way to ensure
that management practices are achieving sustainability is to monitor
the environment on a synoptic scale. Sustainable practices have a significant impact when widely implemented and relevant at a national scale.
Satellite technology could allow for the development of cyanobacteria
early warning indicators. Rapid detection of potential blooms is
essential for protecting the general population from exposure, including
magnitude and duration of contact to the stressor. The National Research Council report on “Exposure Science in the 21st Century” suggested that assessing and mitigating toxic exposures effectively
requires rapid measurement of the stressor on diverse geographic, temporal, and biological scales and an enhanced infrastructure for rapid deployment of resources to address imminent threats (NRC, 2012).
Predictive tools allow forecasting, prevention, and mitigation of the potential effects. The NRC report specifically calls for innovative and expedient assessment approaches that strategically use diverse information,
such as satellite remote sensing, for the identification and quantification
Study Areas and Measurement Locations
90° W
80° W
75° W
70° W
Study Areas
70° W
45° N
New England
40° N
45° N
40° N
Atlantic
Ocean
30° N
40° N
30° N
Gulf of Mexico
90° W
150
80° W
75° W
40°38'30"N
84°26'0"W
70° W
Km
80° W
Florida
Ohio
30° N
40°38'30"N
30° N
5
84°26'0"W
100
Km
80° W
Km
Legend
Field Measurement Locations
Projection: NAD83 USA Contiguous Albers Equal Area
MERIS Imagery Extent
Service Layer Credits: Copyright: ©2012 Esri, DeLorme,
NAVTEQ
Fig. 1. Study area location map indicating the field sampling locations and MERIS scene boundaries.
R.S. Lunetta et al. / Remote Sensing of Environment 157 (2015) 24–34
of relevant exposures that may pose a threat to ecosystems or human
health.
2. Methods
2.1. Study site description
In situ measurement data were assembled for lakes in eight states
across New England, Ohio, and Florida (Fig. 1) to support the evaluation.
MERIS high resolution (300 m) imagery acquisitions were matched
with available field measurement data to support the validation of the
cyanobacteria algorithm to assess the potential for regional scale performance across the eastern US.
The New England lakes used in this study were part of the Northern
Appalachians ecoregion. There were 5226 lakes in the Northern Appalachians included in the EPA National Lakes Assessment Program (NLAP)
inventory, where 54% were man-made reservoirs. Based on the 1992
National Land Cover Dataset, the distribution of land cover was 69%
forested, 17% planted/cultivated, and 14% characterized as other types.
Agricultural and urban runoff from watersheds or drainage basins
were the primary source of excess nitrogen and phosphorus to New England Lakes which exacerbated algae blooms. Lake Champlain, VT was
the largest lake in the Northern Appalachians ecoregion and one of
the largest in the United States. While primarily a recreational lake, it
also served as a source of drinking water and a site for the disposal of
municipal wastes (USEPA, 2009). In 2009, monitoring results indicated
that toxic cyanobacteria blooms continued to be prevalent in two primary areas in the lake (Missisquoi Bay and St. Albans Bay) and at several
locations in the northern region of the lake. Transient algal accumulations were found at scattered sites throughout the lake. During 2008–
2009, microcystin concentrations ranged from 0.01 to N 94 μg/L, with
highest concentrations in Missisquoi Bay (Watzin, Fuller, Bronson,
Gorney, & Schuster, 2011).
Several ecoregions intersect in Ohio including the Eastern Corn Belt
Plains, the Western Alleghany Plateau, and the Erie Drift Plain. The
trophic state of these lakes, as based on chlorophyll-a concentrations,
was 68% hypereutrophic or eutrophic. Averaging the percentage for
all basins sampled in Ohio for major land use types indicates that 22%
of land was developed, 27% was forested upland, and 42% was
planted/cultivated. Impacts to surface water in this region were primarily from agricultural and urban sources. Grand Lake St. Mary's in northwestern Ohio was a man-made reservoir that has been plagued in
recent years with persistent harmful algal blooms. The reservoir was a
source for drinking water with 66% agriculture land use designation,
which resulted in high concentrations of nutrients in the tributaries
that feed into the lake. Ohio has reported high levels of microcystin concentrations (sometimes exceeding 100 ppb) in Grand Lake St. Mary's
that have remained high into the late fall and early winter (USEPA,
2009).
The State of Florida was located in the Coastal Plains ecoregion, of
which 69% of lakes were constructed reservoirs within the ecoregion.
Based on the 1992 National Land Cover, the ecoregion land cover was
39% forested, 30% planted/cultivated, and 16% wetlands with the remaining 15% of land in other types of cover (USEPA, 2009). Anthropogenic nutrient sources included agriculture, phosphate mining and
point sources and have been suggested to increase the occurrence of
cyanobacteria from historical levels in central Florida lakes (Paerl,
2008; Riedinger-Whitmore et al., 2005).
The temporal and spatial stability of cyanobacteria blooms in a particular ecosystem depends on the extent that different environmental
factors influence bloom dynamics. In addition to nutrient supply,
these factors include wind velocity, rainfall amount and intensity, and
the solar radiation (Soranno, 1997). Because these factors vary widely,
cyanobacterial blooms display a range of temporal dynamics. Some
freshwater bodies have seasonal blooms that begin during the summer
and last into fall, some have persistent blooms that cover all four
27
seasons while others have blooms that occur just for days or weeks.
The NLAP assessment data was based on a probabilistic sampling
scheme and did not include methods for collecting data on those factors
favorable to bloom development and sustainability (such as wind velocity, solar radiation, and rainfall), therefore no assessment was conducted on the temporal and spatial stability of cyanobacteria blooms from
the these datasets.
2.2. Cyanobacteria algorithm
Data were processed to Rayleigh-corrected reflectance (R) using
NASA's SeaWiFS Data Analysis System (SeaDAS) standard l2gen software (“rho_s” dimensionless product) and mapped to standard UTM
projection with 300 m pixels using nearest neighbor resampling. The
“rho_s” product includes correction for sun angle, but does not involve
a correction for aerosols. Each MERIS image was masked for land with
the standard 250 m landmask distributed with SeaDAS and for clouds
using the cloud albedo, excluding bright reflectance water that has
strong spectral variation (high reflectance in the near-infrared and
green and low in the red and blue) to avoid masking intense blooms
that would otherwise be flagged as clouds. Clouds were flagged using
threshold algorithms, corrected for turbid water. The cyanobacteria
detection algorithm used second derivative spectral shapes (SS) on
the reflectance spectra with an equation:
n
o ðλ−λ− Þ
−
−
þ
SSðλÞ ¼ RðλÞ−Rðλ Þ þ Rðλ Þ−R λ
þ
:
ðλ −λ− Þ
ð1Þ
Philpot (1991) demonstrated that this form was insensitive to poor
atmospheric correction. Wynne et al. (2008) used the spectral shape
(SS) around the 681 nm, such that λ = 681 nm, λ+ = 709 nm, and
λ− = 665 nm. They derived a Cyanobacteria Index (CI) = −SS(681).
SS(681) is centered on the 681 nm band, which covers peaks in both
chlorophyll absorption and chlorophyll fluorescence. In clearer water
with non-cyanobacterial blooms, a fluorescence peak is observed at
681 nm. When there is negligible fluorescence, as is the case in
cyanobacteria blooms (Seppälä et al., 2007; Wynne et al., 2008), there
is no fluorescence peak. Instead the strong chlorophyll absorption
leads to a dip or “valley” at 681 nm resulting in negative SS(681)
(Binding et al., 2011). As CI = −SS(681), increased chlorophyll absorption leads to increased CI. The CI spectral shape method has proved effective in Lake Erie, where the summer blooms are primarily
composed of Microcystis spp. and provided a robust estimate of
cyanobacteria cell counts (Wynne et al., 2010). The CI can flag other
blooms including chlorophytes (Wynne et al., 2013) and some dense dinoflagellate blooms (Stumpf, unpublished data) causing false positives.
To address the false positive issue, a test had been examined to separate
cyanobacteria from other blooms using a derivative that includes the
620 nm band, which is sensitive to phycocyanin, namely the SS(665)
with λ = 665 nm, λ+ = 681 nm, and λ− = 620 nm. This condition
has been applied by Matthews et al. (2012, Eqs. 3–4) to separate
cyanobacteria from other blooms in African lakes. The resultant, multiple shape algorithm, termed the CI-multi uses the CI for the biomass estimate, but uses the spectral shape around the 665 nm band as an
exclusion criterion. Elevated phycocyanin absorption is presumed to depress the reflectance at 620 nm (e.g., Simis et al., 2005), causing SS(665)
to change from negative to positive. Accordingly, when the CI value
SS(665) b 0, cyanobacteria are presumed absent, when SS(665) N 0,
cyanobacteria are presumed present.
2.3. Algorithm validation
The cyanobacteria cell count reference data used to support MERISderived estimates came from several monitoring programs in Ohio,
Florida, and six New England states. Participating agencies included
the Ohio EPA (Ohio EPA, 2012), St. Johns River Water Management
28
R.S. Lunetta et al. / Remote Sensing of Environment 157 (2015) 24–34
District (Abbott et al., 2009), and the US EPA's National Lakes Assessment Program (NLAP, USEPA, 2007a; USEPA, 2007b). The New England
data sets also included data collected from EPA's Region I Office (USEPA,
2009), Rhode Island Department of Environmental Management (ESS
Group, Inc., 2011), the Massachusetts Departments of Environmental
Protection and Public Health (Beskenis, 2014), the New Hampshire
Department of Environmental Services and the University of New
Hampshire (NH Department of Environmental Services, 2005a,b), Vermont Department of Environmental Conservation (VT Department of
Environmental Conservation, 2014), and the Lake Chaplain Basin Program (Watzin et al., 2011). Databases provided to the USEPA included
site identification numbers, collection dates, geographic coordinates,
and cyanobacteria cell count numbers, which ranged from zero to
approximately 2.0E + 7 (Fig. 2). All 10 data sources used different
methodologies for field sampling and cell count enumerations. The protocols for the NLAP are described here, since it was the largest and most
comprehensive database with sampled points throughout the US.
Details regarding the other sampling programs were found in the associated references mentioned previously. NLAP water samples were collected with an integrated sampler from the upper two meters, or less, of
the water column. The sampler was rinsed in triplicate by submerging it
vertically into the water column. Phytoplankton samples were preserved with Lugol's solution (USEPA, 2007a). Cell counts included 300
natural counts using a magnification of 1000 × or higher. Natural
counting units were defined as one for each colony, filament, or cell,
regardless if colonial, filamentous, or unicellular (USEPA, 2007b).
Picoplankton were not counted throughout the sample set, in addition
chlorophyll-a and phycocyanin fluorescence was lost after preservation
with Lugol's, limiting confirmation of picoplankton counts. Only the
Ohio database included cyanobacteria taxonomic information, but
none of these data were paired with MERIS estimates. Of the available
6014 in situ samples, 3946 (66%) were excluded from the analysis due
to missing data values or sample collection from N 2.0 m depth. Samples
with missing data included 41 samples (12%) in Florida, 3 samples
(4.0%) in Ohio, and 238 samples (4%) in New England flagged with no
data. Additionally, 3664 samples (69%) in New England were flagged
as out of depth range (N 2.0 m), leaving a balance of 2068 samples to
support algorithm validation. The majority of samples only contained
a single surface measurement within b 2.0 m depth. For a consistent
approach across all monitoring programs, it was decided only surface
(b2.0 m) would be used for this effort.
Cyanobacteria cell count ranges (cells/mL) were classified into Low
(10,000–109,999), Medium (110,000–299,999), High (300,000–1,000,000),
and Very High (N1,000,000) values (Table 1). The count ranges were
established as a compromise to correspond with ranges of concern to protect
ecosystem services and public health, and to best fit observed data distributions associated with cyanobacteria occurrences in the eastern US. Fig. 2
illustrates the frequency distribution of the in situ cell counts data for values
N 10,000 cells/mL (n = 939) and all data values (n = 2068), respectively.
A total of 1275 MERIS images were acquired for three study areas including six New England states, Ohio, and Florida over the 39-month
study period beginning in February 2009 through April 2012 (Fig. 1).
The MERIS CI-multi algorithm outputs were converted to cyanobacteria
cell count values ranging from approximately 10,000–3.0 M (cells/mL).
MERIS data were converted to cyanobacteria abundance (CA) using the
following equation (Wynne et al., 2010):
Cyano Abundance ðcells=mLÞ ¼ CI‐multi 1:0E þ 8
ð2Þ
The MERIS-derived cell count values were then matched with approximately 2068 independent in situ measurements taken from lakes
across the study area. Out of the 2068 available samples only 579
(28%) were matched with MERIS-derived cyanobacteria pixels using
the ±7-days window with the majority (82%) corresponding to “Low”
cell count range. Additionally, clouds and land excluded 17% and 34%
of the available data matches, respectively. Virtually all of the MERIS
correspondence pixels classified as land were located in Florida and
New England, where the sampling designs did not incorporate fixed locations, thus a high potential for erroneous coordinates. Visual analysis
of a subset of land pixels demonstrated that their occurrence was closely
tied to shorelines and very small lakes, both of which would be difficult
for MERIS to resolve.
Three separate analyses were conducted to pair the observed measurement data with the MERIS predicted values including: (i) 300 m direct correspondence (single pixel); (ii) 900 m average correspondence
(3 × 3 pixels); and (iii) 900 m closest value correspondence (450 m
buffer). These three analyses were performed to maximize the number
of match-ups between available in-situ data and MERIS derived
cyanobacteria cell counts for algorithm evaluation. All correspondence
analyses were performed for b 1-day, ± 1-day, ± 3-day, ± 5-day, and
± 7-day temporal windows. The 300 m analysis was accomplished
using a simple point-on-pixel overlay operation between in situ measurements and the corresponding 300 m MERIS pixel. A semi-automated GIS
routine was used to assign the values from corresponding MERIS pixels
within the various temporal windows to each in situ measurement. This
process resulted in the addition of as many as 15 fields (±7-day) to derive the value of the corresponding MERIS pixel on each available date
within a given temporal window. Output from this process was merged
Fig. 2. Histogram illustrating the distribution of in situ cell count values for N 10,000 cells/mL (n = 939) for cyanobacteria measurement data from lakes in Ohio, Florida, and New England
(2009–2012). Gray lines indicate break points.
R.S. Lunetta et al. / Remote Sensing of Environment 157 (2015) 24–34
29
Table 1
MERIS derived cyanobacteria cell count estimates versus in situ measurement data values from lakes in Ohio, Florida, and New England. Analysis was performed at multiple spatial scales
(300 × 300 m and 900 × 900 m) using ≤ 1-day, ± 1, ±3, ±5 and ±7 days temporal windows. The 900 m analysis included both an average MERIS-derived value (a) and closest correspondence pixel value (b).
Density range (cells/mL)
300 m
Correspondence (%)
900 ma
Weighted average (%)
900 mb
Sample size (n)
900 ma
300 m
900 mb
≤ 1-day
Not detectable
(b10,000)
Low
(10, 000–109,999)
Medium
(110,000–299,999)
High
(300,000–1,000,000)
Very high
(N1,000,000)
Total weighted correspondence
0
76
19
50
50
67
0
83
30
43
50
74
0
90
15
29
50
76
(NA)
(84)
(21)
(38)
(50)
(73)
178
94
16
2
2
114
278
122
20
7
2
151
304
159
26
7
2
172
± 1-days
Not detectable
(b10,000)
Low
(10, 000–109,999)
Medium
(110,000–299,999)
High
(300,000–1,000,000)
Very high
(N1,000,000)
Total weighted correspondence
0
75
19
63
79
68
0
85
21
35
64
74
0
90
18
27
73
76
(NA)
(85)
(19)
(36)
(72)
(73)
513
199
36
8
33
276
810
297
43
20
36
396
885
346
68
22
30
466
± 3-days
Not detectable
(b10,000)
Low
(10, 000–109,999)
Medium
(110,000–299,999)
High
(300,000–1,000,000)
Very high
(N1,000,000)
Total weighted correspondence
0
77
20
54
76
69
0
84
17
30
66
73
0
90
19
28
86
77
(NA)
(85)
(19)
(34)
(76)
(73)
1224
537
94
26
90
747
1981
715
98
50
94
957
2168
880
162
60
85
1187
± 5-days
Not detectable
(b10,000)
Low
(10, 000–109,999)
Medium
(110,000–299,999)
High
(300,000–1,000,000)
Very high
(N1,000,000)
Total weighted correspondence
0
79
17
59
77
70
0
84
17
34
67
73
0
90
17
33
83
77
(NA)
(85)
(17)
(39)
(75)
(74)
1926
874
151
46
150
1221
3115
1123
160
74
160
1517
3430
1417
256
90
144
1907
± 7-days
Not detectable
(b10,000)
Low
(10, 000–109,999)
Medium
(110,000–299,999)
High
(300,000–1,000,000)
Very high
(N1,000,000)
Total weighted correspondence
0
80
16
58
75
70
0
84
17
35
65
72
0
90
17
35
83
77
(NA)
(85)
(17)
(40)
(74)
(74)
2593
1157
206
62
198
1623
4175
1482
219
96
213
2010
4618
1883
347
113
193
2536
into a single geospatial layer and transposed to produce a unique record
for each date match between the in situ and MERIS data. Records containing no data, land, or cloud values were removed from the database.
Geospatial output was exported to Microsoft Access, where the
multi-temporal tables (e.g., ± 1-day, ± 3-day, etc.) were created and
the three study locations merged. Records were then classified into
bins based on in situ measurements representing Low to Very High
cell count ranges (Table 1). Values b 10,000 were classified as Not
Detectable because they were classified as absent in the MERIS products
and were not included in the subsequent validation process. The in
situ measurements were then plotted versus MERIS-derived values
(cells/mL) and data were analyzed using correspondence tables, histograms, and confusion matrices. Kappa coefficient values were calculated
based on a normalized matrix using the Margit and Kappa programs
(Congalton, 1991). The normalized matrix was calculated using Table 2.
The Kappa value was calculated using a normalized matrix due to the
disproportional number of samples (n = 1883) in the “Low” cell count
category to provide the most representative coefficient value. The
900 m average correspondence analysis incorporated MERIS cell count
values and followed the same 300 m analytical procedures, except the
MERIS imagery was averaged to a 900 m cell prior to performing the
point-on-pixel overlay. The pixels were aggregated to 900 m using a
Table 2
Contingency matrix illustrating the distribution of observed versus estimated cell count values (cells/mL), and commission and omission confusion for cyanobacteria MERIS-derived cell
counts. Field (in situ) measurement and MERIS imagery data were collected within a window of ± 7-days from lakes in Ohio, Florida, and New England. Cyanobacteria cell count ranges
correspond to Low (10,000–109,999), Medium (110,000–299,999, High (300,000–1,000,000) and Very High (N1,000,000). Matching pixels are selected based on closest corresponding
value within a 900 mb radius. The Kappa coefficient was calculated using normalized matrix values to mitigate disproportionate cell count values.
MERIS-derived data (predicted)
Field measurement data (observed)
Low
Medium
High
Very high
Column totals
% Correct
% Commission
Low
Medium
High
Very high
Row totals
% Correct
% Omission
1701
239
63
27
2030
84
16
37
59
3
0
99
60
40
116
43
39
5
203
19
81
29
6
8
161
204
79
21
1883
347
113
193
2536
90
17
35
83
10
83
65
17
77
Kappa = 0.57
30
R.S. Lunetta et al. / Remote Sensing of Environment 157 (2015) 24–34
mean statistical function that ignored ‘no data’ values in the calculation.
This process ensured that no masked pixels were included in the averaging of MERIS pixels. Therefore, samples that fell on land due to imprecise
coordinates from in situ records would not be incorporated in the 900 m
average correspondence analysis, because there was no pixel there for
the sample to intersect.
The 900 m closest correspondence analysis differs from the previous
two analyses because it was not reliant on direct spatial correspondence
with MERIS pixels. This vector-based approach first converted MERIS
imagery to points, and then selected the points that fell within a 450 m
buffer radius of each in situ measurement location. This was accomplished by applying a 450 m one-to-many spatial join to produce a
unique record across multiple temporal windows. The 900 m closest
correspondence analysis was used to search around an in situ measurement within the 450 m buffer to find the MERIS value that was closest to
the in situ cell count, or with an intersecting pixel within the same cell
count category if possible. If none of the pixels within the 450 m buffer
were within the same cell count category as the in situ sample, then the
search would use the closest value pixel within the 450 m buffer. These
data were then analyzed statistically in the same manner as the previous two analyses.
3. Results
Example spectra of MERIS rho_s (Fig. 4) were representative of notdetectable, Medium, High, and Very High count ranges where the CI
used 665, 681, and 709 nm wavebands. The 760 nm waveband was removed because it occurred at the O2 absorption for atmospheric correction. Calculated CI for not-detectable, Medium, and Very High count
ranges were −0.0008, 0.0013, and 0.0126 respectively. The CI for the
High spectra with 306,000 cells/mL was 0.0030 and the CI for the High
spectra with 510,000 cells/mL was 0.0051. The rho_s corrects only for
atmospheric Rayleigh radiance, so that the rho_s reflectance are higher
than expected for the observed water type. Standard atmospheric corrections for turbid water tends to overcorrect the blue bands, leading
to frequent negative reflectance in blooms (Wynne et al., 2010). The
use of the spectral shape on the red and NIR rho_s avoids errors caused
by atmospheric correction (Gower et al., 2008; Hu, 2009; Philpot, 1991;
Stumpf & Werdell, 2010).
There were 1960 paired observations above the not-detectable limit
(10,000 cells/mL) from in situ measures and the MERIS imagery between February 2009 and April 2012. Paired observations included
Fig. 3. Histogram illustrating the distribution of all available in situ data values (n = 2068)
for the five cyanobacteria cell count classifications from lakes in Ohio, Florida, and New
England for 2009–2012. Includes all available data values for sampling depths ≤ 2.0 m
(blank and zero data values omitted). To clarify, the ‘Not Detectable’ classification refers
to the MERIS algorithm result.
Fig. 4. Example MERIS spectra of not-detectable, Medium, High, and Very High count
ranges. The top two spectra were taken from a different scene with more aerosol than
the scene of the other spectra. The two spectra with wider lines have essentially the
same CI value.
any match-up within +/− 7 days of the satellite overpass. There was
a significant positive correlation between MERIS cyanobacteria cell
counts and in situ cyanobacteria cell counts (Fig. 5, slope = 0.94,
R2 = 0.87, p b 0.0001, RMSE = 225,369 cells/mL). Due to the high
RMSE additional correspondence comparisons were undertaken.
Table 1 summarizes validation results across the eastern US for
inland lakes located in New England (RI, MA, NH, ME, NY, and VT),
Florida, and Ohio (Fig. 1). Statistics for lake color, turbidity, and
chlorophyll-a concentrations were summarized to demonstrate the
range of optically active constituents across lakes (Table 3). Correspondence comparisons (MERIS-derived estimates vs. in situ measurement
data) were performed for multiple temporal windows including
≤ 1-day, and ± 1, 3, 5, and 7 days. With the exception of ≤ 1-day,
all temporal windows contained a sufficient number of samples to
support a robust data evaluation. For the ≤ 1-day window, sample
sizes for the Medium, High, and Very High cell count ranges were insufficient for meaningful analysis. Similarly, the ± 1-day window
had too small a sample size for the High cell count range. All other
temporal ranges had sufficient sampling sizes to support a correspondence based assessment of performance. The correspondence
values for ± 1, ± 3, ± 5, and ± 7 days, were very similar. Both
the Low and Very High data range values had the highest
Fig. 5. Regression of measured vs. MERIS estimated cell counts for all data with match-up
to MERIS within +/−7 days.
R.S. Lunetta et al. / Remote Sensing of Environment 157 (2015) 24–34
31
Table 3
Summary statistics of colored dissolved organic matter (CDOM, PCU), turbidity, and chlorophyll-a that demonstrate the range of lake optical complexity. Data was only from surface samples in the National Lakes Assessment Program.
Location
New England (CT,MA,ME,NH,RI,VT)
FL
OH
CDOM (PCU)
Chl-a (mg m−3)
Turbidity (NTU)
Mean
STD
Max
Min
Mean
STD
Max
Min
Mean
STD
Max
Min
17.0
29.0
12.8
16.2
27.8
8.4
140.0
125.0
40.0
0.0
4.0
3.0
1.8
10.3
16.2
1.7
10.3
21.3
7.9
34.7
84.2
0.3
1.2
0.7
5.2
48.9
36.0
6.0
51.7
44.3
31.4
187.9
189.4
0.7
1.5
1.0
correspondence (90% and 86%, respectively) using the closest correspondence approach (900 m), the Medium range was relatively insensitive to the approach used (15–19%), and the High range
performed best using the pixel-wise approach (54–63%). The relative correspondence between the field measurement values and
MERIS-derived cell counts can be characterized based on a correspondence range for any given cell count range metric (e.g., Very
High = 76%–86%), highest correspondence value (e.g., Very High =
86%) or weighted average correspondence (e.g., Very High =
75.5%). Each metric can provide valuable information, but they are
probably best used in combination to evaluate performance. It is important to note that these correspondence values incorporate the inaccuracies (errors) associated with both the reference data (in situ
cell counts) and MERIS-derived estimates (Lunetta, Iiames, Knight,
Congalton, & Mace, 2001). Accordingly, regardless of the correspondence metric applied, the values represent conservative estimates
for algorithm performance.
There was an elevated performance for the High value range 300 m
pixel-wise analysis (Table 1). However, the other cell count ranges
typically performed better in the 900 m analysis. Visual analysis of a
large subset of records indicated that the elevated misclassifications associated with the 900 m analyses were attributable to shoreline (land)
pixels that were included in the 900 m closest correspondence analysis
but were flagged as land in the 300 m pixel-wise correspondence analysis and excluded.
The results of our analysis indicate that the algorithm can estimate
cyanobacteria cell counts for inland lakes in the eastern US at Low
(10,000–109,999) and Very High (N 1,000,000) correspondence levels
of approximately 90% and 83%, respectively. The intermediate cell
count estimates (110,000–299,999 and 300,000–1,000,000) were
below acceptable performance levels at approximately 28% and 40%, respectively (Table 1). Table 2 provides a detailed analysis of confusion
between cell count ranges including the percent correct correspondence, and the percentages for both commission (Type I errors) and
omission (Type II errors). The Low range had the best performance
with an accuracy of 90%, only 10% omission, and moderate commission
errors (16%). The Medium and High cell count ranges both had high
omission (83% and 65%, respectively) and high commission errors
(40% and 81%, respectively). The errors associated with the intermediate count ranges were mostly a result of confusion with the Low count
range. The Very High cell count range has a correspondence accuracy
of 83% with omission (17%) and commission (21%) errors nearly equally
distributed. Overall performance resulted in a Kappa coefficient of 0.57.
This can be interpreted as performing 57% better than random assignments across all cell count classes.
To best interpret the correspondence results, a thorough analysis of
the characteristics associated with the in situ reference data is presented
in Fig. 3. The frequency distribution of the in situ cell count data indicated that 54.6% of the samples (n = 1129) were below the minimal
MERIS detection limit (b 10,000), 32.5% (n = 672) were in the Low
range, 5.8% (n = 121) were in the Medium range, only 2.2% (n = 45)
in the High range, and 4.8% (n = 101) in the Very High range. Wynne
et al. (2010), Wynne et al. (2013) suggested that detection was possible
to 35,000 cells/mL, but had insufficient field data to provide more resolution. MERIS has the sensitivity to detect variations of CI to 0.0001,
which corresponds to 10,000 cells/mL from Eq. 2, at which
concentration there is no public health concern. Of the inland lakes sampled, approximately 45% had cyanobacteria cell count values
N 10,000 cells/mL and only 12.8% N 100,000 cells/mL. The frequency
distribution of the reference data were not normally distributed, but
instead exhibited a distribution that was heavily skewed towards the
low to medium ranges (Fig. 3). The skewed frequency distribution
should be considered when evaluating the nature of the correspondence errors. It's important to properly interpret the count range
confusion in Table 2; which was also skewed towards the Low
range, coinciding with the distribution of the reference data. Also,
the fact that breaks in the reference data were apparent between
the Low and Medium cell count ranges, would make differentiation
using MERIS-derived estimates difficult (Fig. 2). If the break point
between Low and Medium ranges were set at 100,000, it resulted
in a cluster of values around the break point. Therefore, the Low
and Medium breakpoint was moved to 110,000 so that similar values
were not split across categories.
Fig. 6 provides a final illustration of the total number of correspondence values (n = 2536) corresponding to Table 1 for ±7-days. Note
that the high number of samples is a result of multiple correspondence
comparisons (hits) between the MERIS-derived estimated and reference data over the prolonged temporal window of 15-days. The figure
illustrates both the skewed distribution of the reference data and the
relative correspondence between the predicted and observed values
in Table 2.
4. Discussion
The high incidence (45%) of inland lakes with cyanobacteria cell
counts N 10,000 cells/mL confirms that occurrences of cyanobacteria
blooms are now common events across the eastern US. However, due
to the absence of a comprehensive monitoring program, it is not
known if this represents an increased incidence or simply reflects
Fig. 6. Histogram illustrating the distribution of correspondence data values (n = 2536)
for MERIS-derived (gray) and in situ (black) cyanobacteria cell counts for a ± 7-days
window from lakes in Ohio, Florida, and New England. Note that the sample size (n) on
the y-axis is inflated. due to duplication of in situ values to match multiple MERIS image
dates. Each in situ record. has the potential to be duplicated 15 times to account for
the ± 7-days window.
32
R.S. Lunetta et al. / Remote Sensing of Environment 157 (2015) 24–34
long-term baseline conditions (Fristachi & Sinclair, 2008). However,
there is widespread agreement among scientists and managers that
cyanobacteria blooms are increasing (Hudnell, 2010) including one recent study documenting a record setting bloom event in Lake Erie in
2011 that was attributed to climatic and nutrient mediated changes
consistent with anticipated future trends (Michalak et al., 2013). Others
have also documented troubling trends in environmental perturbations
including water temperature increases attributable to climate change
(Paerl & Paul, 2012) and increased nutrient loadings resulting from
intensive agricultural practices (USEPA, 2013b), these trends are indicative of potential risk factors for increased cyanobacteria bloom
occurrences.
The results presented in Table 1 provide an interesting insight
concerning the duration of cyanobacteria bloom events. Very little variability in the correspondence results was observed for temporal windows ranging from 3 to 15 days in duration. This result indicates that
the events tend to be of relatively long duration and are typically stable
for extended periods of time. No attempt was made to go beyond the
maximum 15-day temporal window presented here. Thus, no additional insights concerning viable maximum temporal windows for derived
product calibration and validation efforts can be provided.
This study has validated the robust retrieval capability of cyanobacteria cell counts for inland lakes in the eastern US that coincide
with Low (10,000–109,999) and Very High (N 1,000,000) values. The
confusion with the intermediate cell count ranges is largely attributed
to confusion with the low cell count range (Table 1). However, this
association needs to be tempered due to the skewed nature of the in
situ reference data that was dominated (72%) by low range values
(n = 672). Intermediate cell count estimates (110,000–299,999
and 300,000–1,000,000) were below acceptable performance levels.
A plausible explanation for the substandard performance for the intermediate “Medium and High” cell count values is related to the
cell count measurements used for cyanobacteria monitoring. First,
the uncertainty associated with the cell count laboratory procedures
is unknown and the distinction of natural counting units was limited.
Also, we hypothesize that the algorithm was measuring the intensity
of phycocyanin and Chl-a absorption features (λ = 620 and 665 nm)
to derive the cyanobacteria cell count estimates. As previously mentioned, there were issues in the comparison of two measured quantities with no direct overlap; optical signals of light absorption and
scattering by pigments in cells, and cell counts without biovolume
or biomass information. The ratio of pigmentation and biomass is
not constant, and neither is the ratio of cell number to biomass. Spatial variability can be a factor in mis-match between satellite and
field measurements. Cyanobacteria blooms have extreme spatial
variability as documented by Wynne et al. (2010); thus a 300 m pixel
(9.0 ha) can contain a wide range of cyanobacteria cell densities.
Additionally, while the CI second derivative approach, centered on
the MERIS fluorescence band at 681 nm, is far less sensitive to poor
atmospheric correction, thin clouds can attenuate the CI values,
which would lead to underestimates from satellite. Over the six
state geographic extent of the study area, we would expect that the
cyanobacteria measurement data would incorporate a number of
different cyanobacteria taxa. Importantly, the biovolume (BV) impacts on reflectance as determined by the quantity of pigment and
scattering (cross-sectional area) of individual species can differ by
an order of magnitude (e.g., Microcystis sp. (small) = 19 μm 3 vs.
Anabaena circinalis = 208 μm3) and even Microcystis sp. (large) BVs
exhibit large differences (93 μm3). Accordingly, cyanobacteria cell
counts should optimally be normalized to BV (mL/L) before interspecies comparisons or analysis is performed (MEMH, 2009).
Biovolumes are not currently standard and pigment measurements
are rare. However, for cyanobacteria remote sensing estimates, phycocyanin and Chl-a concentrations would likely be more closely related to BV's than cell counts for heterogeneous cyanobacteria taxa.
Of the available 6000 monitoring samples available over the study
Table 4
Technical characteristics of the ENVISAT MEdium Resolution Imaging Spectrometer
(MERIS) and Sentinel-3 Ocean and Land Colour Instrument (OLCI) systems (Donlon
et al., 2012). Note that the highlighted channels represent new or substantial modifications
to the MERIS configuration. Data are available at 300 m with listed signal-to-noise ratios
(SNRs).
Channel
Wavelength (nm)
Band Width (nm)
SNR
Application
10
10
10
10
10
10
10
7.5
9
7.5
3.75
14
20
10
10
468: 1
413: 1
355: 1
306: 1
289: 1
216: 1
177: 1
147: 1
158: 1
101: 1
375: 1
157: 1
114: 1
113: 1
51: 1
Yellow substance
Chl-a absorption max
Accessory pigments
Sediment/turbidity
Chl-a minimum
Phycocyanin/sediment
Chl-a absorption max
Chl-a fluorescence peak
Chl-a baseline
O2absorption/clouds
O2absorption/aerosols
Atmospheric corrections
Atmospheric corrections
Water vapor reference
Water vapor absorption
MERIS (2–3 days)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
412.5
442.5
490
510
560
620
665
681.3
709
753.8
760.6
775
865
885
900
OLCI (2–3 days, 1 satellite; 1–3 days, 2 satellites)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
400
412.5
442.5
442
510
560
620
665
673.75
681.25
708.75
753.75
761.25
764.375
767.5
778.75
865
885
900
940
1020
15
10
10
10
10
10
10
10
7.5
7.5
10
7.5
2.5
3.75
2.5
15
20
10
10
20
40
2,188:1
2,061:1
1,811:1
1,541:1
1,488:1
1,280:1
997:1
883:1
707:1
745:1
785:1
605:1
232:1
305:1
330:1
812:1
666:1
395:1
308:1
203:1
152:1
Aerosol correction
Chl-a fluorescence
Atmospheric correction
Cloud tops
Atmospheric correction
Atmospheric correction
Aerosol correction
period, only 34% were suitable for validation analysis. Open and
effective discussions and forums between scientists, stakeholders,
policy makers, and environmental managers would be required to
modify existing field sampling designs for providing high quality
reference data for future validation efforts of satellite sensor observations (Schaeffer et al., 2013).
The results of this study show that MERIS products can be used to
monitor the onset of cyanobacteria bloom events and potentially provide a capability for monitoring outbreaks of concern in freshwater
lakes. MERIS, and in the future Sentinel-3, provide a potential for
retrieving data for small lakes at 1–3 day intervals. This represents a
major advancement over current in situ based monitoring programs
by providing the capability for near-real time products over large geographic areas (regional–national) at high temporal intervals. Implementation of an operational satellite cyanobacteria bloom monitoring
program would have the benefit of the European Space Agency's
(ESA) new Ocean Land Colour Imager (OLCI) on the Sentinel-3 satellite
scheduled for launch and operational service beginning in 2015. The
OLCI was designed to complement MERIS by providing operational
system enhancements including improved signal to noise ratio (SNR),
additional bands for improved atmospheric corrections, and pointing
to reduce sun glint (Table 4). Additionally, ESA plans include the eventual deployment of two OLCI sensors that would provide a 1–2 day
revisit capability for most US locations of interest.
An additional application would be to use the robust detection capability for Low cyanobacteria cell counts to trigger predictive model implementation coinciding with the onset of cyanobacteria bloom events
to forecast the potential for intensification. Subsequent OLCI-derived
cell counts could then be used to generate alarm products for lakes
R.S. Lunetta et al. / Remote Sensing of Environment 157 (2015) 24–34
exhibiting Very High cell counts that could be applied to validate modeling predictions by evaluating the percentage of correct predictions and
characterizing Type I and II errors. The application of a predictive model
would potentially provide valuable forecast information to local managers, decision makers and scientists for directing limited field monitoring resources and to provide public recreational advisories. A robust
modeling capability would provide a forecasting capability and would
serve to fill the technological gap for the middle range cyanobacteria
cell counts.
5. Conclusions
The results of this study document the potential for using MERISderived cyanobacteria cell count estimates to monitor freshwater
lakes (N100 ha) across the eastern US. The capability to accurately estimate Low and Very High cell count ranges represent a significant capability for the development of an operational monitoring program. The
substandard performance for intermediate cell count ranges could be
attributed to the possibility that the algorithm does not have sensitivity
to the presence of cyanobacteria in this range, the lack of available in
situ cyanobacteria taxonomic data, or the distinction of natural counting
units to support the conversion of cell counts to cell volumes. Because
the algorithm functions by detecting the impacts of absorption by phycocyanin and Chl-a on the spectral curvature (CI-multi), correlations
would optimally be compared with cell volume (surface area) or pigment concentrations versus cell counts. The potential development of
a cyanobacteria predictive modeling capability for integration with
future OLCI-derived cell count estimates may provide a good solution
for an operational monitoring capability that could support cyanobacteria bloom predictions for the eastern US. This study was a first
attempt to use existing monitoring programs across multiple states
and agencies to validate remote sensing cyanobacteria cell count estimates. Finally, to support future validation efforts of satellite sensor observations of freshwater water quality parameters, modifications and
standardization of existing algal field sampling and cytological enumeration protocols are needed to provide high quality reference data.
Acknowledgments
The research described in this article was partially funded and conducted by the USEPA's Office of Research and Development (ORD)
under a Pathfinder Innovation Project II (PIP2) grant and by the NASA
Applied Science Program announcement NNH08ZDA001N (Human
Health and Air Quality) under contract NNH09AL531. It has been subject to the US EPA programmatic review and has been approved for publication. Mention of any trade names or commercial products does not
constitute endorsement or recommendation for use. The authors
would like to acknowledge the assistance of Keith Loftin (USGS),
Robyn Conmy (USEPA), James Hagy (USEPA), Bryan Milstead (USEPA),
Lesley Vazquez-Coriano (USEPA), and Jan Kurtz (USEPA), Tim Wynne
(NOAA) and Shelly Tomlinson (NOAA), Rick Wilson (Ohio EPA), Russ
Gibson (Ohio EPA), Dan Glomski (Ohio EPA), and Travis Briggs (CSS).
The Florida data used in the study were collected and provided by the
St. Johns River Water Management District (WMD) Division of Water
Resources, Environmental Sciences Bureau by Rolland Fulton, John
Hendrickson, Robert Burks, and Erich Marzolf all with the Suwannee
River WMD.
References
Abbott, G. M., Landsberg, J. H., Reich, A.R., Steidinger, K. A., Ketchen, S., & Blackmore, C.
(2009). Resource guide for public health response to harmful algal blooms in Florida.
Fish and Wildlife Research Institute Technical Report TR-14 (viii + 132 pp.).
Alikas, K., Kangro, K., & Reinart, A. (2010). Detecting cyanobacterial blooms in large North
European lakes using the Maximum Chlorophyll Index. Oceanologia, 52, 237–257.
Backer, L. C. (2002). Cyanobacterial Harmful Algal Blooms (CyanoHABs): Developing a
public health response. Lake and Reservoir Management, 18, 20–31.
33
Becker, R. H., Sultan, M. I., Boyer, G. L., Twiss, M. R., & Konopko, E. (2009). Mapping
cyanobacterial blooms in the Great Lakes using MODIS. Journal of Great Lakes
Research, 35, 447–453.
Beskenis, B. (2014). Draft standard operating procedures for cyanobacteria counts. Worcester,
MA: Division of Watershed Management, Massachusetts Department of Environmental
Protection.
Binding, C. E., Greenberg, T. A., Jerome, J. H., Bukata, R. P., & Letourneau, G. (2011). An
assessment of MERIS algal products during an intense bloom in Lake of the Woods.
Journal of Plankton Research, 33, 793–806.
Budd, J. W., Drummer, T. D., Nalepa, T. F., & Fahnenstiel, G. L. (2001). Remote sensing of
biotic effects: Zebra mussels (Dreissena polymorpha) influence on water clarity in
Saginaw Bay, Lake Huron. Limnology and Oceanography, 46, 213–223.
Chorus, I., & Bartram, J. (1999). Toxic cyanobacteria in water: A guide to their public health
consequences, monitoring and management. Bury St. Edmunds, Suffolk, UK: St.
Edmundsbury Press (400 pp.).
Chorus, I., Salas, H. J., & Bartram, J. (2000). Health risks caused by freshwater
cyanobacteria in recreational waters. Journal of Toxicology and Environmental Health,
Part B, 3, 323–347.
Congalton, R. G. (1991). A review of assessing the accuracy of classifications of remotely
sensed data. Remote Sensing of Environment, 37, 35–46.
Dash, P., Walker, N. D., Mishra, D. R., Hu, C., Pinckney, J. L., & D'Sa, E. J. (2011). Estimation
of cyanobacterial pigments in a freshwater lake using OCM satellite data. Remote
Sensing of Environment, 115, 3409–3423.
Dodds, W. K., Bouska, W. W., Eitzmann, J. L., Pilger, T. J., Pitts, K. L., Riley, A. J.,
Schloesser, J. T., & Thornbrugh, D. J. (2009). Eutrophication of U.S. freshwaters:
Analysis of potential economic damages. Environmental Science and Technology,
43, 12–19.
Donlon, C., et al. (2012). The Global Monitoring for Environment and Security (GEMS)
Sentinel-3 mission. Remote Sensing of Environment, 120, 37–57.
Duan, H., Ma, R., & Hu, C. (2012). Evaluation of remote sensing algorithms for
cyanobacterial pigment retrievals during spring bloom formation in several lakes of
East China. Remote Sensing of Environment, 126, 126–135.
ESS Group, Inc. (2011). Cyanobacteria monitoring program. Rhode Island Department of
Environmental Management. ESS Project No. R298-010. RIDEM REQ# 1180565/
1194117.
Fristachi, A., & Sinclair, J. L. (2008). Occurrence of cyanobacteria harmful algal blooms
working group. Cyanobacterial harmful algal blooms: State of the science and research
needs. Advances in Experimental Medicine and Biology, 619. (pp. 45–103).
Gómez, J. A.D., Alonso, C. A., & García, A. A. (2011). Remote sensing as a tool for monitoring water quality parameters for Mediterranean Lakes of European Union
water framework directive (WFD) and as a system of surveillance of cyanobacterial
harmful algae blooms (SCyanoHABs). Environmental Monitoring and Assessment,
181, 317–334.
Gower, J., King, S., & Goncalves, P. (2008). Global monitoring of plankton blooms using
MERIS MCI. International Journal of Remote Sensing, 29, 6209–6216.
Graham, J. L., Jones, J. R., Jones, S. B., Downing, J. A., & Clevenger, T. E. (2004). Environmental factors influencing microsystin distributions and concentrations in the Midwestern United States. Water Research, 38, 4395–4404.
Graham, J., Loftin, K., & Kamman, N. (2009). Monitoring recreational freshwaters. Lakeline:
North American Lake Management Society, 18–24 (Summer).
Hu, C. (2009). A novel ocean color index to detect floating algae in the global oceans.
Remote Sensing of Environment, 113, 2118–2129.
Hudnell, H. K. (2010). The state of U.S. freshwater harmful algal blooms assessments,
policy and legislation. Toxicon, 55, 1024–1034.
Hunter, P. D., Tyler, A. N., Carvalho, L., Codd, G. A., & Maberly, S.C. (2010). Hyperspectral
remote sensing of cyanobacterial pigments as indicators for cell populations and
toxins in eutrophic lakes. Remote Sensing of Environment, 114, 2705–2718.
Kahru, M., Savchuk, O. P., & Elmgren, R. (2007). Satellite measurement of cyanobacteria
bloom frequency in the Baltic Sea: Interannual and spatial variability. Marine
Ecology Progress Series, 343, 15–23.
Kutser, T. (2009). Passive optical remote sensing of cyanobacteria and other intense phytoplankton blooms in coastal and inland waters. International Journal of Remote
Sensing, 30, 4401–4425.
Lunetta, R. S., Iiames, I., Knight, J., Congalton, R. G., & Mace, T. H. (2001). An assessment of
reference data variability using a “virtual field reference database”. Photogrammetric
Engineering and Remote Sensing, 67(6), 707–715.
Lunetta, R. S., Shao, Y., Ediriwickrema, J., & Lyon, J. G. (2010). Monitoring agricultural
cropping patterns across the Laurentian Great Lakes Basin using MODIS-NDVI data.
International Journal of Applied Earth Observation and Geoinformation, 12, 81–88.
Matthews, M. W., Bernard, S., & Robertson, I. (2012). An algorithm for detecting trophic status (chlorophyll-a), cyanobacteria-dominance, surface scums and floating vegetation in inland and coastal waters. Remote Sensing of Environment,
124, 637–652.
Matthews, M. W., Bernard, S., & Winter, K. (2010). Remote sensing of cyanobacteriadominant algal blooms and water quality parameters in Zeekoevlei, a small hypertrophic lake, using MERIS. Remote Sensing of Environment, 114, 2070–2087.
MEMH (Ministry for the Environment and Ministry of Health) (2009). New Zealand
Guidelines for Cyanobacteria in Recreational Fresh Waters – Interim Guidelines. Prepared by S.A. Wood, W.J. Hamilton, W.J. Paul, K.A. Safi and W.M. Williamson.
Wellington, New Zealand: Ministry for the Environment (82 pp.).
Michalak, A.M., et al. (2013). Record-setting algal bloom in Lake Erie caused by agricultural
and meteorological trends consistent with future conditions (early ed.). PNAS.
NH Department of Environmental Services (2005a). Standard operating procedure for
algae collection/identification. Beach Program.
NH Department of Environmental Services (2005b). Standard operating procedures for
cyanobacteria identification. Beach Program, Limnology Center Laboratory Manual.
34
R.S. Lunetta et al. / Remote Sensing of Environment 157 (2015) 24–34
NRC (National Research Council) (2012). Exposure Science in the 21st Century: A Vision and
a Strategy. Washington, DC. Washington, DC: The National Academies Press (195 pp.).
Ohio EPA (2012). Public water system harmful algal bloom response strategy. http://
www.epa.ohio.gov/Portals/28/documents/PWS_HAB_Response_Strategy_5–30–12.
pdf
Paerl, H. W. (2008). Nutrient and other environmental controls of harmful cyanobacterial
blooms along the freshwater-marine continuum. In H. K. Hudnell (Ed.), Cyanobacterial
harmful algal blooms: State of the science and research needs, 619. (pp. 217–237).
Springer Press.
Paerl, H. W., & Huisman, J. (2008). Blooms like it hot. Science, 320, 57–58.
Paerl, H. W., & Paul, V. G. (2012). Climate change: Links to global expansion of harmful
cyanobacteria. Water Research, 46, 1349–1363.
Philpot, W. D. (1991). The derivative ratio algorithm: Avoiding atmospheric effects in
remote sensing. IEEE Transactions on Geoscience and Remote Sensing, 29, 350–357.
Riedinger-Whitmore, M.A., Whitmore, T. J., Smoak, J. M., Brenner, M., Moore, A., Curtis, J.,
& Schelske, C. L. (2005). Cyanobacteria proliferation is a recent response to eutrophication in many Florida lakes: A paleolimnological assessment. Lake and Reservoir
Management, 21(4), 423–435.
Ruiz-Verdú, A., Simis, S. G. H., Hoyos, C. D., Gons, H. J., & Peña-Martínez, R. (2008). An evaluation of algorithms for the remote sensing of cyanobacterial biomass. Remote
Sensing of Environment, 112, 3996–4008.
Schaeffer, B.A., Schaeffer, K. G., Keith, D., Lunetta, R. S., Conmy, R., & Gould, R. W. (2013).
Barriers to adopting satellite remote sensing for water quality management.
International Journal of Remote Sensing, 34, 7534–7544.
Seppälä, J., Ylostalo, P., Kaitala, S., Hallfors, S., Raateoja, M., & Maunual, P. (2007). Ship-ofopportunity based phycocyanin fluorescence monitoring of the filamentous
cyanobacteria bloom dynamics in the Baltic Sea. Estuarine, Coastal and Shelf Science,
73, 489–500.
Simis, S. G. H., Peters, S. W. M., & Gons, H. J. (2005). Remote sensing of the cyanobacteria
pigment phycocyanin in turbid inland water. Limnology and Oceanography, 50,
237–245.
Smayda, T. J. (1997). What is a bloom? A commentary. Limnology and Oceanography, 42,
1132–1136.
Soranno, P. A. (1997). Factors affecting the timing of surface scums and epilimnetic
blooms of blue-green algae in a eutrophic lake. Canadian Journal of Fisheries and
Aquatic Sciences, 54, 1965–1975.
Stewart, I., Webb, P.M., Schluter, P. J., & Shaw, G. R. (2006). Recreational and occupational field exposure to freshwater cyanobacteria — a review of anecdotal and
case reports, epidemiological studies and the challenges for epidemiologic assessment. Environmental Health, 5:6, http://dx.doi.org/10.1186/147-069X-5-6
(13 pp.).
Stumpf, R. P., & Werdell, P. J. (2010). Adjustment of ocean color sensor calibration through
multi-band statistics. Optics Express, 18(2), 401–412.
Stumpf, R. P., Wynne, T. T., Baker, D. B., & Fahenstiel, G. L. (2012). Interannual variability of
cyanobacteria blooms in Lake Erie. PLoS ONE, 7(8), e4244, http://dx.doi.org/10.1371/
journal.pone.0043444.
USEPA (United States Environmental Protection Agency) (2007a). 2007 National Lakes
Assessment. Washington, DC: Laboratory Operations Manual.
USEPA (United States Environmental Protection Agency) (2007b). Survey of the Nation's
Lakes. Field Operations Manual. Washington, DC. EPA 841-B-07-004.
USEPA (United States Environmental Protection Agency) (2009). National Lakes Assessment:
A collaborative Survey of the Nation's Lakes. Washington, DC. EPA 841-R-09-001.
USEPA (United States Environmental Protection Agency) (2011). Biofuels and the
Environment: The first Technical Report to Congress. U.S. Environmental Protection Agency (2011 Final Report). Washington, DC, EPA/600/R-10/183 F (220 pp.).
USEPA (United States Environmental Protection Agency) (2013a). CCL and Regulatory
Determinations: CCL 3 List. http://water.epa.gov/scitech/drinkingwater/dws/ccl/
index.cfm
USEPA (United States Environmental Protection Agency) (2013b). National Rivers and
Streams Assessment 2008–2009: A Collaborative Survey. USEPA, Washington DC,
EPA Report No. EPA/841/D-13/001 (110 pp.).
Vincent, R. K., Qin, X., Mckay, R. M., & Miner, J. (2004). Phycocyanin detection from
Landsat TM data for mapping cyanobacterial blooms in Lake Erie. Remote Sensing of
Environment, 89, 381–392.
VT Department of Environmental Conservation (2014). Lake Champlain long term water
quality and biological monitoring program. http://www.watershedmanagement.vt.
gov/lakes/docs/lcmonitoring/lp_lclongtermprogdesc.pdf
Watzin, M. C., Fuller, S., Bronson, L., Gorney, R., & Schuster, L. (2011). Monitoring and evaluation of cyanobacteria in Lake Champlain, Summer 2009. Technical Report No. 61.
Rubenstein Ecosystem Science Laboratory. Rubenstein School of Environment and Natural Resources, University of Vermont.
Wynne, T. T., Stumpf, R. P., Tomlinson, M. C., & Dybleb, J. (2010). Characterizing a
cyanobacterial bloom in western Lake Erie using satellite imagery and meteorological
data. Limnology and Oceanography, 55, 2025–2036.
Wynne, T. T., Stumpf, R. P., Tomlinson, M. C., Fahnenstiel, G. L., Dybleb, J., Schwab, D. J., &
Joseph-Joshi, S. (2013). Evolution of a cyanobacteria bloom forecast system in western
Lake Erie: Development and initial evaluation. Journal of Great Lakes Research, http://
dx.doi.org/10.1016/j.jgir.2012.10.003.
Wynne, T. T., Tomlinson, M. C., Warner, R. A., Tester, P. A., Dyble, J., & Fahnensteil, G. L.
(2008). Relating spectral shape to cyanobacterial blooms in the Laurentian Great
Lakes. International Journal of Remote Sensing, 29, 3665–3672.