Novel optical techniques for detecting and classifying toxic dinoflagellate Karenia brevis blooms using satellite imagery

Ruhul Amin; Jing Zhou; Alex Gilerson; Barry Gross; Fred Moshary; Samir Ahmed

doi:10.1364/OE.17.009126

1. Introduction

More than 40 species of toxic microalgae live in the Gulf of Mexico, but the most common is the toxic dinoflagellate Karenia brevis (K. brevis) formerly named Gymnodinium breve [1]. Although K. brevis blooms have been reported throughout the Gulf of Mexico, they are most frequent along the West Florida Shelf (WFS) where they occur nearly every year, usually between late fall and early spring but occasionally at other times of the year as well. K. brevis blooms have many negative impacts due to brevetoxin. This associated toxin causes death in fish, birds, and marine mammals [2]. It also can irritate human eyes and respiratory systems once it becomes airborne in sea spray [3,4].

Many studies have been conducted to explore the optical characteristics of K. brevis [5–12]. Cannizaro et al., (2008) [5] suggested that K. brevis blooms exhibit lower backscattering compared to other phytoplankton and this low backscattering efficiency has been related to its large size (20-40 µm) and low index of refraction (~1.05) [6]. Furthermore, backscattering of light in the ocean is typically dominated by submicron particles [13,14], and thus the low backscatter associated with K. brevis blooms may also reflect lower associated concentration of submicron particles [15,16]. Schofield et al., (2006) [8] suggested that the lower concentration of non-algal particles may be due to the toxicity of K. brevis cells which may directly inhibit bacterial growth and/or alter the organic material available for heterotrophic consumption.

Detecting and monitoring K. brevis blooms from field measurements is labor intensive and suffers from practical limitations on achievable temporal and spatial resolutions. Field measurements are typically made at discrete points and at discrete times, without much temporal continuity. To assist in the detection of blooms and the planning preparation of remedial measures to reduce health risks, etc., detection approaches with higher spatial and temporal resolutions are desirable. It is in this context that satellite Ocean Color sensors offer potential advantages for bloom detection and monitoring.

Stumpf et al., (2003) [17] proposed to use the magnitude of the difference between satellite chlorophyll concentration estimates and a background mean of chlorophyll estimates for the previous 0.5-2.5 months as an index for detecting bloom areas. At NOAA NESDIS CoastWatch, this method is now used operationally to alert for possible blooms in West Florida. Cannizzaro et al., (2008) [5] proposed another technique based on in situ data that uses the backscattering/chlorophyll ratio to discriminate between K. brevis and other blooms. They determined that K. brevis has lower backscatter characteristics than blooms of other diatoms and dinoflagellate species. Both of these methods are based on using the blue-green region of the spectrum. Unfortunately, blue-green reflectance ratio algorithms [18–20] have been found to perform poorly in coastal waters due to increased absorption of colored dissolved organic matter (CDOM), increased particle scattering, inaccurate atmospheric corrections and shallow bottom reflectance. Hu et al., (2005) [21] used Fluorescence Line Height (FLH) to detect and monitor K. brevis bloom on the WFS. However, our studies have shown that FLH strongly overestimates the chlorophyll fluorescence signal under high elastic scattering conditions, resulting in false positives [22–24].

The objective of this study is to develop techniques for bloom detection and K. brevis bloom classification that are less sensitive to CDOM and atmospheric corrections, and apply them to satellite data to detect and classify K. brevis blooms in the Gulf of Mexico. In this study, we illustrate these approaches using the Moderate Resolution Imaging Spectroradiometer (MODIS) ocean color sensor which has several bands in the red and near-infrared (NIR) regions. This complements our previous study [25] where we used the Medium Resolution Imaging Spectrometer (MERIS) ocean color data.

In Section 2 that follows, the procedures for obtaining satellite imagery are briefly described. In Section 3, we present simulations of remote-sensing reflectance spectra ( $R_{r s} (λ)$ ) for K. brevis and non-K. brevis blooms, introduce the proposed detection and classification techniques and discuss their backgrounds. Section 4 applies these techniques to satellite ocean color data and shows examples of detection, tracing, and classification of K. brevis blooms. We also compare the RBD detection and FLH techniques, and carry out an analysis of the impacts of atmospheric correction algorithms on the RBD and KBBI techniques. Section 5 discusses the various advantages of the proposed techniques over other traditional techniques and examines possible false alarm conditions.

2. Satellite Data and Image Processing

MODIS imagery of WFS was obtained for different K. brevis blooms recorded in the literature dating from 2001 to 2006 from the NASA Ocean Color Website [26] and processed to obtain the normalized water-leaving radiance ( $n L w (λ)$ ) for visible and NIR bands, and FLH using SeaDAS version 5.0.3. The top of the atmosphere signals were corrected for the atmosphere using both the standard NIR [27] method and also the recently proposed [28] SWIR method for turbid coastal waters. The data was processed with a pixel size of 1km equal to the nominal pixel size of the sensor’s ocean color bands.

3. Development of Bloom Detection and K. brevis Bloom Classification Techniques

3.1 Optical Modeling of K. brevis and non-K. brevis Blooms

Remote-sensing reflectance, $R_{r s} (λ)$ is defined as the ratio of water-leaving radiance to down welling irradiance, measured just above the sea surface. $R_{r s} (λ)$ can be approximately related to the subsurface remote-sensing reflectance, $r_{r s} (λ)$ [29] by:

R_{r s} (λ) = \frac{0.5 r_{r s} (λ)}{(1 - 1.5 r_{r s} (λ))}

where

r_{r s} (λ)

for optically deep waters is given, according to [30] as:

r_{r s} (λ) = (0.084 + 0.17 \frac{b_{b} (λ)}{a (λ) + b_{b} (λ)}) \frac{b_{b} (λ)}{a (λ) + b_{b} (λ)}

The absorption (

a (λ)

) coefficients can be expressed as the sum of their individual components as

a (λ) = a_{w} (λ) + a_{p h} (λ) + a_{d g} (λ)

where

a_{w} (λ)

is the absorption coefficient of water, obtained from [31],

a_{p h} (λ)

is the absorption coefficients of phytoplankton, and

a_{d g} (λ)

is the absorption coefficient which includes both CDOM (also known as gelbstoff) and detritus since they have similar spectral dependence. In the WFS area

a_{d g} (λ)

is mostly contributed from CDOM (over 90%) [5] and we refer to it below as the CDOM absorption.

The relationship between phytoplankton $a_{p h} (λ)$ and chlorophyll [Chl] can be expressed by a power function

a_{p h} (λ) = A (λ) C h l^{B (λ)}

Here, values for

A (λ)

and

B (λ)

were taken from [5] for K. brevis bloom water containing greater than

10^{4} c e l l / l

(K. brevis bloom) and non-K. brevis bloom water containing less than

10^{4} c e l l / l

(non-K. brevis bloom).

Both CDOM and detritus absorb blue light strongly and absorption decreases with increasing wavelength as expressed by:

a_{d g} (λ) = a_{d g} (λ_{\circ}) \exp (- S_{d g} (λ - λ_{\circ}))

where

S_{d g}

is the slope of absorption spectrum, and the reference wavelength

λ_{\circ}

is 440nm. CDOM absorption at 440nm,

a_{d g} (440)

, was taken as ranging from 0 to 3

m^{- 1}

for our simulation. Based on the documented field measurement data, S is in the range of 0.1 to 0.2 nm⁻¹ with CDOM slope larger than that of detritus [5]. Here

S_{d g}

is taken as 0.014 nm⁻¹. Note here that

a_{d g} (440)

does not correlate with chlorophyll concentration, as is the case for most optically complex coastal waters [32].

Similarly, the backscattering coefficient can be expanded as

b_{b} (λ) = b_{b w} (λ) + b_{b p} (λ)

where

b_{b w} (λ)

is the backscatter coefficient of water obtained according to [33], and

b_{b p} (λ)

is the total particulate backscatter coefficient.

Particulate backscattering spectra can be fitted to the form

b_{b p} (λ) = b_{b p} (550) {(\frac{550}{λ})}^{γ}

where γ is the Angstrom exponent that describes the spectral shape and can be modeled according to [5] by a modified relationship from [34] as:

γ = 0.1 + \frac{1.9}{(1 + C h l)}

Particulate backscattering at the reference wavelength for K. brevis and non-K. brevis blooms based on the cell concentrations of K. brevis was modeled according to [5] as follows:

b_{b p} (550) = 0.0051 C h l^{0.180}

b_{b p} (550) = 0.0098 C h l^{0.977}

Non-K. brevis blooms modeled in accordance with [5], which will be referred hereafter as non-K. brevis-1, were found to give high reflectance values for

C h l

>

5 m g / m^{3}

. They are represented by the total particulate backscattering model in Eq. (9)b, which may be justified, since the non-K. brevis bloom model was developed from a data set which was mostly dominated by diatoms, and in general, diatom blooms are associated with upwelling and freshwater inputs from rivers which contain large amounts of sediments [35].

However, to further investigate our classification technique, we use a second method which is based on a standard bio-optical model [32] to compute the particulate backscattering of non-K. brevis blooms while keeping the absorption term the same as for the non-K. brevis-1. This second category will be referred hereafter as the non-K. brevis-2. To model the non-K. brevis-2 blooms, we separated the particulate backscattering term $b_{b p} (λ)$ in Eq. (6) as follows:

b_{b p} (λ) = {\tilde{b}}_{b, n a p} * b_{n a p} (λ) + {\tilde{b}}_{b, p h} * b_{p h} (λ)

where

b_{n a p} (λ)

is the scattering coefficient of non-algal particles (NAP),

{\tilde{b}}_{b, n a p}

is the backscattering ratio of NAP,

b_{p h} (λ)

is the scattering coefficient of phytoplankton, and

{\tilde{b}}_{b, p h}

is the backscattering ratio of phytoplankton .

The NAP scattering is modeled by a power law function as follows:

b_{n a p} (λ) = b_{n a p} (550) {(\frac{550}{λ})}^{γ_{2}}

where

b_{n a p} (550) = b_{n a p}^{*} (550) \cdot C_{n a p}

The specific scattering of NAP at 550nm, $b_{n a p}^{*} (550)$ , was assumed to be $0.5 m^{2} / g$ , and $γ_{2} = 1$ . $C_{n a p}$ represents the concentration of NAP in $m g / l$ . A constant backscattering ratio ${\tilde{b}}_{b, n a p} = 2 %$ was assumed over the whole spectral band for the NAP [32]. We assumed a NAP concentration of $2 m g / l$ as typically co-existing with the non-K. brevis blooms. This should be a reasonably conservative assumption for NAP concentrations associated with diatom blooms (with increasing concentrations of NAP the separation between K. brevis blooms and non-K. brevis blooms becomes even more defined) [35].

Both high spectral resolution field measurements and simulations, which take into account the imaginary part of refractive index of algal cells, have shown that the phytoplankton attenuation decreases gradually with wavelength while its scattering spectrum is affected by absorption features due to anomalous dispersion [16,36]. Following Lee [32], phytoplankton scattering is modeled as:

b_{p h} (λ) = c_{p h} (λ) - a_{p h} (λ)

where the attenuation spectrum

c_{p h} (λ)

can be modeled as power law function [16,37]:

c_{p h} (λ) = c_{p h} (550) {(\frac{550}{λ})}^{γ_{3}}

where

c_{p h} (550) = 0.3 {[c h l]}^{{0.62}_{}}

Here,

γ_{3} = 1

and the backscattering ratio

{\tilde{b}}_{b, p h} = 1 %

were assumed constant over the whole spectral band for phytoplankton [32].

The total amount of fluorescence radiance just above the surface can be modeled according to [38] as follows:

L F = 0.54 \cdot \frac{1}{4 π} \cdot ϕ \cdot \int_{400}^{700} \frac{a_{p h} (λ) \cdot E (λ, 0)}{K (λ) + a_{f}} \cdot d λ

where the 0.54 factor accounts for the propagation of the fluorescence through the air-water interface, the

\frac{1}{4 π} (s r^{- 1})

factor converts an isotropic fluorescence field to radiance, ϕ is the effective quantum yield of chlorophyll fluorescence which takes reabsorption into account [38] and was assumed to be 0.75%. This value falls on the upper side of the range determined for coastal waters [39].

E (λ, 0)

is the downwelling scalar irradiance just below surface, and was taken from [40],

K (λ)

is the attenuation coefficient in the fluorescence excitation zone, and

a_{f}

is the attenuation coefficient in the fluorescence emission zone. The fluorescence spectrum is modeled as

L_{f} (λ) = L F \cdot G (λ)

where

G (λ)

is Gaussian function centered at 685nm with a 25nm width. The absorption value at 685nm

a (685)

, was used to approximate the

a_{f}

term in the calculation. To estimate the diffuse attenuation coefficient, we use [41] to obtain:

K (λ) = 1.0547 \cdot \frac{a (λ) + b_{b} (λ)}{\cos (θ_{s})}

where

θ_{s}

is the solar zenith angle.

L_{f} (λ)

is added to the elastic remote sensing reflectance obtained from Eq. (1) after normalizing with respect to the surface downwelling irradiance

E_{d} (λ)

.

From the above microphysical model, the construction of simulated remote sensing reflectances, $R_{r s} (λ)$ , including both elastic and inelastic chlorophyll fluorescence components, can be performed. The first two simulated data sets were created for K. brevis cell concentrations, in accordance with [5], using the relationship for absorption of phytoplankton and particulate backscattering developed for K. brevis and non-K. brevis bloom waters. We also created a third data set for non-K. brevis blooms using the non-K. brevis-2 model, where we separated backscattering of NAP and algal particles, and assumed NAP concentrations of $2 m g / l$ to be always co-existing with the non-K. brevis blooms. We also created a fourth data set for Case-1 (Open Ocean) waters in accordance with [42], in which $C h l$ is the only input parameter:

b_{b p} (λ) = [0.002 + 0.02 (0.5 - 0.25 \log_{10} C h l) \frac{550}{λ}] \cdot 0.3 C h l^{0.62}

a_{p h} (λ) = 0.06 C h l^{0.65}

a_{d g} (λ) = 0.2 [a_{w} (440) + 0.06 C h l^{0.65}] \exp (- 0.014 (λ - 440))

Since these waters normally have a pronounced fluorescence component, it is important to distinguish them from waters with K. brevis blooms. Finally, the $n L w (λ)$ was derived by multiplying the $R_{r s} (λ)$ by the extraterrestrial solar constant in accordance with [43].

3.2 Background of Detection and Classification Algorithms

The essence of our approach is that the water-leaving radiance spectra of K. brevis and non-K. brevis blooms have distinctive features in the red region of the spectrum which can be used to detect and classify K. brevis blooms. The red spectral region is particularly attractive since it is less contaminated by CDOM and bottom reflectance, and is less susceptible to atmospheric correction difficulties than the blue-green region. As a consequence, uncertainties in bloom detection algorithms are reduced if this spectral region is used instead of the blue-green region. The distinguishing optical features of K. brevis and non-K. brevis blooms are demonstrated in Fig. 1 . The simulated elastic reflectance, $R_{r s} (λ)$ without chlorophyll fluorescence (green spectra) shows a trough around 675nm due to the absorption of chlorophyll for both types of blooms. When a fluorescence signal is included in the simulation (red spectra), the trough of the K. brevis bloom shifts toward shorter wavelengths around 667nm, or less, depending on chlorophyll concentrations and its quantum yield while the trough of non-K. brevis bloom remains around 675nm. The shift in the K. brevis spectra is due to the fact that K. brevis exhibit lower backscattering efficiency, so the fluorescence signal dominates the red reflectance spectral region. Because of the overlap of the phytoplankton absorption and chlorophyll fluorescence emission, when fluorescence is a significant portion of the reflectance signal, the trough in the red region shifts towards shorter wavelengths, which is the case for K. brevis. On the other hand, non-K. brevis blooms (mostly dominated by diatoms) have higher backscattering efficiency, so reflection is dominated by the elastic backscattering component, therefore the fluorescence signal represents only a very small portion of the total reflectance, and is too weak a contributor (compared to the backscatter signal) to result in any significant spectral changes. As a consequence, the trough of non-K. brevis blooms reflectance remains around the maximum of the phytoplankton absorption spectra (Fig. 1).

Fig. 1 Modeled remote sensing reflectance spectra for K. brevis cell concentrations (a) greater than $10^{4} c e l l s / l$ (K. brevis bloom) and (b) less than $10^{4} c e l l s / l$ (non-K. brevis-1 bloom) for the $C h l = 3 m g / m^{3}$ and $a_{d g} (440) = 0.25 m^{- 1}$ . The solid green spectra are when chlorophyll fluorescence is excluded (“F OFF”) from the simulation and solid red spectra are when fluorescence is included (“F ON”) in the simulation assuming 0.75% quantum yield. Band 13 and 14 are MODIS bands centered at 667nm and 678nm respectively.

Download Full Size | PDF

3.3 Detection Algorithm

Based on the above observations showing that the minimum of $R_{r s} (λ)$ can shift from the phytoplankton absorption maximum, around 678nm, to shorter wavelengths, around 667nm, with significant chlorophyll fluorescence contributions in the red spectral region, we can define a bloom detection technique which we identify simply as the Red Band Difference (RBD) as follows:

R B D = n L w (678) - n L w (667)

Simulation shows that the positive RBD values (> $1 m g / m^{3}$ of Chlorophyll) are primarily due to the fluorescence signal which correlates strongly with the chlorophyll concentration of the K. brevis. Because of this strong correlation, it may be possible to quantify K. brevis and other blooms with similar characteristics in terms of the chlorophyll concentrations more accurately than the standard reflectance band ratio algorithms [18–20] by developing some empirical relationship between the RBD and the bloom (K. brevis and other low backscattering blooms) chlorophyll using in situ data.

Since the RBD technique may also be able to detect blooms of other species, particularly the low backscattering ones, we further propose, below, an additional K. brevis bloom classification technique, KBBI, to discriminate K. brevis blooms from other blooms and bloom like features such as CDOM plumes, sediment plumes and bottom reflectance.

3.4 Classification Algorithm

We define the K. brevis bloom index (KBBI) as follows:

K B B I = \frac{n L w (678) - n L w (667)}{n L w (678) + n L w (667)}

The KBBI technique is primarily based on the fact that total particulate backscattering associated with K. brevis is different from that for non-K. brevis blooms. Since K. brevis bloom water is known to have lower total particulate backscattering [5–8] than the non-K. brevis bloom waters, the water-leaving radiance signal is much weaker for K. brevis blooms than for the non-K. brevis blooms since it is largely proportional to backscattering. As a consequence, the denominator of Eq. (20), which is just the sum of the two MODIS red bands (band 13 and band 14), becomes much larger for non-K. brevis blooms than for K. brevis blooms. Furthermore, the numerator of Eq. (20), which is the RBD, is much more pronounced for K. brevis blooms than the non-K. brevis blooms. Therefore, the KBBI values for K. brevis are usually higher than that of non-K. brevis, thus permitting the separation.

4. Results

4.1 Detection of K. brevis Blooms

Using the RBD detection technique, we detected various K. brevis blooms in the Gulf of Mexico. Figure 2 demonstrates two examples of K. brevis bloom detection on the WFS. The RBD image in Fig. 2a is created for MODIS (Terra) image 17 Sep 2001 [44,48] while the RBD image in Fig. 2b is created for MODIS (Aqua) image 21 Jan 2005 [45]. At the same time their corresponding KBBI images are displayed in Fig. 2c,d and will be discussed further below. To further investigate the potential of the RBD technique, the K. brevis bloom that took place between October and December 2004 on WFS, was traced over several weeks period. According to [21] the bloom by mid-late November contained high concentrations ( $> 10^{5} c e l l s / l$ ) of K. brevis cells and caused higher mortalities of fish and dolphins. Figure 3 shows an example of bloom tracing where the bloom drifted southward and expanded to form a large curved patch around 25.5 °N 82.5 °W from early November to mid December, and in subsequent weeks the bloom moved further to the south and formed a continuous band parallel to the Florida Keys [21]. These results are found to match reasonably well with cell count data from in situ measurements, obtained from [46] which were overlaid on top of images in Figs. 2 and 3 with H (black) and L (magenta) representing $> 10^{6} c e l l s / l$ and $< 10^{5} c e l l s / l$ respectively.

Fig. 2 K. brevis blooms detected using the RBD technique on the WFS on (a) 17 Sep 2001, and (b) 21 Jan 2005. These blooms are classified as K. brevis blooms using the KBBI classification technique with appropriate thresholds applied on (c) 17 Sep 2001 and (d) 21 Jan 2005. The 17 Sep 2001 image is an example when K. brevis and Trichodesmium blooms were co-occurring spatially but only K. brevis bloom is detected.

Download Full Size | PDF

Fig. 3 MODIS RBD image series show the progression of a K. brevis bloom. Land and cloud pixels are shown in white. The bright pixels next to the cloud are noise from clouds.

Download Full Size | PDF

The RBD images created from MODIS data often shows striping at the end of each scan line due to the variations in detector response. Also significant noise was noticed at the cloud edge pixels due to scattering from clouds affecting nearby pixels. However, both the striping and the noise at the cloud edge (which are retained in all figures) have significantly different appearances from bloom patches and can be easily distinguished in observation. Our analysis of the satellite data shows that using a threshold of RBD> $0.15 W / m^{2} / μ m / s r$ readily identifies legitimate bloom areas.

The modeled RBD values as a function of $C h l$ for various types of blooms with $a_{d g} (440)$ fixed at $0.25 m^{- 1}$ are demonstrated in Fig. 4a . This simulation shows that to reach the RBD threshold values stated above, may require chlorophyll as high as $5 m g / m^{3}$ with the assumed 0.75% fluorescence quantum yield. By using a $C h l > 1 m g / m^{3}$ to define a blooming condition [21], the simulation results show that an RBD threshold value lower than $0.15 W / m^{2} / μ m / s r$ is possible. However this lower threshold will increase false positives from satellite data analysis. Thus a compromise between simulations and satellite data analysis was made to arrive at the $0.15 W / m^{2} / μ m / s r$ RBD threshold.

Fig. 4 Modeled (a) RBD ( $W / m^{2} / μ m / s r$ ) and (b) KBBI values as a function of chlorophyll concentrations generated for K. brevis blooms (magenta), non-K. brevis-1 (green), non-K. brevis-2 (purple), and Case-1 (cyan) waters. $a_{d g} (440) = 0.25 m^{- 1}$

Download Full Size | PDF

The distinction between K. brevis and the non-K. brevis-1 can be clearly seen in Fig. 4a. However the K. brevis and Case-1 waters both give nearly the same RBD responses for $C h l$ ranging from 1 to 100 $m g / m^{3}$ . Even the non-K. brevis-2 model gives similar RBD values especially when $C h l$ goes beyond 20 $m g / m^{3}$ . The high backscattering level for non- K. brevis-1 leads to a dominant elastic signal in the reflectance at the red spectral region with a minima around 675nm, even at high $C h l$ . As a result the reflectance signal in the 678nm is lower than that at 667nm, which results in negative RBD values. On the other hand, the backscattering level for non-K. brevis-2 and Case-1 waters is not high enough to be distinguished from the K. brevis cases. Therefore the RBD itself is not sufficient to identify whether a bloom is produced by K. brevis or other species although it may discriminate between K. brevis and non-K. brevis-1 bloom waters (assuming backscattering relationship developed by [5] represents true non-K. brevis blooms backscattering and true fluorescence quantum yield is close to the assumed value). It is also reasonable to expect that the RBD technique maybe used to detect blooms of other species as well that have somewhat similar optical characteristics as the K. brevis blooms.

As discussed and shown in Fig. 4a, the RBD technique alone cannot differentiate between K. brevis bloom waters and Case-1 waters with high concentration of chlorophyll (even though it might be argued that it is very unlikely to have high concentrations of chlorophyll (> $5 m g / m^{3}$ ) in Case-1 waters unless if there is some sort of algal bloom present). To deal with the ambiguity that the RBD technique alone is not sufficient to distinguish K. brevis blooms, application of the KBBI technique is proposed for selective classification of K. brevis blooms, and is discussed next.

4.2 Classification of K. brevis Blooms using KBBI

In Fig. 4b, KBBI values are plotted as a function of $C h l$ . As can be seen, there are higher KBBI values for K. brevis blooms than for other species with similar $C h l$ , which opens up possibilities for distinguishing between K. brevis and other species. Therefore, to achieve effective separation of the K. brevis blooms from other blooms, we need to combine the RBD and KBBI techniques together. This approach is discussed below for a variety of coastal water conditions.

Since coastal waters are often impacted by CDOM absorption (causing algorithms using simple blue-green reflectance ratios to perform poorly), it is important that we explore the sensitivity of our proposed indices to CDOM impacts and other water properties typically expected in turbid waters. To explore these issues, the simulation was performed for two extreme values: $a_{d g} (440)$ = 0 and $3 m^{- 1}$ respectively, while keeping all other parameters constant as shown in Fig. 5 . Although these extreme values may not reflect conditions that naturally occur on the WFS, they at least provide a general idea of the boundaries for RBD and KBBI values due to CDOM absorption, which suffices for our detection and classification purpose. Increasing concentrations of CDOM in the water column are seen to result in a reduction in fluorescence signal. This cause the RBD and KBBI values to decrease slightly, given the prominence of chlorophyll fluorescence in these techniques. Since CDOM is assumed to have zero scatter, this decrease can be attributed to the absorption effects of CDOM, reducing the availability of photons in the chlorophyll fluorescence excitation waveband, and the consequent reduction of the fluorescence signal. The discrimination of K. brevis blooms from other blooms can also be observed in Fig. 5 where slopes for K.Brevis lines are greater than that for all other cases under consideration. The slopes are lower at $a_{d g} (440) = 0$ (Fig. 5: 1a & 2a) than that at $3 m^{- 1}$ (Fig. 5: 1b & 2b). This is due to the normalization factor in the KBBI technique, which increases with the decrease of $a_{d g}$ .

Fig. 5 Relationship between modeled RBD ( $W / m^{2} / μ m / s r$ ) and KBBI values for K. brevis blooms with $a_{d g} (440) = 0 m^{- 1}$ (red, “1a”) and $a_{d g} (440) = 3 m^{- 1}$ (red, “1b”), non-K. brevis-2 with $a_{d g} (440) = 0 m^{- 1}$ (blue, “2a”) and $a_{d g} (440) = 3 m^{- 1}$ (blue, “2b”), and Case-1 (cyan, “3”) waters. Dotted points represent MODIS (Aqua) data corrected using the NIR atmospheric correction algorithm and collected on 13 Nov 2004 (25.5°N – 25..9°N, 81.9°W - 82.3°W) (cyan) and 18 Nov 2004 (25.39°N - 25.5°N, 81.683°W - 82.266°W) (yellow). Also shown is a RBD threshold line (magenta) equal to $0.15 W / m^{2} / μ m / s r$ and the relationship KBBI = 0.3*RBD (magenta). The satellite data includes K. brevis bloomed water and also closed by waters which may contain K. brevis cells but in low concentrations.

Download Full Size | PDF

To further examine the efficacy of the classification technique over a wide range of water compositions, we used non-K. brevis-2 simulated data along with the case-1 simulated data as tests for separating K. brevis blooms from other types of water and blooms. The dotted points in Fig. 5 are taken from MODIS Aqua sensor images which contains data from the regions between (25.5°N - 25.9°N) and (81.9°W - 82.3°W) for 13 Nov 2004 (cyan) image and (25.39°N - 25.5°N) and (81.683°W - 82.266°W) for the 18 Nov 2004 (yellow) image. These regions included K. brevis blooms as well as neighboring pixels which may have contained K. brevis cells, but in low concentrations. The K. brevis values obtained from satellite data in Fig. 5 show slightly higher KBBI values than those from the simulated K. brevis data. This is probably due to the uncertainties caused by data retrieval using standard NIR atmospheric correction algorithm [47].

After analyzing simulation data coupled with the satellite data, we obtained the following thresholds for K. brevis bloom classification.

(1) RBD > 0.15 W/m²/µm/sr, and
(2) KBBI > 0.3*RBD

However, since these thresholds were derived assuming a chlorophyll fluorescence quantum yield of 0.75% and only the central waveband of each channel was used rather than the weighted average of the bandwidth, these values need to be further refined with in situ data.

To broaden the examination, twenty cloud free K. brevis bloom images from seven different bloom events (5 from WFS and 1 from the Texas Coast) were taken from literature dating between 2001 and 2006. The results are summarized in Table 1 . All twenty RBD images detect the bloom locations documented in the literature at least as well as other techniques such as chlorophyll anomaly and FLH. Furthermore in several occasions RBD removes false positive given by FLH imagery due to high scattering of NAP. Three of the KBBI images without thresholds gave false positives in the areas where RBD didn’t detect any bloom. This is probably due to the contamination from thin clouds or atmospheric uncertainties. However, when proposed thresholds were applied to create K. brevis classification images, those false positives disappeared.

Table 1. Studied MODIS bloom images. Detected and classified blooms with RBD and KBBI are shown with “√” while the symbol “√/×” under KBBI represents images where RBD and KBBI agreed in the bloomed area but KBBI gave false positives close to the coast where RBD didn’t detect any bloom. The Chlorophyll Anomaly [17] method is referred as ChlA.

View Table

Figure 2c & 2d are for the same conditions as shown in Fig. 2a & 2b respectively, and apply the proposed threshold conditions for both RBD and KBBI. Figure 2a & 2c show an example where K. brevis and Trichodesmium blooms were both known to co-exist (identified by red arrows in 2a) on the WFS [44]. As can be seen in Fig. 2c, the proposed K. brevis classification technique only extracts the K. brevis bloomed area and ignores the Trichodesmium bloomed area. This distinction is primarily due to the relative backscattering characteristic of K. brevis and Trichodesmium blooms. Trichodesmium blooms are well known to backscatter strongly [54]. Whenever backscattering is strong, the KBBI values are negative or very close to zero, so these blooms are not detected.

4.3 Comparison between RBD and FLH

Although FLH can sometimes be used to detect blooms [21], it breaks down in highly scattering waters, where high red peak values are primarily due to contributions from elastic scattering modulated by chlorophyll absorption rather than the fluorescence, thus falsely indicating possible blooms. Thus, as the concentration of NAP increases, radiance generally raises as well, and the fluorescence peak becomes a less prominent component of increased total signal.

In contrast, the RBD technique is found to easily differentiate between the two effects, giving positive values in truly bloomed waters and negative values in highly scattering waters. The performance of MODIS FLH calculations can be assessed for turbid waters by comparing FLH values with true fluorescence values at 685nm. Simulations show that the true fluorescence signal at 685 nm decreases as a fraction of the total signal for increasing concentration of NAP. However, MODIS FLH shows an opposite trend and a significant overestimation of the true fluorescence signal. The most dramatic effect is the overestimation of true fluorescence when chlorophyll is low and NAP increases. Similar results were found in [23,24,55]. Tomlinson, et al., (2008) [56] also pointed out that FLH was unreliable in the area surrounding the Florida Keys; giving elevated values in all of the images they examined which is in agreement with our analysis of satellite imagery for this region. Clearly, the FLH algorithm can breakdown for turbid waters and if used for K. brevis bloom detection would also flag pixels with turbid waters as possible K. brevis blooms. On the other hand, the RBD technique only detects true blooms and gives negative or near zero values for highly scattering waters, so does the KBBI classification technique. This is more clearly illustrated in Fig. 6 , which shows MODIS Aqua bloom images from November 13, 2004 on the WFS. To examine whether the intense regions in the FLH image are due to blooms or highly scattering waters, we took three spectra from the three supposedly bloomed regions (bloomed, turbid-1, and turbid-2) as indicated by FLH (Fig. 6a) and plotted the resultant spectra in Fig. 6c. We see that the true K. brevis bloom spectrum (Fig. 6c; red) differs significantly from the other two spectra particularly in the blue-green region of the optical spectrum where they both give significantly higher values than the bloomed spectra. So the spectra (Fig. 6c; green and blue) taken from turbid-1 and turbid-2 region of Fig. 6a are due to highly scattering waters, and not characteristic of K. brevis. Furthermore, it is seen that the signals in the red bands of these spectra are also significantly different where the K. brevis bloom spectra (Fig. 6c; red) has a positive slope from 667nm band to 678nm band while other two have negative slopes. The slope is negative only when water is highly scattering. Therefore, the region in Fig. 6a indicated by turbid-1 and turbid-2 must be due to highly scattering waters and not K. brevis blooms. Those false blooms signals based on FLH clearly disappear in the RBD image (Fig. 6b).

Fig. 6 MODIS (Aqua) bloom image from 13 November 2004 for the WFS (a) FLH ( $W / m^{2} / μ m / s r$ ) image, (b) RBD ( $W / m^{2} / μ m / s r$ ) image and (c) Normalized-water leaving radiance spectra taken from the bloomed and turbid waters indicated by “circle” and “squares” respectively in the FLH image.

Download Full Size | PDF

4.4 Atmospheric Corrections by NIR and SWIR and Impacts on RBD and KBBI

In coastal waters, the standard NIR atmospheric correction [27] often fails due to higher turbidity and consequently significant higher radiance contributions in the NIR bands. Since the water-leaving radiance at NIR can no longer be considered negligible for the use of atmospheric correction for extreme blooms [47,57], negative readings may result in the blue-green bands due to the over-correction of the atmosphere [58]. More recently an atmospheric correction scheme using the short-wave infrared (SWIR) bands has been proposed [28] for turbid coastal waters, and has become part of the operational SeaDAS environment. Clearly, the less a bloom retrieval index is sensitive to atmospheric correction uncertainties the better, so we next examine performance of our RBD and KBBI techniques for different atmospheric correction algorithms.

Figure 7a shows two normalized water leaving radiance spectra averaged over 3 by 3 pixels taken from the same MODIS (Aqua) image of 13 Nov 2004 from the area around (25.73°N and 82.11°W) but atmospherically corrected using standard NIR and SWIR algorithms. The negative signal at 412nm band, shown with an “orange, circle” in Fig. 7a, is an indication of the atmospheric correction failure, meaning that the signal is overcorrected for the atmosphere which is often the case in bloomed and coastal waters. Although SWIR seems to do a little bit better than NIR in the bloomed region (Fig. 7a), they both give a negative signal at the blue band edges, an indication of the limitations of both algorithms, with errors in the atmospheric correction schemes clearly larger for shorter wavelengths. Thus, retrievals using blue-green band ratio algorithms with the imperfect atmospheric corrections will result in different and inaccurate results for different atmospheric correction algorithms.

Fig. 7 (a) Normalized-water leaving radiance spectra of K. brevis bloom taken from the MODIS (Aqua) image of 13 Nov 2004 (25.73°N and 82.11°W) using different atmospheric correction: NIR and SWIR algorithms. The orange circled area indicates the atmospheric correction failure and the red circled area indicates the fluorescence peak. (b) RBD value with SWIR vs. RBD with NIR (c) KBBI value with SWIR vs. KBBI with NIR. The data are from the same image in (a) containing the region between (25.9°N - 25.5°N) and (81.9°W - 82.3°W).

Download Full Size | PDF

Figure 7b demonstrates the relative insensitivity of the RBD technique, which is focused on a small spectral range near the red chlorophyll fluorescence spectrum, on NIR and SWIR atmospheric correction algorithms. The data for Fig. 7b are taken from MODIS Aqua sensor image containing data from the regions between (25.9°N - 25.5°N) and (81.9°W - 82.3°W) on 13 Nov 2004. This region includes K. brevis bloomed areas as well as neighboring pixels which may contain K. brevis cells but in low concentrations. Our analysis shows that the RBD values are nearly the same with either atmospheric correction algorithm (Fig. 7b) for the selected regions. This is not surprising, since with RBD we are calculating the difference between the two bands which doesn’t change if the spectrum is shifted up or down by different atmospheric correction algorithms, as oppose to spectral band ratios which change significantly even with a small shift in the absolute magnitudes of the bands.

Sensitivity of the K. brevis classification technique KBBI on NIR and SWIR atmospheric correction is demonstrated Fig. 7c. The data are taken from the same region as the RBD data in Fig. 7b. Since KBBI is a ratio, the correlation between NIR and SWIR KBBI data is somewhat reduced mainly because of the normalization while the RBD which retains its strong correlations. Thus, although the numerator (same as the RBD) remains nearly the same, the sum of the two red bands (denominator of the KBBI) changes when the spectrum is shifted up or down with different atmospheric correction algorithms. Because of the changes in the denominator of KBBI with the atmospheric correction algorithms, the KBBI values changes somewhat, but still give reasonable enough correlations.

5. Discussion

Chlorophyll retrieval from reflectance spectra remains a challenge in coastal waters, and may even be an impossible task in some cases, particularly when CDOM concentrations are high [59]. MODIS red bands 13 and 14 were designed with high signal-to-noise ratios to avoid various problems including CDOM for retrievals of fluorescence, and hence chlorophyll. However, in highly scattering waters, typical in coastal areas, the fluorescence component represents a small portion of the total reflectance signal, and FLH algorithms erroneously extract a combined elastic scattering and fluorescence signal, rather than the much smaller fluorescence signal.

When low concentrations of NAP and low scattering conditions exist, the red spectral region becomes chlorophyll fluorescence dominated. Furthermore, this spectral region is less affected by CDOM, shallow bottom, and even atmospheric correction uncertainties than the blue-green region. Since the K. brevis bloom is known to have a much lower backscattering efficiency, the red signals from K. brevis blooms are usually largely dominated by the K. brevis chlorophyll fluorescence. The RBD technique takes advantage of this fluorescence dominated signal to detect these types of blooms. Normalization then makes it possible for the KBBI technique to discriminate K. brevis from non-K. brevis blooms. When applied to existing satellite imagery, we find that applying the thresholds proposed in section 4.2, the K. brevis blooms were detected and classified successfully over a wide time span in the WFS. This included the bloom event of 2004 on the WFS which was successfully tracked for several weeks.

The techniques developed in this study for detecting and classifying K. brevis blooms using satellite ocean color measurements appear to have advantages over other techniques developed using blue-green bands [5,17] and even red-NIR bands such as FLH. Recently it has been suggested in [56] that future studies should examine the use of a FLH anomaly as a potential replacement to the chlorophyll anomaly method. However, it can be seen from our work here and previously [22–24] that any significant NAP scattering gives red peaks where the main contribution is from elastic reflection modulated by the confluence of chlorophyll and water absorption rather than from fluorescence, which results in false identification of fluorescence peaks in NAP rich waters. In contrast, the RBD technique eliminates association of an elastic peak with a bloom, by giving a negative value when these circumstances occur. Both RBD and KBBI techniques are much less affected by uncertainties in atmospheric correction algorithms. This is particularly true for the RBD detection technique because it uses band differences of two spectrally close lying bands, for which the atmospheric impact is little changed, and more importantly because differences rather than ratios are used. Thus analysis shows that the detection technique gives nearly the same result with different atmospheric correction algorithms. Classification of K. brevis bloom from Trichodesmium bloom is also possible if K. brevis blooms spatially co-exist with high-backscattering Trichodesmium blooms [44,54] since KBBI gives negative or nearly zero values for highly backscattering blooms. Furthermore, Hu et al., (2005) [21] suggested that using FLH data along with knowledge of local water and ERGB composite imagery, an operator may make it possible to identify whether a feature is shallow bottom, resuspended sediment, a phytoplankton bloom, or a CDOM-rich plume. However, the RBD and KBBI techniques seem to be capable, possibly with some additional tuning, of eliminating or at least minimizing most of these issues.

We also would point out that the RBD or the KBBI imagery generated from MODIS often shows striping due to the variation in the response among detectors and significant noise near the cloud edges which is consistence with the previous studies of FLH [21]. However, if all threshold conditions are used simultaneously, cloud edge noise or striping effects should not be a problem. Furthermore, cloud edge noise can be flagged out by flagging out cloud edge pixels and the striping is very easily distinguishable using spatial pattern techniques and can be flagged out as well. Additional errors in the classification technique may be introduced due to normalization in highly absorbing waters, or from inappropriate atmospheric correction algorithms, which may affect the KBBI index somewhat. However, since the RBD indicator used at the same time would not give false signal in such waters, these problems can be avoided by checking agreement between the RBD and KBBI imagery or using the proper thresholds.

6. Conclusion

We have introduced two red band indices RBD and KBBI which appear to be effective for assisting K. brevis bloom detection and classification respectively. These techniques take advantage of the low backscattering characteristics of K. brevis which allows RBD values to increase with increasing chlorophyll concentrations due to chlorophyll fluorescence, while at the same time, KBBI values also increase with increasing chlorophyll concentration for K. brevis blooms but remains relatively small for non-K. brevis blooms.

The thresholds for bloom detection and K. brevis bloom classification were derived using simulated data coupled with the satellite data of known bloom events. Both techniques were successfully applied to MODIS data to detect, monitor and classify K. brevis blooms in the Gulf of Mexico to be efficient if used together. Different K. brevis bloom events documented in the literature were successfully detected and in addition, the K. brevis bloom event that took place in late 2004 on the WFS was traced for several weeks using the proposed bloom detection technique.

Our results provide a way for assisting the detection of toxic dinoflagellate K. brevis blooms and differentiating them from other blooms or bloom like features such as CDOM-rich plumes using satellite ocean color measurements. In particular, we show that these indicators improve on traditional algorithms such as FLH to correctly identify the potential bloomed areas and to distinguish K. brevis blooms from other blooms, plumes, sediments, and even shallow bottom reflectance. Furthermore, we also show that our bloom detection technique is largely insensitive to atmospheric correction algorithms, unlike the traditional band ratio algorithms, such as standard blue-green chlorophyll retrieval algorithms. The KBBI classification technique is also found to perform reasonably well with different atmospheric correction algorithms for the bloomed regions. While the general thresholds developed provide a reasonably robust set of indicators, more extensive observations are needed to fine tune these and explore potential difficulties.

Acknowledgments

This research has been supported by grants from NOAA, NASA and the Office of Naval Research. We are very grateful to the anonymous reviewers whose multiple comments and suggestions helped to improve this paper significantly.

References and links

1. http://www.floridaconservation.org/

2. J. H. Landsberg, and K. A. Steidinger, “A historical review of Gymnodium breve red tides implicated in mass mortalities of the manatee (Trichechus manatus latiriostris) in Florida, USA,” In: Reguera, B., Blanco, J., Fernandez, M.L., Wyatt, T. (Eds.), Proceedings of the 8th International Conference on Harmful Algal Blooms, Vigo, Spain, pp. 97–100, (1998)

3. W. H. Hemmert, “The public health implications of Gymnodinium breve red tides, A review of the literature and recent events” In: LoCicero, V.R. (Eds.), Proceedings of the First International Conference on Toxic Dinoflagellate Blooms, pp. 489–497, (1975)

4. S. Asai, J. J. Krzanowski, W. H. Anderson, D. F. Martin, J. B. Polson, R. F. Lockey, S. C. Bukantz, and A. Szentivanyi, “Effects of the toxin of red tide, Ptychodiscus brevis, on cannie tracheal smooth muscle: a possible new asthma-triggering mechanism,” J. Allergy Clin. Immunol. 69(5), 418–428 (1982). [CrossRef] [PubMed]

5. J. P. Cannizzaro, K. L. Carder, F. R. Chen, C. A. Heil, and G. A. Vargo, “A novel technique for detection of the toxic dinoflagellate Karenia brevis in the Gulf of Mexico from remotely sensed ocean color data,” Continent. Shel. Res. 28(1), 137–158 (2008). [CrossRef]

6. K. L. Mahoney, “Backscattering of light by Karenia brevis and implications for optical detection and monitoring,” PhD dissertation, Univ. of South. Miss., Stennis Space Cent. (2003)

7. K. L. Carder and R. G. Steward, “A remote-sensing reflectance model of a red-tide dinoflagellate off west Florida,” Limnol. Oceanogr. 30, 286–298 (1985). [CrossRef]

8. O. Schofield, J. Kerfoot, K. Mahoney, M. Moline, M. Oliver, S. Lohrenz, and G. Kirkpatrick, “Vertical migration of the toxic dinoflagellate Karenia brevis and the impact on ocean optical properties,” J. Geophys. Res. 111(C6), C06009 (2006). [CrossRef]

9. G. J. Kirkpatrick, D. F. Millie, M. A. Moline, and O. Schofield, “Optical discrimination of a phytoplankton species in natural mixed populations,” Limnol. Oceanogr. 45, 467–4718 (2000). [CrossRef]

10. D. F. Millie, G. J. Kirkpatrick, and B. T. Vinyard, “Relating photosynthetic pigments and in vivo optical density spectra to irradiance for the Florida red-tide dinoflagellate Gymnodinium breve,” Mar. Ecol. Prog. Ser. 120, 65–75 (1995). [CrossRef]

11. D. F. Millie, O. M. Schofield, G. J. Kirkpatrick, G. Johnsen, P. A. Tester, and B. T. Vinyard, “Detection of harmful algal blooms using photopigments and absorption signatures: a case study of the Florida red tide dinoflagellate, Gymnodinium breve,” Limnol. Oceanogr. 42, 1240–1251 (1997). [CrossRef]

12. S. E. Lohrenz, G. L. Fahnenstiel, G. J. Kirkpatrick, C. L. Carroll, and K. A. Kelly, “Microphotometric assessment of spectral absorption and its potential application for characterization of harmful algal species,” J. Phycol. 35(6), 1438–1446 (1999). [CrossRef]

13. A. Morel and Y. Ahn, “Optics of heterotrophic nanoflagellates and ciliates: A tentative assessment of their scattering role in oceanic waters compared to those of bacteria and algal cells,” J. Mar. Res. 49(1), 177–202 (1991). [CrossRef]

14. D. Stramski and D. A. Kiefer, “Light scattering in microorganisms in the open ocean,” Prog. Oceanogr. 28(4), 343–383 (1991). [CrossRef]

15. O. Ulloa, S. Sathyendranath, and T. Platt, “Effect of the particle-size distribution on the backscattering ratio in seawater,” Appl. Opt. 33(30), 7070–7077 (1994). [CrossRef] [PubMed]

16. M. S. Twardowski, E. Boss, J. B. Macdonald, W. S. Pegau, A. H. Barnard, and J. R. V. Zaneveld, “A model for estimating bulk refractive index from the optical backscattering ratio and the implications for understanding particle composition in case I and case II waters,” J. Geophys. Res. 106(C7), 14129–14142 (2001). [CrossRef]

17. R. P. Stumpf, M. E. Culver, P. A. Tester, M. Tomlinson, G. J. Kirkpatrick, B. A. Pederson, E. Truby, V. Ransibrahmanakul, and M. Soracco, “Monitoring Karenia brevis blooms in the Gulf of Mexico using satellite ocean color imagery and other data,” Harmful Algae 2, 147–160 (2003). [CrossRef]

18. J. E. O’Reilly, S. Maritorena, D. Siegel, M. C. O’Brien, and D. Toole, B. G., Mitchell, et al., “Ocean color chlorophyll a algorithms for SeaWiFS, OC2, and OC4: version 4,” In: Hooker, S. B. & Firestone, E. R. (Eds.), SeaWiFS Postlaunch Calibration and Validation Analyses, Part 3 SeaWiFS Postlaunch Technical Report Series, (pp. 9–23) Greenbelt, Maryland: NASA, Goddard Space Flight Center (2000)

19. H. R. Gordon, D. K. Clark, J. W. Brown, O. B. Brown, R. H. Evans, and W. W. Broenkow, “Phytoplankton pigment concentrations in the Middle Atlantic Bight: comparison of ship determinations and CZCS estimates,” Appl. Opt. 22(1), 20–36 (1983). [CrossRef] [PubMed]

20. K. L. Carder, F. R. Chen, Z. P. Lee, S. K. Hawes, and D. Kamykowski, “Semi-analytic moderate-resolution imaging spectrometer algorithms for chlorophyll a and absorption with bio-optical domains based on nitrate-depletion temperatures,” J. Geophys. Res. 104(C3), 5403–5421 (1999). [CrossRef]

21. C. Hu, F. E. Muller-Karger, C. Taylor, K. L. Carder, C. Kelble, E. Johns, and C. A. Heil, “Red tide detection and tracing using MODIS fluorescence data: A regional example in SW Florida coastal waters,” Remote Sens. Environ. 97(3), 311–321 (2005). [CrossRef]

22. A. Gilerson, J. Zhou, S. Hlaing, I. Ioannou, R. Amin, B. Gross, F. Moshary, and S. Ahmed, “Fluorescence contribution to reflelctance spectra for a variety of coastal waters,” Proc. SPIE 6680, (2007). [CrossRef]

23. A. Gilerson, J. Zhou, S. Hlaing, I. Ioannou, J. Schalles, B. Gross, F. Moshary, and S. Ahmed, “Fluorescence component in the reflectance spectra from coastal waters. Dependence on water composition,” Opt. Express 15(24), 15702–15721 (2007). [CrossRef] [PubMed]

24. A. Gilerson, J. Zhou, S. Hlaing, I. Ioannou, B. Gross, F. Moshary, and S. Ahmed, “Fluorescence component in the reflectance spectra from coastal waters. II. Performance of retrieval algorithms,” Opt. Express 16(4), 2446–2460 (2008). [CrossRef] [PubMed]

25. R. Amin, J. Zhou, A. Gilerson, B. Gross, F. Moshary, and S. Ahmed, “ Detection of Karenia brevis harmful algal blooms in the West Florida Shelf using red bands of MERIS Imagery,” OCEANS 08 MTS/IEEE Quebec, Canada, 15–18 Sept. (2008)

26. http://oceancolor.gsfc.nasa.gov

27. H. R. Gordon and M. Wang, “Retrieval of water-leaving radiance and aerosol optical thickness over the oceans with SeaWiFS: a preliminary algorithm,” Appl. Opt. 33(3), 443–452 (1994). [CrossRef] [PubMed]

28. M. Wang and W. Shi, “Estimation of ocean contribution at MODIS near infrared wavelengths along the east coast of the U.S.: two case studies,” Geophys. Res. Lett. 32(13), L13606 (2005), doi:. [CrossRef]

29. Z. P. Lee, K. L. Carder, C. D. Mobley, R. G. Steward, and J. S. Patch, “Hyperspectral remote sensing for shallow waters. I. A semianalytical model,” Appl. Opt. 37(27), 6329–6338 (1998). [CrossRef]

30. Z. P. Lee, K. L. Carder, C. D. Mobley, R. G. Steward, and J. S. Patch, “Hyperspectral remote sensing for shallow waters. 2. Deriving bottom depths and water properties by optimization,” Appl. Opt. 38(18), 3831–3843 (1999). [CrossRef]

31. R. M. Pope and E. S. Fry, “Absorption spectrum (380-700 nm) of pure water. II. Integrating cavity measurements,” Appl. Opt. 36(33), 8710–8723 (1997). [CrossRef]

32. Z. P. Lee, http://www.ioccg.org/groups/OCAG_data.html

33. A. Morel, “Optical properties of pure water and pure seawater,” in Optical Aspects of Oceanography, N. G. Jerlov and E. S. Nielsen, eds., (Academic, New York, 1974)

34. Z. P. Lee and K. L. Carder, “Band-ratio or spectral-curvature algorithms for satellite remote sensing?” Appl. Opt. 39(24), 4377–4380 (2000). [CrossRef]

35. T. J. Smayda, “Harmful algal blooms: their ecophysology and general relevance to phytoplankton blooms in the sea,” Limnol. Oceanogr. 42, 1137–1153 (1997). [CrossRef]

36. D. Stramski, A. Bricaud, and A. Morel, “Modeling the inherent optical properties of the ocean based on the detailed composition of the planktonic community,” Appl. Opt. 40(18), 2929–2945 (2001). [CrossRef]

37. C. S. Roesler and E. Boss, “Spectral beam attenuation coefficient retrieved from ocean color inversion,” Geophys. Res. Lett. 30(9), 1468 (2003), doi:. [CrossRef]

38. Y. Huot, C. A. Brown, and J. J. Cullen, “New algorithms for MODIS sun-induced chlorophyll fluorescence and a comparison with present data products,” Limnol. Oceanogr. Methods 3, 108–130 (2005). [CrossRef]

39. J. Zhou, A. Gilerson, I. Ioannou, S. Hlaing, J. Schalles, B. Gross, F. Moshary, and S. Ahmed, “Retrieving quantum yield of sun-induced chlorophyll fluorescence near surface from hyperspectral in-situ measurement in productive water,” Opt. Express 16(22), 17468–17483 (2008). [CrossRef] [PubMed]

40. W. W. Gregg and K. L. Carder, “A simple spectral solar irradiance model for cloudless maritime atmospheres,” Limnol. Oceanogr. 35, 1657–1675 (1990). [CrossRef]

41. A. Albert and C. D. Mobely, “An analytical model for subsurface irradiance and remote sensing reflectance in deep and shallow case-2 waters,” Opt. Express 11, 2873–2890 (2003). [CrossRef] [PubMed]

42. A. Morel and S. Maritorena, “Bio-optical properties of ocean waters: A reappraisal,” J. Geophys. Res. 106(C4), 7163–7180 (2001). [CrossRef]

43. http://oceancolor.gsfc.nasa.gov/DOCS/MSL12/master_prodlist.html/#nLw

44. J. J. Walsh, J. K. Jolliff, B. P. Darrow, J. M. Lenes, S. P. Milroy, A. Remsen, D. A. Dieterle, K. L. Carder, F. R. Chen, G. A. Vargo, R. H. Weisberg, K. A. Fanning, F. E. Muller-Karger, E. Shinn, K. A. Steidinger, C. A. Heil, C. R. Tomas, J. S. Prospero, T. N. Lee, G. J. Kirkpatrick, T. E. Whitledge, D. A. Stockwell, T. A. Villareal, A. E. Jochens, and P. S. Bontempi, “Red tides in the Gulf of Mexico: where, when, and why?” J. Geophys. Res. 111(C11), C11003 (2006). [CrossRef]

45. K. L. Carder, J. P. Cannizzaro, F. R. Chen, and C. Hu, “ Detecting HAB’s in the Gulf of Mexico: Problems with atmospheric correction and shallow waters,” MODIS Science Team Meeting, (2005)

46. FWRI, http://www.floridamarine.org

47. R. Amin, A. Gilerson, J. Zhou, B. Gross, F. Moshary, and S. Ahmed, “ Impacts of atmospheric corrections on algal bloom detection techniques,” 89th AMS Annual Meeting, Phoenix, Arizona, 11–15 Jan, (2009)

48. M. C. Tomlinson, R. P. Stumpf, V. Ransibrahmanakul, E. W. Truby, G. J. Kirkpatrick, B. A. Pederson, G. A. Vargo, and C. A. Heil, “Evaluation of the use of SeaWiFS imagery for detecting Karenia brevis harmful algal blooms in the eastern Gulf of Mexico,” Remote Sens. Environ. 91(3-4), 293–303 (2004). [CrossRef]

49. http://www.nasa.gov/centers/goddard/news/topstory/2004/0826planktonglow.html

50. http://nasascience.nasa.gov/earth-science/applied-sciences/national-applications/coastal-managment

51. http://coastwatch.noaa.gov/hab/bulletins_ns.htm

52. C. Hu, F. E. Muller-Karger, and P. W. Swarzenski, “Hurricanes, submarine groundwater discharge, and Florida’s red tides,” Geophys. Res. Lett. 33(11), L11601 (2006), doi:. [CrossRef]

53. http://www.tpwd.state.tx.us/landwater/water/environconcerns/hab/redtide/status.phtml

54. A. Subramaniam, E. J. Carpenter, and P. G. Falkowski, “Bio-optical properties of the marine diazotrophic cyanobacteria Trichodesmium spp. II. A reflectance model for remote sensing,” Limnol. Oceanogr. 44, 618–627 (1999). [CrossRef]

55. D. McKee, A. Cunningham, D. Wright, and L. Hay, “Potential impacts of nonalgal materials on water-leaving Sun induced chlorophyll fluorescence signals in coastal waters,” Appl. Opt. 46(31), 7720–7729 (2007). [CrossRef] [PubMed]

56. M. C. Tomlinson, T. T. Wynne, and R. P. Stumpf, “An evaluation of remote sensing techniques for enhanced detection of the toxic dinoflagellate, Karenia brevis,” Remote Sens. Environ. (2008), doi:.

57. D. A. Siegel, M. Wang, S. Maritorena, and W. Robinson, “Atmospheric correction of satellite ocean color imagery: the black pixel assumption,” Appl. Opt. 39(21), 3582–3591 (2000). [CrossRef]

58. C. Hu, K. L. Carder, and F. E. Muller-Karger, “Atmospheric correction of SeaWiFS imagery over turbid coastal waters: a practical method,” Remote Sens. Environ. 74(2), 195–206 (2000). [CrossRef]

59. C. Hu, Z. P. Lee, F. E. Muller-Karger, and K. L. Carder, “Application of an optimization algorithm to satellite ocean color imagery: A case study in Southwest Florida coastal waters,” In J. Frouin, Y. Yuan, & H. Kawamura (Eds.) SPIE Proceedings. Ocean Remote Sensing and Applications, vol. 4892 (pp. 70–79). Bellingham, WA: SPIE, (2003)

Bloom events	Date	Regions in the Gulf of Mexico	RBD	KBBI	Other techniques	References
1	09/17/01	WFS	√	√	ChlA	[48,44]
2	02/04/02 02/17/02	WFS WFS	√ √	√ √	Enhanced RGB True-color	[49] [50]
3	10/09/03 10/19/03	WFS WFS	√ √	√/ × √	Enhanced RGB Enhanced RGB / FLH	[49] [49]
4	10/30/04 11/07/04 11/13/04 11/18/04 11/21/04 11/22/04 12/04/04 12/13/04	WFS WFS WFS WFS WFS WFS WFS WFS	√ √ √ √ √ √ √ √	√ √ √/ × √ √/ × √ √ √	FLH FLH ChlA / FLH ChlA / FLH FLH ChlA / FLH FLH ChlA / FLH	[21] [21] [51,21] [51,21] [21] [51,21] [21] [51,21]
5	01/21/05 02/08/05 02/09/05	WFS WFS WFS	√ √ √	√ √ √	ChlA / FLH FLH FLH	[51,45] [45] [45]
6	09/26/05 09/27/05 09/29/05	WFS WFS WFS	√ √ √	√ √ √	FLH FLH ChlA / FLH	[52] [52] [51,52]
7	10/06/06	Texas	√	√	True-color	[53]

Novel optical techniques for detecting and classifying toxic dinoflagellate Karenia brevis blooms using satellite imagery

Abstract

1. Introduction

2. Satellite Data and Image Processing

3. Development of Bloom Detection and K. brevis Bloom Classification Techniques

3.1 Optical Modeling of K. brevis and non-K. brevis Blooms

3.2 Background of Detection and Classification Algorithms

3.3 Detection Algorithm

3.4 Classification Algorithm

4. Results

4.1 Detection of K. brevis Blooms

4.2 Classification of K. brevis Blooms using KBBI

4.3 Comparison between RBD and FLH

4.4 Atmospheric Corrections by NIR and SWIR and Impacts on RBD and KBBI

5. Discussion

6. Conclusion

Acknowledgments

References and links

Cited By

Figures (7)

Tables (1)

Equations (25)

Optics Express