A hybrid approach to estimate chromophoric dissolved organic matter in turbid estuaries from satellite measurements: A case study for Tampa Bay

Chengfeng Le; Chuanmin Hu

doi:10.1364/OE.21.018849

1. Introduction

Chromophoric dissolved organic matter (CDOM) in the ocean, also called yellow substance or Gelbstoff, is an optically active component that plays an important role in marine aquatic ecosystems. CDOM absorbs light over a wide spectrum from the ultra-violet (UV) to the visible, thus modulating the underwater light field and growth rates of phytoplankton and other aquatic organisms [1]. In addition, CDOM is a component of the dissolved organic carbon (DOC) pool and plays an important role in carbon cycling [2–5]. In river-driven estuaries and coastal waters, CDOM often shows a strong relationship with salinity (an important water quality index) due to conservative mixing between fresh and oceanic waters [6–8]. Thus, it is desirable to know CDOM abundance, source, transport and transformation from local, regional, to global scales.

Traditionally, studies on CDOM were mainly based on shipboard measurements [3, 6, 8, 9]. Although shipboard measurements are generally accurate, their coverage is often limited in both space and time. On the other hand, CDOM and other water quality parameters in estuaries can change substantially in short periods of time (e.g. 1-2 days in Tampa Bay) due to tidal mixing, cold fronts, and summer storms [10–13]. These short-term variations might be missed by the infrequent field measurements, leading to potentially biased results in assessing mean and anomaly conditions. Satellite ocean color measurements may complement shipboard surveys through more synoptic and frequent observations [14]. However, although some preliminary success has been achieved in deriving several water quality parameters (e.g. water turbidity, light attenuation coefficient, Chlorophyll a concentration) from satellite measurements in turbid estuaries such as Tampa Bay and Chesapeake Bay [15–17], estimating CDOM distributions from space still remains a challenge primarily due to two reasons: uncertainties in the algorithm design, and uncertainties in the satellite-derived remote sensing reflectance (Rrs(λ), sr⁻¹, used as the algorithm input) due to imperfect atmospheric correction.

During the past few decades, several empirical and semi-analytical algorithms have been proposed for estimating CDOM from satellite measurements. Among the empirical approaches are LUT (look-up table)-based algorithms [18, 19], PCA (principal component analysis)-based algorithms [20, 21], and band-ratio algorithms [22–27]. Generally, empirical algorithms are difficult to extrapolate to other regions as the governing optical relationships between the various optically significant components may vary [28]. In contrast, semi-analytical algorithms address this problem by taking into account the underlying physics through governing radiative transfer equations [29–32]. These algorithms have been implemented by the U.S. NASA Ocean Biology Processing Group (OBPG) for processing global ocean color data, with data products generated and shared with the research community. However, these algorithms do not differentiate between CDOM and non-algal particulate (or detritus) absorption, and therefore may lead to high uncertainties in coastal and estuarine waters where CDOM and non-algal particle loads do not co-vary. Further, atmospheric correction errors often lead to high uncertainties in satellite-derived Rrs(λ) in the blue bands (412 and 443 nm) that are often utilized by semi-analytical algorithms for estimating CDOM and other water quality parameters (e.g., chlorophyll-a concentrations or Chla in mg m⁻³). As a result, similar to the empirical approaches, semi-analytical approach can also lead to high uncertainties in the CDOM retrievals in turbid estuaries.

Recently, several studies have attempted to separate CDOM absorption from other absorbing components (e.g., non-algal particles, phytoplankton pigments) by combining empirical relationships and total absorption derived from a quasi-analytical algorithm (QAA, see [32]) [33,34]. The empirical relationships, based on field measurements, are between non-algal particulate (detritus) absorption at 443 nm (a_d(443), m⁻¹) and backscattering coefficient at 555nm (b_b(555), m⁻¹), or between spectral shapes of CDOM absorption (a_g(λ)) and phytoplankton pigment absorption (a_ph(λ)) at 412, 443, and 490 nm. Although the case studies showed some success using mostly field data and limited satellite data, a potential caveat is the strong reliance these approaches have on Rrs(412) and Rrs(443), which may have substantial uncertainties in estuarine waters due to problems in atmospheric correction and other factors affecting the optical complexity (e.g., changes in particle size distribution). In addition, most semi-analytical algorithms (including QAA) are implemented with global parameterization for absorption spectral slopes and mass-specific absorption coefficients, which may not be appropriate for turbid estuaries [16, 35, 36], and therefore require regional tuning.

Thus, despite the recent progress in the published approaches for algorithm development, a reliable, general-purpose inversion algorithm for CDOM retrievals over turbid estuaries from satellite measurements is still unavailable. The work described here is therefore intended to fill some of the technology gaps, with the following specific objectives:

(1) Develop and demonstrate a novel, hybrid approach to derive CDOM absorption from satellite measurements over turbid estuaries using Tampa Bay as an example;
(2) Establish a long-term CDOM environmental data record (EDR) for Tampa Bay using SeaWiFS and MODIS Aqua (MODISA) measurements between 1998 and 2012;
(3) Discuss the potential applicability of the hybrid approach to other estuaries such as Chesapeake Bay.

The manuscript is arranged as follows. First, the study region is briefly introduced to show the general environmental setting, followed by a description of the data sets and the data reduction methods. Then, the hybrid algorithm approach and its validation are presented in detail. Next, a 15-year CDOM EDR for Tampa Bay is presented and described. Finally, the applicability of the hybrid approach to other estuaries is discussed.

2. Study region

Located on the west coast of Florida in the eastern Gulf of Mexico, Tampa Bay is the largest estuary in Florida, with a surface area of ~1000 km² and an average water depth of 4 m [37]. It is conventionally divided into four geographical segments, namely Old Tampa Bay (OTB), Hillsborough Bay (HB), Middle Tampa Bay (MTB), and Lower Tampa Bay (LTB) (Fig. 1(a)). Water column turbidity is generally < 10 NTU [38], with Chla in surface waters (1-2 m) ranging between 1.0 and 100 mg m⁻³ [15, 39]. A recent comprehensive study using field data collected over >10 years showed that although the variability in light absorption and attenuation is dominated by particles, CDOM plays a dominant role in affecting the total light absorption at blue-green wavelengths [9]. Similar to many other coastal environments [6–8], CDOM has a strong relationship with salinity. The physical and bio-optical properties of this moderate-turbid estuary are driven by tide, wind, and river discharge [11, 40], yet to date CDOM data are only available from a limited number of field samples, with its spatial distributions and temporal changes generally unknown.

Fig. 1 (a) Study area of Tampa Bay, Florida, USA, in the eastern Gulf of Mexico (inset figure). Following convention, Tampa Bay is divided into four segments separated by the dotted lines: Old Tampa Bay (OTB), Hillsborough Bay (HB), Middle Tampa Bay (MTB), and Lower Tampa Bay (LTB). Several major rivers that discharge into Tampa Bay are also annotated: Alafia River (AR), Hillsborough River (HR), Little Manatee River (LMR), and Manatee River (MR). (b) Station locations in Tampa Bay where bio-optical data were collected between October 2004 and November 2010.

Download Full Size | PDF

3. Data and methods

Four data sets were used in this study for algorithm development and validation and for establishing a long-term CDOM EDR.

The first data set included 71 field measurements of Chla, Rrs(λ), and absorption coefficients of water constituents during five cruise surveys in Tampa Bay from October 2004 to November 2010 (Fig. 1(b)). The details regarding the data collection, quality control, and processing methods for this data set can be found in Le et al. [9]. These data were used to develop and tune the hybrid approach for estimating CDOM from in situ Rrs(λ).

The second data set was a subset of the first data set, containing field measurements of Chla and absorption coefficients of water constituents and near-concurrent ( ± 3 hours) MODISA (n = 33) and SeaWiFS (n = 23) Rrs(λ) measurements. This data set also included CDOM and salinity measurements collected continuously along a cruise track using a flow-through system in LTB. These data were used to validate the performance of the hybrid CDOM-retrieval approach using MODISA and SeaWiFS data as input. Details concerning the collection of flow-through data and the calibration of relative CDOM fluorescence to CDOM absorption coefficients were described in Hu et al. [41].

The third data set contained all MODISA and SeaWiFS measurements for Tampa Bay obtained daily from September 1997 to December 2012. This data set was used to establish a long-term CDOM EDR for Tampa Bay. SeaWiFS and MODISA low-level (L0) daily data were obtained from the NASA Goddard Space Flight Center (http://oceancolor.gsfc.nasa.gov/), and processed with the most recent calibration and algorithms using SeaWiFS Data Analysis System (SeaDAS, Version 6.4) software. SeaWiFS data covered the period of September 1997 – December 2010, and MODISA data covered the period of July 2002 – December 2012. The outputs of the processing included Rrs(λ) which was used for validation of the newly developed hybrid CDOM-retrieval algorithm and EDR development.

The fourth data set contained 32 field measurements of Chla and Rrs(λ), and 24 field measurements of absorption coefficients of water constituents from Chesapeake Bay obtained in July 2011. These data were used to test whether the hybrid approach could be extended to other estuaries. The data collection, quality control, and processing methods are the same as those for Tampa Bay [9].

In addition to the field-measured bio-optical data and SeaWiFS and MODISA satellite data, monthly mean river discharge data were obtained from the United States Geological Survey National Water Information System (USGS NWIS) for the period of January 1997– September 2012. These included data for the four major rivers discharging into Tampa Bay: Alafia River (AR), Hillsborough River (HR), Little Manatee River (LMR), and Manatee River (MR) (Fig. 1(a)). Annual means and multi-year monthly climatology were derived from these data in order to understand the spatial patterns and temporal changes of CDOM distributions in Tampa Bay.

For algorithm validation, several criteria were used to find satellite and in situ data pairs: 1) a narrow time window of ± 3 h was used to find near-concurrent satellite and in situ measurements [42]; 2) a median value from a 3 × 3 pixel box centered at the in situ site was used to filter sensor and algorithm noise [43]; and 3) a homogeneity test was used to assure that the satellite pixel was not chosen from patchy waters. Only when the number of valid pixels (i.e., after discarding pixels associated with various quality control flags) in the 3x3 box exceeded 4 and the coefficient of variation (CV) was < 0.4 was the satellite data used to compare with in situ data [44]. The quality control flags used to discard data are those specified in the “l2_flags” data field: ATMFAIL (1), LAND (2), HIGLINT(4), HILT (5), HISATZEN (6), STRAYLIGHT (9), CLDICE (10), COCCOLITH (11), HISOLZEN (13), LOWLW (15), CHLFAIL (16), NAVWARN (17), MAXAERITER (20), CHLWARN (22), ATMWARN (23), NAVFAIL (26), and FILTER (27), where the numbers in the parenthesis represent the bit position in the 32-bit data field. To minimize possible interference from the shallow bottom, pixels with bottom depths < 2 m were excluded for both algorithm validation and EDR development [8]. After applying the above quality control criteria, 33 and 23 match-up data pairs were found for MODISA and SeaWiFS measurements, respectively.

The algorithm accuracy was assessed by calculating the mean relative error (MRE), and root mean square error (RMSE) between the measured and estimated quantities:

MRE (%) = \frac{1}{N} \sum | X_{m e a s u r e d} - X_{d e r i v e d} | / X_{m e a s u r e d} * 100

RMSE (%) = \sqrt{\frac{1}{N} \sum {[(X_{m e a s u r e d} - X_{d e r i v e d}) / X_{m e a s u r e d}]}^{2}} * 100

where X_measured represents the field measured parameter of interest (Chla, a_t-w, a_d and a_g), X_derived is the same parameter derived using the hybrid approach, and N is the number of the samples. While MRE was used to represent the mean relative difference between the two data sets, RMSE was used to represent the algorithm uncertainty.

4. Development of the hybrid CDOM-retrieval approach

Assuming that total absorption at 443nm, a_t(443), can be derived from the existing QAA [32] with acceptable uncertainties, the challenge then is how to partition a_t(443) into a_g(443) and other absorption components. Figure 2 shows the schematic flow chart of the hybrid approach developed here to derive a_g(443) from Rrs(λ), where the individual steps are listed in Table 1.

Fig. 2 Schematic flow chart showing the hybrid approach for deriving a_g(443) from Rrs(λ). The gray dashed arrows indicate the steps in the original QAA. The bold text shows the processing not provided in the original QAA. The step numbers (Sx) correspond to those listed in Table 1.

Download Full Size | PDF

Table 1. Steps of the hybrid approach for deriving a_g(443) from Rrs(λ). See Fig. 2 for a schematic flow chart of these steps.

View Table | View all tables in this article

Three general processing levels were used in the hybrid approach:

Level 0 – the ratio of backscattering coefficient to the sum of absorption and backscattering coefficients, R = b_b/(a + b_b), was calculated from Rrs on the basis of radiative transfer theory, following Lee et al. [32]:

r_{r s} (λ) = \frac{R r s (λ)}{0.52 + 1.7 R r s (λ)}

R (λ) = - 0.02 + \frac{{[0 .007 + 0 .68 r_{r s} (λ)]}^{0 .5}}{0.34}

where r_rs(λ) is the remote sensing reflectance just below the sea surface.

Level 1 – Chla was first derived with a recently developed Red-Green-Chlorophyll-Index (RGCI) algorithm [39]. Then, a_t(λ₀) and b_b(λ₀) at reference wavelength λ₀ (in this case, λ₀ = 670 nm) were derived, assuming that

a_{t} (670) = a_{g} (670) + a_{d} (670) + a_{ph} (670) + a_{w} (670) » a_{ph} (670) + a_{w} (670),

where a_w(670) is the absorption coefficient from pure water and is constant [45], and a_ph(670) was estimated empirically from Chla as:

a_{ph} (670) = A {Chla}^{B}

Here the regional coefficients A and B were determined through numerical fitting between measured a_ph(670) and Chla for Tampa Bay (A = 0.0169, B = 0.8649 [9],). The approximation in Eq. (5) to discard absorption by CDOM and detrital particles is because that for Tampa Bay, a_ph(670) contributes to ~>80% of the total absorption at this wavelength.

Then, b_bp(670) was derived from R(670) as [32]:

b_{b p} (670) = \frac{R (670) * a_{t} (670)}{1 - R (670)} - b_{b w} (670)

where b_bw(λ) is the backscattering coefficient of pure seawater and is constant [46].

Level 2 – Next, b_bp(443) and a_t(443) were derived as:

b_{b p} (443) = \frac{b_{b p} (670)}{{(443 / 670)}^{Y}}

a_{t} (443) = \frac{[1 - R (443)] * [b_{b p} (443) + b_{b w} (443)]}{R (443)}

using outputs from Level 1, R(443), and the spectral power of the particulate backscattering coefficient, Y, which was set to 1.3 [9]. In order to decompose a_t(443) into the various absorption components, empirical relationships derived from field measurements were used to derive a_ph(443) from Chla and a_d(443) from b_bp (443) [9]:

a_{ph} (443) = 0.051 * c h l a^{0.74},

a_{d} (443) = 3.32 * b_{b p} (443) + 0.0098

Finally, a_g(443) was derived as:

a_{g} (443) = a_{t} (443) - a_{d} (443) - a_{ph} (443) - a_{w} (443) .

The technical steps are detailed in Table 1, where the nature of the steps (empirical, semi-analytical, or analytical) is also annotated. The relationships in Steps 1, 2, 7, and 8 are from empirical regressions based on field measurements of Tampa Bay waters [9], while other steps are from the original QAA algorithm [32] or recent modifications made to this approach [47]. Thus, the term hybrid approach is employed here to describe the mixture of empirical, semi-analytical, and analytical steps used in this study.

5. Results

5.1 Validation of the hybrid CDOM-retrieval approach using in situ Rrs

The performance of the hybrid approach was first evaluated using the Tampa Bay in situ data set, where in situ Rrs was used as the algorithm input. Figure 3 shows the algorithm outputs (Chla, a_t-w(443), a_d(443), a_g(443)) as compared with those determined from discrete water samples, where the corresponding retrieval statistics are listed in Table 2. Chla derived from the RGCI band-ratio algorithm (Step 1 in Table 1) showed a mean relative error (MRE) of 24.4%, a relative root mean square error (RMSE) of 30.8%, a mean ratio of 0.97, and R² = 0.97 for Chla ranging between 1.0 and 80.0 mg m⁻³ (N = 71, p<0.01). Such a performance meets the SeaWiFS mission goal of achieving Chla retrievals to within ± 35% uncertainty [48], and is better than the performance of the NASA default blue/green band ratio algorithm for the global open oceans [49].

Fig. 3 Comparisons between measured and Rrs(λ)-derived (a) Chla, (b) a_t-w(443), (c) a_d(443), and (d) a_g(443) for Tampa Bay. Rrs(λ)-derived values were obtained using the hybrid approach with in situ Rrs(λ) as input. Algorithm performance is summarized in Table 2.

Download Full Size | PDF

Table 2. Algorithm performance statistics for Tampa Bay using the hybrid approach with both in situ and satellite-derived Rrs data as input

View Table | View all tables in this article

For the retrievals of absorption coefficients, except for a_d(443), the Rrs derived a_t-w(443) and a_g(443) showed excellent agreement with those determined from water samples after the default QAA was locally tuned. a_t-w(443), total absorption minus the absorption due to pure water, showed a MRE of 16.2%, a RMSE(%) of 21.0%, a mean ratio of 1.04, and R² = 0.94 for a range of 0.45 – 9.0 m⁻¹ (N = 71, p<0.01). Although the performance of a_g(443) retrieval was slightly degraded (Fig. 3(d)), the uncertainties still met the ± 35% mission goals (MRE = 19.0%, RMSE(%) = 23.1%, mean ratio = 1.04, R² = 0.97, N = 71, p<0.01). Compared to a_t-w(443) and a_g(443), estimations of a_d(443) showed the highest uncertainty, especially for a_d(443) < 0.2 m⁻¹. For the entire range, a_d(443) showed a mean ratio of 1.29, a MRE of 43.0%, and a RMSE of 65.2%. The departure from the 1:1 line mainly occurred in the low range (< 0.2 m⁻¹). The poor performance for the low range, however, did not have a significant impact on a_g(443) retrievals because a_d(443) was only a small portion of a_g(443) (on average, <~30% of a_g(443)).

For comparison, a_t-w(443) retrieved from the default QAA was also presented in Fig. 3(b), where 640 nm was selected as the reference wavelength [32]. The default parameterization of QAA did not lead to satisfactory retrievals of a_t-w(443) because the optical properties of Tampa Bay waters (e.g. absorption and scattering spectral shapes, Chla mass-specific absorption coefficients, etc.) differ greatly from the default parameterization.

5.2 Validation of the hybrid CDOM-retrieval approach using satellite-derived Rrs

If satellite-derived Rrs had negligible uncertainties, then the hybrid approach would yield similar results regardless of whether in situ or satellite Rrs were used as algorithm inputs. Unfortunately, this is typically not the case because satellite-derived Rrs(412) and Rrs(443) often show much higher uncertainties compared to longer wavelengths due to imperfect atmospheric correction over turbid coastal waters [50, 51]. Figure 4(a) presents the comparison between MODISA-derived and in situ Rrs at 443nm and 547nm for Tampa Bay. The uncertainties in the MODISA Rrs at 443nm were more than double those at 547 nm. Consequently, if such MODISA Rrs(443) data were used as the algorithm inputs in the hybrid approach (Table 1), large uncertainties would result in the retrieved a_t-w(443) and a_g(443). However, the excellent relationship between in situ Rrs(443) and in situ Rrs(550) (R² = 0.92, N = 71, p<0.001) suggests that MODISA Rrs(547) (SeaWiFS Rrs(555)) may be used to derive MODISA (SeaWiFS) Rrs(443), which can then be used as the algorithm input. Indeed, utilizing the derived MODISA Rrs(443) in place of the measured MODISA Rrs(443) resulted in a significant decrease in uncertainty, from ~45% to ~20%. Thus, in the satellite data processing, MODISA and SeaWiFS Rrs(443) were derived from MODISA Rrs(547) and SeaWiFS Rrs(555), respectively, using the relationship established in Fig. 4(b). Note that in the future when atmospheric correction is improved and uncertainties in the satellite-derived Rrs(443) are significantly reduced, this additional step would not be necessary.

Fig. 4 (a) Comparison between in situ measured Rrs(λ) and MODISA-derived Rrs(λ) at 443nm and 547nm; (b) Relationship between in situ Rrs(443) and in situ Rrs(550). All data were collected in Tampa Bay.

Download Full Size | PDF

Figure 5 presents the performance of the hybrid approach for Tampa Bay when MODISA and SeaWiFS Rrs data were used as the algorithm inputs. Retrieval statistics are summarized in Table 2. Although the performance was slightly degraded as compared to when in situ Rrs data were used as algorithm input (Fig. 3), most of the statistical measures still met the ± 35% mission goals. Chla showed uncertainties ~30% for a range of ~1.5 – 20.0 mg m⁻³, while a_t-w(443) showed lower uncertainties between 14.9% and 27.1% with a mean ratio ~1.0 for both MODISA and SeaWiFS. Once again, a_d(443) showed the highest uncertainty, but its relatively small values (compared to a_g(443)) would not affect the a_g(443) retrievals significantly. Finally, the end product, a_g(443), showed similar accuracy as a_t-w(443), with about 30% uncertainty and ~1.1 mean ratio (Fig. 5(d)). Note that a_g(443) derived from MODISA and SeaWiFS was consistent with each other. This is an important result, meaning that both MODISA and SeaWiFS can be used to form a consistent a_g(443) EDR, as shown below.

Fig. 5 Comparisons between measured and Rrs(λ)-derived (a) Chla, (b) a_t-w(443), (c) a_d(443), and (d) a_g(443) for Tampa Bay. Rrs(λ)-derived values were obtained using the hybrid approach with satellite Rrs(λ) as input. Algorithm performance is summarized in Table 2.

Download Full Size | PDF

The performance of hybrid approach was further evaluated using continuous flow-through data to test spatially whether the satellite-derived a_g(443) patterns were correct. Figure 6 shows a_g(443) from in situ measurements (3 repeated measurements along the same transect on 17 April 2008) and from same-day MODISA retrievals collected within ± 3 hours of one another. Although minor differences (RMS ~10%) were found in the absolute magnitudes, the a_g(443) patterns observed along the transect agreed well spatially between MODISA estimates and in situ measurements, with a MRE of 9.89%, RMSE of 11.68%, and a mean ratio of 1.03. Note that the range of in situ a_g(443) shown here was relatively small (0.3 – 0.5 m⁻¹) and values were low compared to the full dynamic range of values tested in Figs. 3(d) and 5(d). Recall that algorithm performance declined for these lower a_g(443) values. This may be the reason why there is still some difference between the two measurements although their general spatial patterns agree with each other. For other bay segments where a_g(443) is typically higher (e.g., > 1 m⁻¹), algorithm performance is expected to improve.

Fig. 6 (a) In situ a_g(443) (m⁻¹) estimated from CDOM fluorescence that was measured during a cruise survey on 17 April 2008 in LTB, where the same transect was measured three times with an underway flow-through system between 16:45 and 20:30 GMT; (b) a_g(443) derived from MODISA Rrs on the same day (19:05 GMT) using the hybrid approach, with the cruise track overlaid as a red line; (c) Comparison between in situ measured a_g(443) and MODISA-derived a_g(443) along the transect.

Download Full Size | PDF

5.3 Long-term CDOM EDR

Given the satisfactory performance of the hybrid approach in deriving a_g(443) from satellite measurements, the approach was applied to available MODISA data (July 2002 – December 2012) and SeaWiFS data (September 1997 – December 2010) to establish a 15-year CDOM EDR for Tampa Bay.

Figure 7 shows the 15-year climatological monthly mean a_g(443) derived from MODISA and SeaWiFS using the hybrid approach. a_g(443) exhibited significant variability in both space and time, which was also clearly visible in Fig. 8(a) for individual bay segments. In general, a_g(443) decreased from the upper and middle bay segments (HB, OTB, MTB) to the lower bay segment (LTB) as the latter received more influence of the much clearer water from the Gulf of Mexico [40]. The highest a_g(443) was found in HB, followed by a_g(443) in OTB and MTB, and then by a_g(443) in LTB. Seasonally, a_g(443) in the wet season (July to October) was significantly higher than in the dry season (November to June) for all bay segments. These results suggest that the seasonality of a_g(443) appears to be associated with river discharge (Fig. 8(b)), similar to the Chla patterns derived from MODISA [17]. The month of May showed the lowest a_g(443), corresponding to the lowest river discharge. Monthly mean a_g(443) in all four bay segments showed significant correlation (p<0.01) with monthly mean river discharge. For example, the Pearson correlation coefficient (r) between monthly mean a_g(443) in HB and monthly mean river discharge was 0.59 and 0.65 for the AR and HR, respectively. a_g(443) in MTB showed r = 0.43, 0.56, and 0.58 for the correlation with monthly mean discharge from the LMR, AR, and HR, respectively. For LTB, a_g(443) showed r = 0.52 for the correlation with monthly mean discharge from the MR. An interesting result from this analysis is that although LTB showed significantly lower a_g(443) than the other three bay segments for all months, the monthly mean a_g(443) in other bay segments was similar (Fig. 8(a)).

Fig. 7 Monthly mean a_g(443) (m⁻¹) in Tampa Bay derived from a combined SeaWiFS and MODISA climatology (1997-2002: SeaWiFS; 2003 – 2010, SeaWiFS and MODISA; 2011-2012: MODISA).

Download Full Size | PDF

Fig. 8 (a) Monthly means and standard deviations of a_g(443) (m⁻¹) in Tampa Bay for individual bay segments (HB, OTB, MTB, and LTB) derived from a combined SeaWiFS and MODISA climatology (1997-2002: SeaWiFS; 2003 – 2010, SeaWiFS and MODISA; 2011-2012: MODISA); (b) Monthly means and standard deviations of river discharge from the four main rivers (AR, HR, LMR, and MR) for the same period.

Download Full Size | PDF

Figure 9 shows the annual mean a_g(443) derived from MODISA and SeaWiFS using the hybrid approach. Substantial inter-annual variability was found in a_g(443), as is also shown in Fig. 10(a) for individual bay segments. The highest a_g(443) occurred in 1998 as a result of higher river runoff during the 1997-1998 El Niño event (Fig. 10(c)). The lowest a_g(443) was found in 2000 for all bay segments as a result of lower river runoff during a strong La Niña event (1999-2000). Annual mean a_g(443) in HB ranged from 0.41 m⁻¹ in 2000 to 1.38 m⁻¹ in 1998, with a 15-year mean a_g(443) of 0.69 ± 0.26 m⁻³. The 15-year mean a_g(443) values for OTB (0.66 ± 0.20 m⁻¹) and MTB (0.65 ± 0.32 m⁻¹) were comparable to that observed in HB. a_g(443) in LTB was the lowest, ranging from 0.21 m⁻¹ in 2000 to 0.71 m⁻¹ in 1998 (15-year mean = 0.39 ± 0.14 m⁻¹). Of the four bay segments, MTB showed the highest inter-annual variability, with a 15-year standard deviation (SD) of 0.32 m⁻¹, followed by HB (0.26 m⁻¹), OTB (0.21 m⁻¹), and LTB (0.14 m⁻¹). This is likely because MTB has the most dynamic environment where waters from the upper bay segments with high a_g(443) (HB and OTB) mix with waters from LTB with lower a_g(443).

Fig. 9 Annual mean a_g(443) (m⁻¹) in Tampa Bay derived from a combined SeaWiFS and MODISA climatology (1998-2002: SeaWiFS; 2003-2010, SeaWiFS and MODISA; 2011-2012: MODISA).

Download Full Size | PDF

Fig. 10 Annual means and standard deviations of (a) a_g(443) and (b) Chla for individual bay segments (HB, OTB, MTB, and LTB) derived from a combined SeaWiFS and MODISA climatology (1997-2002: SeaWiFS; 2003 – 2010, SeaWiFS and MODISA; 2011-2012: MODISA) using the hybrid CDOM-retrieval approach and the RGCI algorithm, respectively. (c) Climatological annual means and standard deviations of river discharge from the four main rivers (AR, HR, LMR, and MR) for the same period (discharge data for 2012 was not available).

Download Full Size | PDF

The inter-annual variability of a_g(443) shared similar trends with Chla derived from MODISA and SeaWiFS using the RGCI algorithm (Fig. 10(b)). Increased Chla during high discharge years likely results from increased nutrient loading supplied by the various rivers emptying into Tampa Bay. The highest a_g(443) occurred in the El Niño years of 1998 and 2003 when the highest river discharges were also recorded as a response to ENSO [52]. Likewise, the lowest a_g(443) occurred in the La Niña years of 1999-2000 and 2007 when the lowest river discharges were recorded. Significant correlation between annual mean a_g(443) and river discharge was found for the 15-year period for each bay segment (p<0.01), with r = 0.91 and 0.96 between HB a_g(443) and discharges by the HR and AR, respectively. a_g(443) in MTB and LTB showed r of 0.88 and 0.84 for their correlation with LMR and MR discharges, respectively. The coefficient of determination (r²) suggests that between 73% and 92% of the interannual variability in a_g(443) of the various bay segments can be explained by local river discharges. Furthermore, significant correlation was found between monthly mean a_g(443) anomalies and Multivariate ENSO Index (MEI) [53] (r = 0.55, p<0.01, n = 184), suggesting that CDOM variability at both monthly and inter-annual scales was related to climate variability.

One striking difference found between a_g(443) and Chla was in their cross-bay (north–south) gradients (Figs. 10(a) and 10(b)). Annual mean a_g(443) did not show any significant difference between HB, OTB, and MTB (p>0.4) although they were all significantly higher than LTB. In contrast, cross-bay differences in the annual mean Chla was all statistically significant (p<0.05). In other words, a_g(443) was more homogeneous than Chla spatially between HB, OTB, and MTB. Similar results were also obtained from the climatological monthly means (Fig. 8(a)). This suggests that compared to Chla, CDOM was well mixed between these bay segments. Similarly, the higher correlation between a_g(443) and river discharge (~0.9) than between Chla and river discharge (~0.8) [39] indicates that CDOM in Tampa Bay is more directly related to river discharge while other factors (e.g., nutrients and light) may also contribute to Chla variability.

6. Discussion

6.1 Algorithm performance and general applicability

Retrieving CDOM in coastal and estuarine waters from satellite measurements has been challenging because CDOM and non-algal particles have similar absorption spectral shapes and because satellite-derived Rrs in the blue bands tend to exhibit higher uncertainties than in other bands. The hybrid approach demonstrated here addressed these two issues using locally-derived empirical relationships to tune a community-accepted QAA algorithm [32] for improved a_g(443) retrieval accuracies in coastal and estuarine waters. Specifically, absorption coefficients of phytoplankton pigments (a_ph(443)) and non-algal particles (a_d(443)) were estimated from Rrs(λ) and its derived Chla and b_bp data based on pre-determined empirical relationships from a large in situ data set, and then subtracted from the QAA-derived a_t-w(443) to obtain a_g(443). To deal with the high uncertainties of the satellite-derived Rrs(443), MODISA Rrs(547) (SeaWiFS Rrs(555)) was first used to derive Rrs(443) from an empirical relationship between in situ Rrs(443) and Rrs(550), and then used in the hybrid approach. The various statistical measures suggested that the hybrid approach was robust for Tampa Bay for a large dynamic range in a_g(443). Indeed, uncertainties in the retrieved a_g(443) were well below the ± 35% satellite ocean color mission goals for uncertainty requirements. In contrast, if satellite-derived Rrs(443) were used directly in the approach, the uncertainties in the retrieved a_g(443) would reach ~85%. Clearly, improved atmospheric correction is required in future work to improve the satellite Rrs performance in the blue bands.

The question then asked was can the hybrid approach developed for Tampa Bay be applied to other estuaries? This will depend primarily on 1) whether a_d(443) can be derived empirically from b_bp and 2) whether satellite-derived Rrs(443) is sufficiently accurate or at least can be derived empirically from one or more other bands. Additional constraints that may apply include whether Chla can be accurately estimated from satellite measurements for deriving a_ph(443). Although each estuary may have its unique optical characteristics, an example using Chesapeake Bay is given below to demonstrate that this approach can be extended to other estuaries.

With a surface area of ~11,601 km² and average bottom depth of 14 m, the Chesapeake Bay is the largest estuary in the U.S. An extensive watershed contributes an annual average of 2.3 × 10³ m³ s⁻¹ freshwater flow to the estuary, associated with dissolved and particulate matters including nutrients and sediments. Similar to Tampa Bay, the Chesapeake Bay is optically complex with absorption at blue wavelengths influenced by all three optically significant constituents (phytoplankton, CDOM, and non-algal particles) and scattering mainly regulated by suspended sediments [13, 54].

The same hybrid approach was applied to in situ data collected in Chesapeake Bay in July 2011, with results presented in Fig. 11. Note that the empirical algorithms in Step 1 and Step 7 in Table 1 were locally-tune relationships (present in Figs. 11(a) and 11(c)). While statistical measures in the retrieved Chla, a_t-w(443), and a_g(443) were similar to those for Tampa Bay, R² values were much lower due to a narrower dynamic range in values . One striking difference between Chesapeake Bay and Tampa Bay is that while a_d(443) is only a small portion of a_g(443) in Tampa Bay, a_d(443) in Chesapeake Bay is comparable or even higher than a_g(443). The 25% or less uncertainties in the retrieved a_g(443) (range from 0.3 m⁻¹ to 0.6 m⁻¹) are deemed acceptable, validating the applicability of the hybrid approach for Chesapeake Bay.

Fig. 11 Comparisons between in situ measured and in situ Rrs(λ)-derived (a) Chla, (b) a_t-w(443), (c) a_d(443), and (d) a_g(443) using the hybrid approach in Chesapeake Bay.

Download Full Size | PDF

However, cautions must be taken on whether the assumptions used in the hybrid approach are still valid when the approach is applied to other estuaries. In particular, whether a_t-w(670) can be approximated by a_ph(670) needs to be verified, and whether an empirical relationship exists between a_d(443) and b_bp(443) also needs to be verified. These can only be achieved through targeted field measurements, from which empirical relationships may be derived. In the extreme cases where a_t-w(670) is dominated by suspended sediments (e.g., Taihu Lake and Yangtze River estuary in China [55, 56];), the approach is likely to fail to decompose a_g(443) from a_t-w(443) accurately.

6.2 Implications for environmental management

The value of the hybrid approach goes well beyond the CDOM retrievals for estuaries, as it may be potentially useful for deriving surface salinities.

Salinity is one of the most important water quality parameters for assessing ecosystem health in estuaries, providing valuable information for environmental management. For example, one of the two variables used in statistical models in predicting Vibrio spp. distributions is salinity [57–59]. Salinity in estuaries is often difficult to monitor at synoptic scales because of the often strong salinity gradients. Although the recent launch of the Aquarius satellite (http://aquarius.nasa.gov) made it possible to estimate surface salinity at global scale, the resolution (50 km or lower) is too coarse to be applicable for estuaries. Some statistical approaches using neural networks or Principal Component Analysis (PCA) have been proposed to estimate salinity directly from satellite-derived Rrs(λ) (e.g., see [60] for Chesapeake Bay), yet the underlying physics and governing equations are unclear and their application to different estuaries would require re-training of the neural network or PCA.

The availability of satellite-derived a_g(443) may lead to an alternative way to estimate salinity from ocean color measurements, as salinity appears to be related to CDOM for estuaries and coastal waters due to water mixing [6–8]. Once a robust CDOM product is available, the only required data is a locally-tuned salinity-CDOM relationship that is often driven by mixing but can also be modulated by local chemical or biological processes [7]. For Tampa Bay, salinity showed a strong relationship with CDOM (Fig. 12; also see [8]). Figure 12(a) shows the relationship between salinity and a_g(443) measured along the east–west cross-bay transect using a flow-through system (Fig. 6), while Fig. 12(b) shows the relationship between in situ salinity and MODISA derived a_g(443) from 2007 to 2010. In this latter figure, the salinity data were collected by the Environment Protection Commission of Hillsborough County (EPCHC) once a month at fixed locations throughout Tampa Bay, and a_g(443) was derived from MODISA Rrs measurements collected on the same day as the ship measurements. Relatively strong relationships were found for both cases, with less uncertainties found for the in situ a_g(443) (~1%) than for the MODISA a_g(443) (~6%). The latter case would translate to a salinity uncertainty of about 2 for the salinity range of 25-36. Note that some of the uncertainties may result from the inherent difference between satellite and in situ measurements (1 km² versus a point measurement, and data collected hours apart), and some data scatter may also be caused by the different CDOM and salinity sources from different rivers. Nevertheless, the preliminary test shows the potential of using satellite ocean color measurements to derive salinity in Tampa Bay with moderate accuracy (uncertainty of ~2 for salinity between 25 and 36). We expect to expand this work to focus on salinity retrievals in both Tampa Bay and Chesapeake Bay in the near future.

Fig. 12 (a) Relationship between field-measured salinity and a_g(443) along a transect in LTB on 17 April 2008 (see Fig. 6(b) for transect location); (b) Relationship between field-measured salinity and MODISA-derived a_g(443) in Tampa Bay. Salinity data in (b) were collected once a month by EPCHC from January 2007 to September 2010, and a_g(443) was derived from MODISA Rrs(λ) measurements using the hybrid algorithm on the same day as the field measurements.

Download Full Size | PDF

7. Summary and conclusion

A hybrid approach was developed to estimate CDOM absorption in a turbid estuary, Tampa Bay using satellite measurements. The approach combines a modified QAA, which is based on physical principles governing the radiative transfer equations, and empirical relationships among various bio-optical properties. While the former derives total absorption coefficient at 443 nm (a_t(443)) and particulate backscattering coefficient (b_bp) from either in situ or satellite-derived Rrs(λ), the latter derives the absorption coefficients at 443 nm by phytoplankton pigments (a_ph(443)) and by non-algal (detrital) particles (a_d(443)) based on pre-determined empirical relationships between these properties and Chla and b_bp, respectively. a_ph(443) and a_d(443) can then be explicitly removed from a_t(443), leading to the estimation of a_g(443).

The hybrid approach has been validated using field data collected from Tampa Bay covering a large dynamic range (a_g(443) ~0.3 – 8 m⁻¹), with overall uncertainties of ~20% when in situ Rrs data are used as the algorithm input. Application of the same approach to MODISA and SeaWiFS data leads to a_g(443) retrieval uncertainties of ~30% when Rrs(443) is derived from satellite Rrs(550) and used as the algorithm input. This is to minimize problems in atmospheric correction for the blue bands. The 15-year CDOM EDR established from MODISA and SeaWiFS measurements between 1998 and 2012 shows substantial variability in both space and time, which can be explained by river discharge and water mixing. The strong relationship between salinity and a_g(443) also suggests that satellite-based salinity distribution maps may be derived for this estuary.

The work described here demonstrates a new approach in overcoming the traditional difficulties in remote sensing of CDOM distributions in optically complex waters. The approach also shows potentials for applications in other estuaries, although rigorous evaluation and validation have yet to be conducted.

Notations

MODISA Moderate Resolution Imaging Spectroradiometer / Aqua
SeaWiFS Sea-viewing Wide Field-of-view Sensor
Chla Chlorophyll-a concentration (mg m⁻³).
CDOM Chromophoric Dissolved Organic Matter.
a_ph(λ) Absorption coefficient of phytoplankton pigments (m⁻¹).
a_d(λ) Absorption coefficient of detrital particles (m⁻¹).
a_g(λ) Absorption coefficient of CDOM (m⁻¹).
a_t(λ) Total absorption coefficient (water constituents + water, m⁻¹).
a_t-w(λ) Total absorption coefficient without pure water (m⁻¹)
b_b(λ) Total backscattering coefficient (particulate + water, m⁻¹).
b_bp(λ) Particulate backscattering coefficient (m⁻¹)
Y Spectral slope of b_bp (dimensionless)
Rrs(λ) Above-water surface remote sensing reflectance (sr⁻¹).
OTB Old Tampa Bay
HB Hillsborough Bay
MTB Middle Tampa Bay
LTB Lower Tampa Bay

Acknowledgements

This work was supported by NASA’s Water and Energy Program, Ocean Biology and Biogeochemistry Program, and Gulf of Mexico Program. This work benefited from the long-term efforts from several agencies and groups to collect, quality control, and share the field and satellite data. These include the EPCHC, Florida DEP, USGS, Chesapeake Bay Program, Tampa Bay Estuary Program, and the U.S. NASA. We thank David English, Jennifer Cannizzaro, Jun Zhao, Hongtao Duan, Daniel Sensi, Ryan Lloyd for their help in collecting and processing the Tampa Bay and Chesapeake Bay data, and thank the NASA OBPG for providing all satellite data and processing software. We appreciate the substantial comments and suggestions from two anonymous reviewers who helped improve the quality of this manuscript.

References and links

1. J. T. O. Kirk, Light and Photosynthesis in Aquatic Ecosystems, 2nd ed., (Cambridge Univ. Press, Cambridge, U. K., 1994) p. 509.

2. H. Gao and R. G. Zepp, “Factors influencing photoreactions of dissolved organic matter in a coastal river of the southeastern United States,” Environ. Sci. Technol. 32(19), 2940–2946 (1998). [CrossRef]

3. N. B. Nelson, D. A. Siegel, and A. F. Michaels, “Seasonal dynamics of colored dissolved material in the Sargasso Sea,” Deep Sea Res. Part I Oceanogr. Res. Pap. 45(6), 931–957 (1998). [CrossRef]

4. C. A. Stedmon, S. Markager, M. Søndergaard, T. Vang, A. Laubel, N. H. Borch, and A. Windelin, “Dissolved organic matter (DOM) export to a temperate estuary: seasonal variations and implications of land use,” Estuaries Coasts 29, 388–400 (2006).

5. P. G. Coble, “Marine optical biogeochemistry: The chemistry of ocean color,” Chem. Rev. 107(2), 402–418 (2007). [CrossRef] [PubMed]

6. G. M. Ferrari and M. D. Dowell, “CDOM absorption characteristics with relation to fluorescence and salinity in coastal areas of the Southern Baltic Sea,” Estuar. Coast. Shelf Sci. 47(1), 91–105 (1998). [CrossRef]

7. D. G. Bowers and H. L. Brett, “The relationship between CDOM and salinity in estuaries: an analytical and graphical solution,” J. Mar. Syst. 73(1-2), 1–7 (2008). [CrossRef]

8. Z. Chen, C. Hu, R. N. Conmy, F. E. Muller-Karger, and P. Swarzenski, “Colored dissolved organic matter in Tampa Bay, Florida,” Mar. Chem. 104(1-2), 98–109 (2007a). [CrossRef]

9. C. Le, C. Hu, D. English, J. Cannizzaro, Z. Chen, C. Kovach, C. J. Anastasiou, J. Zhao, and K. L. Carder, “Inherent and apparent optical properties of the complex estuarine waters of Tampa Bay: what controls light?” Estuar. Coast. Shelf Sci. 117, 54–69 (2013a). [CrossRef]

10. K. Oubelkheir, L. A. Clementson, I. T. Webster, P. W. Ford, A. G. Dekker, L. C. Radke, and P. Daniel, “Using inherent optical properties to investigate biogeochemical dynamic in a tropical macrotidal coastal system,” J. Geophys. Res. 111(C7), C07021 (2006), doi:. [CrossRef]

11. Z. Chen, C. Hu, F. E. Muller-Karger, and M. E. Luther, “Short-term variability of suspended sediment and phytoplankton in Tampa Bay, Florida: observations from a coastal oceanographic tower and ocean color satellites,” Estuar. Coast. Shelf Sci. 89(1), 62–72 (2010). [CrossRef]

12. J. E. Cloern, “Our evolving conceptual model of the coastal eutrophication problem,” Mar. Ecol. Prog. Ser. 210, 223–253 (2001). [CrossRef]

13. L. W. Harding Jr, A. Magnuson, and M. E. Mallonee, “Bio-optical and remote sensing observations in Chesapeake Bay,” Estuar. Coast. Shelf Sci. 62, 75–94 (2005). [CrossRef]

14. J. Udy, M. Gall, B. Longstaff, K. Moore, C. Roelfsema, D. R. Spooner, and S. Albert, “Water quality monitoring: a combined approach to investigate gradients of change in the Great Barrier Reef, Australia,” Mar. Pollut. Bull. 51(1-4), 224–238 (2005). [CrossRef] [PubMed]

15. Z. Chen, F. E. Muller-Karger, and C. Hu, “Remote sensing of water clarity in Tampa Bay,” Remote Sens. Environ. 109(2), 249–259 (2007b). [CrossRef]

16. M. Wang, S. Son, and L. W. Harding Jr., “Retrieval of diffuse attenuation coefficient in the Chesapeake Bay and turbid ocean regions for satellite ocean color applications,” J. Geophys. Res. 114(C10), C10011 (2009), doi:. [CrossRef]

17. C. Le, C. Hu, D. English, J. Cannizzaro, Z. Chen, L. Feng, R. Boler, and C. Kovach, “Towards a long-term chlorophyll-a data record in a turbid estuary using MODIS observations,” Prog. Oceanogr. 109, 90–103 (2013b). [CrossRef]

18. Z. P. Lee, K. L. Carder, T. G. Peacock, C. O. Davis, and J. L. Mueller, “Method to derive ocean absorption coefficients from remote-sensing reflectance,” Appl. Opt. 35(3), 453–462 (1996). [CrossRef] [PubMed]

19. C. C. Liu and R. L. Miller, “Spectrum matching method for estimating the chlorophyll-a concentration, CDOM ratio, and backscatter fraction from remote sensing of ocean color,” Can. J. Rem. Sens. 34(4), 343–355 (2008). [CrossRef]

20. J. Fischer, “On the information content of multispectral radiance measurements over an ocean,” Int. J. Remote Sens. 6(5), 773–786 (1985). [CrossRef]

21. R. Doerffer and H. Schiller, “Determination of case 2 water constituents using radiative transfer simulation and its inversion by neural networks”, in Proceedings of Ocean Optics XIV [CD-ROM], S. G. Ackleson and J. Campbell, Kailua-kona, ed. (academic 1998), 1–13.

22. E. J. D’Sa and R. L. Miller, “Bio-optical properties in waters influenced by the Mississippi River during low flow conditions,” Remote Sens. Environ. 84(4), 538–549 (2003). [CrossRef]

23. D. Doxaran, R. C. N. Cherukuru, and S. J. Lavender, “Use of reflectance band ratios to estimate suspended and dissolved matter concentrations in estuarine waters,” Int. J. Remote Sens. 26(8), 1763–1769 (2005). [CrossRef]

24. P. Kowalczuk, L. Olszewski, M. Darecki, and S. Kaczmarek, “Empirical relationships between coloured dissolved organic matter (CDOM) absorption and apparent optical properties in Baltic Sea waters,” Int. J. Remote Sens. 26(2), 345–370 (2005). [CrossRef]

25. A. Mannino, M. E. Russ, and S. B. Hooker, “Algorithm development and validation for satellite-derived distributions of DOC and CDOM in the US Middle Atlantic Bight,” J. Geophys. Res. 113(C7), C07051 (2008), doi:. [CrossRef]

26. S. P. Tiwari and P. Shanmugam, “An optical model for the remote sensing of coloured dissolved organic matter in coastal/ocean waters,” Estuar. Coast. Shelf Sci. 93(4), 396–402 (2011). [CrossRef]

27. N. C. Tehrani, E. J. D’Sa, C. L. Osburn, T. S. Bianchi, and B. A. Schaeffer, “Chromophoric dissolved organic matter and dissolved organic carbon from Sea-Viewing Wide Field-of-View Sensor (SeaWiFS), Moderate Resolution Imaging Spectroradiometer (MODIS) and MERIS Sensors: case Study for the Northern Gulf of Mexico,” Remote Sens. 5(3), 1439–1464 (2013). [CrossRef]

28. International Ocean-Colour Coordinating Group (IOCCG), “Remote sensing of inherent optical properties: Fundamentals, tests of algorithms, and applications,” Z. P. Lee (Ed.), Reports of the International Ocean-Colour Coordinating Group, No. 5. Dartmouth, Canada: IOCCG (2006).

29. K. L. Carder, F. R. Chen, Z. P. Lee, S. K. Hawes, and D. Kamykowski, “Semianalytic 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]

30. S. Maritorena and D. A. Siegel, “Consistent merging of satellite ocean color data sets using a bio-optical model,” Remote Sens. Environ. 94, 429–440 (2005). [CrossRef]

31. D. A. Siegel, S. Maritorena, N. B. Nelson, D. A. Hansell, and M. Lorenzi-Kayser, “Global distribution and dynamics of colored dissolved and detrital organic materials ,” J. Geophys. Res. 107, 3228, DOI:. (2002). [CrossRef]

32. Z. P. Lee, K. L. Carder, and R. A. Arnone, “Deriving inherent optical properties from water color: a multiband quasi-analytical algorithm for optically deep waters,” Appl. Opt. 41(27), 5755–5772 (2002). [CrossRef] [PubMed]

33. Q. Dong, S. Shang, and Z. Lee, “An algorithm to retrieve absorption coefficient of chromophoric dissolved organic matter from ocean color,” Remote Sens. Environ. 128, 259–267 (2013). [CrossRef]

34. W. Zhu, Q. Yu, Y. Tian, R. Chen, and G. B. Gardner, “Estimation of chromophoric dissolved organic matter in the Mississippi and Atchafalaya river plume regions using above-surface hyperspectral remote sensing,” J. Geophys. Res. 116(C2), C02011 (2011), doi:. [CrossRef]

35. Y. Qin, V. E. Brando, A. G. Dekker, and D. Blondeau-Patissier, “Validity of SeaDAS water constituents retrieval algorithms in Australian tropical coastal waters,” J. Geophys. Res. Lett. 34(21), L21603 (2007), doi:. [CrossRef]

36. C. Le, Y. Li, Y. Zha, D. Sun, and B. Yin, “Validation of a quasi-analytical algorithm for highly turbid eutrophic water of Meiliang Bay in Taihu Lake, China,” IEEE Trans. Geosci. Rem. Sens. 8, 2490–2500 (2009).

37. M. A. Harwell, J. S. Ault, and J. H. Gentile, Comparison of the ecological risks to the Tampa Bay ecosystem from spills of fuel #6 and Orimulsion. Comparative Ecological Risk Assessment, Volume 1, Center for Marine and Environmental Analyses. (University of Miami, Miami, Florida, 1995.)

38. Z. Chen, C. Hu, and F. E. Muller-Karger, “Monitoring turbidity in Tampa Bay using MODIS/Aqua 250-m imagery,” Remote Sens. Environ. 109(2), 207–220 (2007c). [CrossRef]

39. C. Le, C. Hu, D. English, J. Cannizzaro, and C. Kovach, “Climate-driven chlorophyll-a changes in a turbid estuary: observations from satellites and implications for management’,” Remote Sens. Environ. 130, 11–24 (2013c). [CrossRef]

40. R. H. Weisberg and L. Zheng, “Circulation of Tampa Bay driven by buoyancy, tides, and winds, as simulated using a finite volume coastal ocean model,” J. Geophys. Res. 111(C1), C01005 (2006), doi:. [CrossRef]

41. C. Hu, Z. Chen, T. D. Clayton, P. Swarzenski, J. C. Brock, and F. E. Muller-Karger, “Assessment of estuarine water-quality indicators using MODIS medium-resolution bands: initial results from Tampa Bay, FL,” Remote Sens. Environ. 93(3), 423–441 (2004). [CrossRef] [PubMed]

42. S. W. Bailey and P. J. Werdell, “A multi-sensor approach for the on-orbit validation of ocean color satellite data products,” Remote Sens. Environ. 102(1-2), 12–23 (2006). [CrossRef]

43. C. Hu, K. L. Carder, and F. E. Muller-Karger, “How precise are SeaWiFS ocean color estimates? Implications of digitization-noise errors,” Remote Sens. Environ. 76(2), 239–249 (2001). [CrossRef]

44. L. W. Harding Jr, A. Magnuson, and M. E. Mallonee, “SeaWiFS retrievals of chlorophyll in Chesapeake Bay and the mid-Atlantic bight,” Estuar. Coast. Shelf Sci. 62(1-2), 75–94 (2005b). [CrossRef]

45. 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] [PubMed]

46. A. Morel, Optical properties of pure water and pure seawater. E. Steeman Nielsen ed. (Academic, 1974, pp 1–24).

47. Z. P. Lee, B. Lubac, J. Werdell, and R. Arnone, “An update of the Quasi-Analytical Algorithm (QAA v5),” http://www. ioccg.org/groups/Software OCA/QAA v5.pdf (2009).

48. S. B. Hooker, W. E. Esaias, G. C. Feldman, W. W. Gregg, and C. R. McClain, An overview of SeaWiFS and ocean color. NASA Tech. Memo., vol. 104566. (National Aeronautics and Space Administration, Goddard Space Flight CenterGreenbelt, MD, 1992).

49. W. W. Gregg and N. W. Casey, “Global and regional evaluation of the SeaWiFS chlorophyll dataset,” Remote Sens. Environ. 93(4), 463–479 (2004). [CrossRef]

50. P. J. Werdell, S. W. Bailey, B. A. Franz, L. W. Harding Jr, G. C. Feldman, and C. R. McClain, “Regional and seasonal variability of chlorophyll-a in Chesapeake Bay as observed by SeaWiFS and MODIS-Aqua,” Remote Sens. Environ. 113(6), 1319–1330 (2009). [CrossRef]

51. S. Son and M. Wang, “Water properties in Chesapeake Bay from MODIS-Aqua measurements,” Remote Sens. Environ. 123, 163–174 (2012). [CrossRef]

52. N. Schmidt, E. K. Lipp, J. B. Rose, and M. E. Luther, “ENSO influences on Seasonal Rainfall and River Discharger in Florida,” J. Clim. 14(4), 615–628 (2001). [CrossRef]

53. K. Wolter and M. S. Timlin, “El Niño/Southern Oscillation behavior since 1871 as diagnosed in an extended multivariate ENSO index (MEI.ext),” Int. J. Climatol. 31(7), 1074–1087 (2011). [CrossRef]

54. M. Tzortziou, A. Subramanian, J. R. Herman, C. L. Gallegos, P. J. Neale, and L. W. Harding, “Remote sensing reflectance and inherent optical properties in the mid Chesapeake Bay,” Estuar. Coast. Shelf Sci. 72(1-2), 16–32 (2007). [CrossRef]

55. D. Sun, Y. Li, Q. Wang, C. Le, C. Huang, and L. Wang, “Parameterization of water component absorption in an inland eutrophic lake and its seasonal variability: a case study in Lake Taihu,” Int. J. Remote Sens. 30(13), 3549–3571 (2009). [CrossRef]

56. F. Shen, Y. X. Zhou, D. J. Li, W. J. Zhu, and M. S. Salama, “Medium resolution imaging spectrometer (MERIS) estimation of chlorophyll-a concentration in the turbid sediment-laden waters of the Changjiang (Yangtze) Estuary,” Int. J. Remote Sens. 31(17-18), 4635–4650 (2010). [CrossRef]

57. V. R. Louis, E. Russek-Cohen, N. Choopun, I. N. G. Rivera, B. Gangle, S. C. Jiang, A. Rubin, J. A. Patz, A. Huq, and R. R. Colwell, “Predictability of Vibrio cholerae in Chesapeake Bay,” Appl. Environ. Microbiol. 69(5), 2773–2785 (2003). [CrossRef] [PubMed]

58. G. C. Magny, W. Long, C. W. Brown, R. R. Hood, A. Huq, R. Murtugudde, and R. R. Colwell, “Predicting the distribution of Vibrio spp. in the Chesapeake Bay: a vibrio cholera case study,” EcoHealth (2010), doi:. [CrossRef]

59. J. M. Jacobs, M. Rhodes, C. W. Brown, R. R. Hood, A. Leigh, and W. L. R. Wood, “Predicting the distribution of Vibrio vulnificus in Chesapeake Bay,” NOAA Technical Memorandum NOS NCCOS 112. 1–12 (2010).

60. E. A. Urquhat, B. F. Zaitchik, M. J. Hoffman, S. D. Guikema, and E. F. Geiger, “Remote sensed estimates of surface salinity in the Chesapeake Bay: a statistical approach,” Remote Sens. Environ. 123, 522–531 (2012). [CrossRef]

Step	Properties	Math formula	Approach
0	R(λ)	= f(Rrs(λ))	Semi-analytical
1	Chla	= 36.55*RGCI^1.98 (RGCI = Rrs(670)/Rrs(550))	Empirical
2	a_ph(670)	= 0.017*Chla^0.86	Empirical
3	a_t(670)	= a_ph(670) + a_w(670)	Semi-analytical
4	b_bp(670)	= $\frac{R (670) * a_{t} (670)}{1 - R (670)} - b_{b w} (670)$	Analytical
5	b_bp(443)	= $\frac{b_{b p} (670)}{{(443 / 670)}^{Y}}$	Analytical
6	a_t(443)	= $\frac{[1 - R (443)] * [b_{b p} (443) + b_{b w} (443)]}{R (443)}$	Analytical
7	a_d(443)	= 3.32*b_bp(443) + 0.0098	Empirical
8	a_ph(443)	= 0.051*Chla^0.74	Empirical
9	a_g(443)	= a_t(443)-a_d(443)-a_ph(443)-a_w(443)	Analytical

Algorithm input	Statistical measure	Chla	a_t-w	a_d	a_g
In situ Rrs	MRE(%)	24.2	16.2	43.0	19.0
N = 71	RMSE(%)	30.8	21.0	65.2	23.1
	Mean ratio	0.97	1.04	1.29	1.04
	R²	0.94	0.95	0.83	0.97
MODISA Rrs	MRE(%)	32.6	21.4	42.4	23.7
N = 33	RMSE(%)	36.3	27.1	52.2	29.6
	Mean ratio	0.85	1.0	1.0	1.07
	R²	0.70	0.73	0.45	0.72
SeaWiFS Rrs	MRE(%)	32.4	14.9	41.7	17.6
N = 23	RMSE(%)	38.6	23.3	60.6	25.7
	Mean ratio	0.76	1.03	1.20	1.06
	R²	0.73	0.89	0.42	0.83

Step	Properties	Math formula	Approach
0	R(λ)	= f(Rrs(λ))	Semi-analytical
1	Chla	= 36.55*RGCI^1.98 (RGCI = Rrs(670)/Rrs(550))	Empirical
2	a_ph(670)	= 0.017*Chla^0.86	Empirical
3	a_t(670)	= a_ph(670) + a_w(670)	Semi-analytical
4	b_bp(670)	= $\frac{R (670) * a_{t} (670)}{1 - R (670)} - b_{b w} (670)$	Analytical
5	b_bp(443)	= $\frac{b_{b p} (670)}{{(443 / 670)}^{Y}}$	Analytical
6	a_t(443)	= $\frac{[1 - R (443)] * [b_{b p} (443) + b_{b w} (443)]}{R (443)}$	Analytical
7	a_d(443)	= 3.32*b_bp(443) + 0.0098	Empirical
8	a_ph(443)	= 0.051*Chla^0.74	Empirical
9	a_g(443)	= a_t(443)-a_d(443)-a_ph(443)-a_w(443)	Analytical

Algorithm input	Statistical measure	Chla	a_t-w	a_d	a_g
In situ Rrs	MRE(%)	24.2	16.2	43.0	19.0
N = 71	RMSE(%)	30.8	21.0	65.2	23.1
	Mean ratio	0.97	1.04	1.29	1.04
	R²	0.94	0.95	0.83	0.97
MODISA Rrs	MRE(%)	32.6	21.4	42.4	23.7
N = 33	RMSE(%)	36.3	27.1	52.2	29.6
	Mean ratio	0.85	1.0	1.0	1.07
	R²	0.70	0.73	0.45	0.72
SeaWiFS Rrs	MRE(%)	32.4	14.9	41.7	17.6
N = 23	RMSE(%)	38.6	23.3	60.6	25.7
	Mean ratio	0.76	1.03	1.20	1.06
	R²	0.73	0.89	0.42	0.83

A hybrid approach to estimate chromophoric dissolved organic matter in turbid estuaries from satellite measurements: A case study for Tampa Bay

Abstract

1. Introduction

2. Study region

3. Data and methods

4. Development of the hybrid CDOM-retrieval approach

5. Results

5.1 Validation of the hybrid CDOM-retrieval approach using in situ Rrs

5.2 Validation of the hybrid CDOM-retrieval approach using satellite-derived Rrs

5.3 Long-term CDOM EDR

6. Discussion

6.1 Algorithm performance and general applicability

6.2 Implications for environmental management

7. Summary and conclusion

Notations

Acknowledgements

References and links

Cited By

Figures (12)

Tables (2)

Equations (12)

Optics Express