2.1 Study area

Galveston Bay (GB), a shallow water estuary (∼2.1 m average depth), encompasses two major sub-estuaries: San Jacinto Estuary (also divided as Upper GB and Lower GB) and Trinity Estuary (Trinity Bay) (Fig. 1). It is located adjacent to the heavily urbanized and industrialized metropolitan areas of Houston, Texas (Dorado et al., 2015). The deep (∼14 m) narrow Houston Ship Channel connects the bay to the northern Gulf of Mexico (nGoM) through a narrow entrance, the Bolivar Roads Pass. Tidal exchange between GB and the nGoM occurs through the entrance channel with diurnal tides ranging from ∼0.15 to ∼0.5 m. The major freshwater sources to GB are the Trinity River (55 %), the San Jacinto River (16 %) and Buffalo Bayou (12 %) (Guthrie et al., 2012). The San Jacinto River was frequently observed to transport greater amounts of dissolved nutrients into GB than the Trinity River (Quigg, 2011); however, the negative relationship between nitrate concentrations and salinity observed in the mid-bay area (adjacent to Smith Point) (Santschi, 1995) indicated Trinity River to be a major source of nitrate in GB. The catastrophic flooding of Houston and surrounding areas associated with Hurricane Harvey resulted in strong freshwater inflows into GB from the San Jacinto River (>3300 m³ s⁻¹; USGS 08067650) on 29 August 2017 and the Trinity River (>2500 m³ s⁻¹; USGS 08066500 site at Romayor, Texas) on 30 August 2017, respectively. Although the discharge from the two rivers in the upstream returned to normal conditions (∼50–120 m³ s⁻¹) in about 2 weeks after the Hurricane passage, salinity remained low for over a month following the hurricane passage (D'Sa et al., 2018).

2.2 Sampling and data collection

Surface water samples were collected at a total of 34 stations during two surveys on 29 September and 29–30 October 2017 (Fig. 1). Samples at stations 1 to 14 (red asterisk on top of green circle; Fig. 1) along the Trinity River transect were collected repeatedly on 29 September and 29 October 2017, respectively. An additional nine sampling sites (blue squares; Fig. 1) around the upper Galveston Bay and in the East Bay were sampled on 30 October 2017. The surface water samples were stored in coolers and filtered on the same day. The filter pads were immediately frozen and stored in liquid nitrogen for laboratory absorption spectroscopic and HPLC measurements of the samples. An optical package equipped with a conductivity–temperature–depth recorder (SBE; Sea-Bird Scientific) and a fluorescence induction and relaxation system (FIRe; Satlantic Inc.) were used to obtain profiles of salinity, temperature, pressure and phytoplankton physiological parameters (F_v∕F_m and σ_PSII). Measurements of backscattering were also made at each station using the WETLabs VSF-3 (470, 530, 670 nm) backscattering sensor (D'Sa et al., 2006). Included in the optical package was also a hyperspectral downwelling spectral irradiance meter (HyperOCR; Satlantic Inc.). The irradiance data from the HyperOCR were processed using ProSoft 7.7.14 and the photosynthetically active radiation (PAR) were estimated from the irradiance measurements. The above-water reflectance measurements were collected using a GER 1500 512iHR spectroradiometer in the 350–1050 nm spectral range. At each station, sky radiance, plate radiance and water radiance were recorded (each repeated three times) and processed to obtain the above-water remote-sensing reflectance (Joshi et al., 2017). A total of 43 Sentinel-3A OLCI full resolution mode, cloud free level-1 images were obtained over GB between 1 August 2016 and 1 December 2017 from the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) website and pre-processed using Sentinel-3 Toolbox Kit Module (S3TBX) version 5.0.1 in Sentinel Application Platform (SNAP). These Sentinel-3A OLCI data were further atmospherically corrected to obtain remote-sensing reflectance ($R_{{\mathrm{rs}}_{\mathrm{O}}\mathrm{LCI}}$, sr⁻¹) using Case-2 Regional/Coast Color (C2RCC) module version 0.15 (Doerffer and Schiller, 2007). River discharge information during August 2016–December 2017 was downloaded from the USGS Water Data (USGS) for the Trinity River at Romayor, Texas (USGS 08066500), and the west flank of the San Jacinto River (USGS 08067650). Individual pictures of microplankton (10 to 150 µm) recorded by an Imaging FlowCytobot (IFCB) located at the entrance to Galveston Bay were downloaded (http://dq-cytobot-pc.tamug.edu/tamugifcb, last access: 8 November 2018) for pigment validation.

2.3 Absorption spectroscopy

Surface water samples were filtered through 0.2 µm nuclepore membrane filters, and the colored dissolved organic matter (CDOM) absorbance (A_CDOM) was obtained using a 1 cm path length quartz cuvette on a PerkinElmer Lambda-850 UV–VIS spectrophotometer equipped with an integrating sphere. The quantitative filter technique (QFT) with 0.7 µm GF/F filters was used to measure absorbance of particles (A_total) and non-algal particles (A_NAP) inside an integrating sphere at 1 nm intervals from 300 to 800 nm. The absorption coefficients of CDOM (a_CDOM), NAP (a_NAP), particles (a_total) and phytoplankton (a_PHY) were calculated using the following equations:

where L is the path length in meters. The a_CDOM were corrected for scattering, temperature and baseline drift by subtracting an average value of absorption between 700 and 750 nm for each spectrum (Joshi and D'Sa, 2015).

where V_filtered is the filtered volume of sample, S_filter is the area of filter pads and the path length correction for filter pad was applied according to Stramski et al. (2015).

2.4 Pigment absorption spectra

The water samples were filtered with 0.7 µm GF/F filter. The filter pads were stored in liquid nitrogen until transferred into 30 mL vials containing 10 mL cold 96 % ethanol (Ritchie, 2006). The vials were spun evenly to ensure full exposure of the filter pad to the ethanol and then kept in the refrigerator (in the dark) overnight. The pigment solutions at room temperature were poured off from vials into 1 cm cuvettes and measured on a PerkinElmer Lambda-850 UV–VIS spectrophotometer to obtain pigment absorbance spectra (A_pig) while 90 % ethanol was used as a blank (Thrane et al., 2015). The total absorption coefficients of pigments a_pig(λ) were calculated as follows:

where L is the path length in meters, V_ethanol is the volume of ethanol and V_filtered is the filtered volume of the water samples.

2.5 HPLC measurements

Water samples were filtered through 0.7 µm GF/F filters and immediately frozen in liquid nitrogen for HPLC analysis using the methods reported by Barlow et al. (1997). The detected pigments along with their abbreviations are listed in Table 1. Diagnostic biomarker pigments are marked in bold letters (Table 1).

Table 1Pigment information acquired from HPLC samples in Galveston Bay. Diagnostic biomarker pigments are marked in bold.

Note: (1) [Chl b], [Allo], [Fuco], [Peri], [Zea], [Buta] and [Hexa] are considered as diagnostic pigments for PFTs (Moisan et al., 2017).

Download Print Version | Download XLSX

2.6 FIRe measurements

An in situ FIRe system (Satlantic Inc.) was used to characterize phytoplankton photosynthetic physiology during the two surveys in GB. The FIRe system is based on illuminating a sample with an intense flash of light to instantaneously saturate the reaction centers of photosystem II (PSII); under these light conditions, reaction centers do not accept electrons and most of the absorbed light energy is dissipated as fluorescence. The fundamental parameter measured by the FIRe system is fluorescence yield F(t), which is the emitted fluorescence divided by the irradiance intensity (no unit). In contrast to strong flashes, dark adaption enables all reaction centers of PSII to be open with least fluorescence emitted, thus resulting in minimal fluorescence yield (F_o). Maximum fluorescence yield (F_m) can be obtained after sufficient irradiation when all reaction centers are closed. Maximum photochemical efficiency, which quantifies the potential of converting light to chemical energy for the PSII reaction centers (Moore et al., 2006), was calculated as ($F_{\mathrm{m}}-F_o)/F_{\mathrm{m}}=F_{\mathrm{v}}/F_{\mathrm{m}}$. The functional absorption cross section σ_PSII (Ųquantum⁻¹) measures the capability of reaction centers to absorb light from the ambient environment. The FIRe system was deployed at a slow descent rate, with 12 and 20 vertical profiles obtained during the first and second surveys, respectively. All measurements were programmed using standard protocols of single saturating turnover (ST) flash saturation of PSII (Kolber et al., 1998). Flashes were generated from highly uniform blue LEDs at 455 nm with approximately 30 nm half-bandwidth. Chl a fluorescence was stimulated using a saturating sequence of 80 1.1 µs flashes applied at 1 µs intervals, eight sequences were averaged per acquisition, and the fluorescence signal was detected at 668 nm. All data were processed using standard FIReCom software (Satlantic Inc.). In addition, samples of 0.2 µm filtered sea water at each station were used as “blank” to remove the background fluorescence signals (Cullen and Davis, 2003); in this step, the fluorescence from the filtered samples (without phytoplankton) was subtracted from in situ fluorescence signals to get more accurate values of F_v∕F_m.

2.7 Retrieving phytoplankton groups from above-water R_rs

A fundamental relationship that links subsurface remote-sensing reflectance (r_rs) and the IOPs was expressed using a quadratic function developed by Gordon et al. (1988):

where the parameters g₁ (∼0.0788) and g₂ (∼0.2379) are values for turbid estuarine waters (Joshi and D'Sa, 2018); r_rs is the subsurface remote-sensing reflectance that was obtained from above-water remote-sensing reflectance (R_rs) using the following equation (Lee et al., 2002)

The total backscattering coefficient b_b is comprised of water (b_bw) and particulates including both organic and inorganic particles (b_bp), while the total absorption coefficients (a_total) can be further separated into four sub-constituents (Roesler and Perry, 1995) as indicated by

where a_w, a_phy, a_CDOM, and a_NAP represent the absorption coefficients of pure water, phytoplankton, colored dissolved organic matter and non-algal particles, respectively.

The IOP inversion algorithm for retrieving IOPs from R_rs require known spectral shape (eigenvector) of each component in Eq. (8) to estimate the magnitude (eigenvalue) of each component (Table 2). The spectral shape can be adjusted by changing the values of slope based on characteristics of the study area. It is worth noting that a single averaged phytoplankton eigenvector does not provide species information, whereas a set of several species-specific phytoplankton eigenvectors allow estimates of species composition. The IOP inversion algorithm applied in this study makes use of mass-specific phytoplankton absorption spectra of 10 groups, namely dinoflagellates, diatoms, chlorophytes, cryptophytes, haptophytes, prochlorophytes, raphidophytes, rhodophyta, red cyanobacteria and blue cyanobacteria; these were obtained from Dierssen et al. (2006) and Dutkiewicz et al. (2015) as eigenvectors rather than using one averaged a_phy(λ) spectrum. Subsequently, the inversion algorithm iterates repeatedly to minimize the difference between modeled R_rs and in situ measured R_rs (R_{rs_insitu}) until a best fit is achieved while allowing for alterations of all parameters listed in Table 2 (Chase et al., 2017). The absolute percent errors between modeled and measured values of R_rs, a_phy, a_CDOM, a_NAP and b_bp were calculated as

2.8 Retrieving pigments from Sentinel-3A OLCI R_rs

2.9 Processing approach

Sentinel-3A OLCI pigment concentration maps were generated using the processing pathway 1 (Fig. 2) that includes the following: (1) developing empirical relationships between HPLC-measured Chl a and R_{rs_insitu} band ratio for Sentinel-3A OLCI band 9 (673 nm) and band 11 (709 nm) to generate Sentinel-3A OLCI Chl a maps and (2) converting Chl a concentration to a_{pig_OLCI}(λ) maps and subsequently decomposing a_{pig_OLCI}(λ) into individual pigment spectra to generate phytoplankton pigment maps for GB. In processing pathway 2, phytoplankton taxonomic composition at each sampling station was obtained from a 10-species IOP inversion model, which uses R_{rs_insitu} as input and estimates Chl a concentration of each phytoplankton group (Fig. 2). Finally, CDOM-corrected FIRe measurements of F_v∕F_m and σ_PSII were combined with phytoplankton taxonomy to assess photosynthetic physiology of different phytoplankton groups.

Table 2Parameters and eigenvectors used in the semi-analytical inversion algorithm.

Note: $a_{\mathrm{phi}}^{*}\left(\mathit{\lambda }\right)$ for 10 different groups of phytoplankton used in this study were extracted from Dierssen et al. (2006) and Dutkiewicz et al. (2015).

Download Print Version | Download XLSX

Figure 2Flowchart showing the three processing steps as follows: (1) retrieving pigments spatial distribution maps from OLCI, (2) distinguishing phytoplankton groups, and (3) assessing phytoplankton physiological parameters and their linkages to taxonomic groups.

Download

3.1 Phytoplankton taxonomy and physiological state from field observations

3.1.1 Measurements of above-water remote-sensing reflectance

Above-water remote-sensing reflectances (R_{rs_insitu}) from the two surveys (Fig. 3) reflect the influence of the absorbing and scattering features of water constituents. Low reflectance (∼675 nm) caused by Chl a red light absorption and maximum reflectance at green wavelength (∼550 nm) were observed. Obvious dips at ∼625 nm versus reflectance peaks ∼650 nm were observed in spectra during both surveys, which could be attributed to cyanobacteria modulation of the spectra (Hu et al., 2010). The reflectance peak around 690–700 nm was obvious at most sampling sites except at stations 13 and 14 adjacent to the nGOM and were likely due to the effect of Chl a fluorescence (Gitelson, 1992; Gilerson et al., 2010). The peak positions at stations with lower Chl a concentration (∼5 µg L⁻¹) were observed at 690–693 nm; however, the peaks shifted to longer wavelengths of 705 and 710 nm for stations 23 and 19 with extremely high Chl a concentrations of ∼31 and 50 µg L⁻¹, respectively (Fig. 3).

Figure 3R_{rs_insitu} spectra at stations in GB on 29 September and 29–30 October 2017; vertical bars represent Sentinel-3A OLCI spectral bands.

Download

3.1.2 Performance of the IOP inversion algorithm

The IOP inversion algorithm was applied to R_{rs_insitu} data (Fig. 3) obtained during the two surveys in GB. The mean errors for modeled a_CDOM, a_NAP, a_phy and b_{bp_470} at all wavelengths for the 34 stations were 5.86 %, 6.83 %, 12.19 % and 10.79 %, respectively (Table 3). A total of eight phytoplankton groups (dinoflagellate, diatom, chlorophyte, cryptophyte, haptophyte, prochlorophyte, raphidophyte and blue cyanobacteria) were spectrally detected from IOP inversion algorithm. The sum of eight eigenvalues of Chl_i (Table 2) represents the modeled total Chl a (TChl a_mod) of the whole phytoplankton community. The TChl a_mod is better correlated with HPLC-measured total Chl a (TChl a_HPLC) for survey 2 (green circle; Fig. 4a) with R²∼0.92, compared to survey 1 (red color; Fig. 4a). In addition, the TChl a_mod appears to be slightly higher than TChl a_HPLC for survey 2. The modeled a_CDOM (a_{CDOM_mod}) are in close agreement with the spectrophotometrically measured a_CDOM at 412 nm (Fig. 4b), with the a_CDOM obtained on 29 September 2017 much higher than that obtained on 29–30 October 2017. The modeled b_bp (b_{bp_mod}) are well correlated with in situ b_bp (b_{bp_insitu}) at 470 nm (Fig. 4c) with higher R² (0.81) observed on 29 September 2017. In addition, both modeled and field-measured b_bp showed stronger backscattering at most stations on 29 September 2017 than those on 29–30 October 2017.

Figure 4(a) Validation of TChl a_mod via HPLC-measured TChl a; individual %Chl a of each detected taxa versus corresponding %DP shown with (d) cryptophyte, (e) chlorophyte, (f) cyanobacteria, (g) diatom, (h) dinoflagellate and (l) haptophyte; red and green dots indicate the samples on 29 September and 29–30 October 2017, respectively. Comparison between in situ measurements and modeled results with (b) a_CDOM (412) and (c) b_bp (470).

Download

Table 3Error statistics over all wavelengths and sampling stations ($N=\mathrm{301}\times \mathrm{34}=\mathrm{10}\mathrm{234}$; 12 and 22 stations on 29 September and 29–30 October 2017) from a semi-analytical IOP inversion algorithm.

Download Print Version | Download XLSX

The Chl a percentage (%Chl a), which is Chl_i ∕ TChl a, was also compared with diagnostic pigment percentage (%DP), which is the specific DP for each phytoplankton group over the sum of DP (∑DP). The DP for diatoms (fucoxanthin), dinoflagellates (peridinin), cryptophytes (alloxanthin), chlorophytes (Chl b), haptophytes (19^′-hexanoyloxyfucoxanthin) and cyanobacteria (zeaxanthin) referred in Moisan et al. (2017) were used in this study. The R² between %Chl a and %DP for different phytoplankton groups ranges from 0.15 to 0.81 (Fig. 4). The %Chl a of cryptophytes is between 5 % and 42 % and is well correlated with alloxanthin ∕ ∑DP (R²∼0.62–0.72; Fig. 4d) for both surveys. In addition, the cryptophyte %Chl a at station 19 and 23 on 30 October 2017 was highest (∼40 %) in coincidence with the highest value of alloxanthin ∕ ∑DP (Fig. 4d). Furthermore, the relationship between chlorophyte %Chl a and Chl b ∕ ∑DP (R²∼0.55; Fig. 4e) shows that chlorophytes during survey 1 contributed a higher fraction to the whole phytoplankton community compared to survey 2. The %Chl a of cyanobacteria highly correlated with zeaxanthin ∕ ∑DP with R² larger than 0.7 (Fig. 4f) for both surveys. Low %Chl a of dinoflagellates in coincidence with low peridinin ∕ ∑DP (R²∼0.78) were observed at stations along the transect; however, an increased contribution of dinoflagellates appeared adjacent to the entrance during both surveys (Fig. 4g).

3.1.3 Variations in phytoplankton community structure

Reconstruction of the phytoplankton absorption coefficients spectra revealed variations in phytoplankton community structure (Fig. 5) even several weeks after Hurricane Harvey. The modeled a_phy spectra (a_{phy_mod}) at stations 6, 13, 17 and 19 (Fig. 5a–f) yielded spatiotemporal differences of phytoplankton taxonomic composition in GB. The strong absorption peak around 625 nm induced by cyanobacteria was observed at most of the stations for both modeled results and in situ measurements (Fig. 5a, c and e) except at stations adjacent to the entrance (Fig. 5b and d). The a_{phy_mod} at station 6 was primarily dominated by group of cyanobacteria (blue line) and chlorophytes (green line) on 29 September 2017 (Fig. 5a); in contrast, the spectrum of chlorophytes contributed very little at station 6 on 29 October 2017 (green line; Fig. 5c). Furthermore, the shape of spectra for samples obtained at station 13 shows strong dinoflagellate modulation versus extremely low cyanobacteria contribution during survey 1 (red line; Fig. 5b). However, small-size groups like the haptophytes and prochlorophytes displayed increasing proportions at station 13 on 29 October 2017 (Fig. 5d). Station 17 in the East Bay was dominated by cyanobacteria (blue line; Fig. 5e) and cryptophyte (pink line; Fig. 5e) absorption spectra, whereas, on 30 October 2017, the main spectral features at station 19 in the upper GB was from cryptophytes (pink line) and chlorophytes (green line; Fig. 5f).

Figure 5Reconstruction of phytoplankton absorption coefficients spectra at stations 6 (a) and 13 (b) on 29 September 2017, at stations 6 (c) and 13 (d) on 29 October 2017 and at stations 17 (e) and 19 (f) on 30 October 2017 based on the mass-specific absorption spectra of different phytoplankton groups including diatoms, chlorophytes, dinoflagellates, cryptophytes, cyanobacteria (blue), haptophytes, prochlorophytes and raphidophytes presented using different colors.

Download

The corresponding taxa-specific %Chl a derived from IOPs inversion algorithm for the two surveys on 29 September and 29–30 October 2017 are shown in Fig. 6a and b, respectively. Cyanobacteria (blue bars) and chlorophytes (green bars) constituted over 55 % of the phytoplankton communities during survey 1 (29 September 2017; Fig. 6a). In addition, chlorophytes, known to proliferate in freshwater environments, were present in a higher fraction than that observed in survey 2 (Fig. 6). Further, chlorophytes together with diatoms (Fig. 6a) accounted for ∼60 % of TChl a_mod at many stations with a well-mixed water column (e.g., stations 7, 8 and 9 as inferred from salinity profiles; not shown) on 29 September 2017. Cryptophytes, haptophytes and raphidophytes became minor components of the community and accounted in total to ∼25 % of TChl a_mod (Fig. 6a). Furthermore, contribution by the dinoflagellate group to TChl a_mod was low inside the bay, but showed increasing %Chl a (∼30 %) in higher-salinity waters adjacent to the nGOM (red color; Fig. 6a). Cyanobacteria (blue color; Fig. 7) exhibited a slightly elevated percentage during survey 2 (∼60 d after the hurricane passage, 29–30 October 2017) and were quite abundant in East Bay (stations 16, 17 and 18) where the water was calm and stratified (as indicated by the salinity profiles – not shown). In addition, cyanobacteria were not prevailing adjacent to the nGOM (stations 12, 13 and 14) and close to San Jacinto (stations 19, 20, 21, 23 and 24), where cryptophytes and chlorophytes showed dominance (Fig. 6b). The %Chl a of chlorophytes obtained at stations along the Trinity River transect decreased by ∼10 % on 29–30 October 2017 compared to 29 September 2017. Small-size groups like haptophytes and prochlorophytes increased on 29–30 October 2017 and were more abundant adjacent to the nGOM, accounting for more than 25 % of the TChl a_mod.

Figure 6Phytoplankton taxonomic compositions detected from IOP inversion algorithm on (a) 29 September and (b) 29–30 October 2017 in Galveston Bay; phytoplankton groups are represented in different colors as shown in the legend.

Download

3.1.4 Environmental conditions and physiological state of phytoplankton community

The surface salinity presented a pronounced seaward increasing gradient along the transect (stations 3–14) during both surveys (Fig. 7a) with primarily lower salinity throughout the bay during survey 1 in comparison to survey 2, which indicated the freshening impact was still ongoing even 4 weeks after Hurricane Harvey. The salinity was ∼15 at station 16 and decreasing when going further into East Bay (∼10 at stations 17 and 18; Fig. 7a). In upper GB, salinity at stations 19–24 did not vary significantly (∼15), increasing along with the distance away from the San Jacinto River mouth with highest value (∼17.5) at station 24. During both surveys, lowest Chl a (Fig. 7b) were observed adjacent to the nGOM, and the highest Chl a were closest to the river mouth. The photosynthetically active radiation (PAR), which were calculated from down-welling irradiance (not shown here), decreased significantly with depth, but surface PAR (Fig. 7c) were similar in magnitude at all stations. Pigment ratios including TChl a ∕ TP (0.58–0.68), PSC ∕ Chl a (0.07–0.26) and AP ∕ TP (0.34–0.42) were obtained from HPLC measurements and shown in Fig. 7d, e and f, respectively.

Figure 7Environmental conditions (salinity, light field), pigment composition and physiological state in GB surface waters (red bars indicating samples from 29 September 2017 and blue bars representing samples from 29 and 30 October 2017). (a) Salinity, (b) Chl a concentration, (c) PAR, (d) Chl a ∕ TP, (e) PSC ∕ Chl a, (f) AP ∕ TP, (g) F_v∕F_m and (h) σ_PSII.

Download

The CDOM-calibrated and 0–0.5 m depth-averaged photosynthetic parameters F_v∕F_m varied from 0.41 to 0.64 (Fig. 7g) while σ_PSII was in the range of 329–668 Ųquantum⁻¹(Fig. 7h). The highest σ_PSII and lowest F_v∕F_m appeared adjacent to the nGOM (stations 12–14). Conversely, values of F_v∕F_m at stations 7–9 with a well-mixed water column were high with low values of σ_PSII. Both F_v∕F_m and σ_PSII did not directly correlate with Chl a (e.g., high Chl a∼51 µg L⁻¹ at station 19 corresponded to a relatively low level of $F_{\mathrm{v}}/F_{\mathrm{m}}\sim \mathrm{0.45}$ versus high σ_PSII∼550 Ųquantum⁻¹). However, the stations with high F_v∕F_m coincided with the high fraction of Chl a (Chl a ∕ TP) and the low fraction of AP (AP ∕ TP) (Fig. 7d and f). In contrast, σ_PSII showed an overall positive relationship with AP ∕ TP but altered negatively with Chl a ∕ TP during both surveys. The lowest (highest) value of σ_PSII (F_v∕F_m) were observed at station 9, corresponding to the highest Chl a ∕ TP value (∼0.64) on 29 October 2017. The highest AP ∕ TP and PSC ∕ Chl a were obtained from stations adjacent to the nGOM.

3.1.5 F_v∕F_m and σ_PSII taxonomic signatures

Distinct pigments housed within phytoplankton light-harvesting antennae can strongly influence PSII light-harvesting capability and the photosynthetic quantum efficiency of phytoplankton (Lutz et al., 2001). In this study, we observed an inverse relationship (R²∼0.63–0.81; Fig. 8a and d) between the F_v∕F_m and σ_PSII that appeared related to taxonomic signals during surveys 1 and 2 in GB. Stations 1–9 along the transect were considered as a well-mixed group with no dominance by any particular group (black circles; Fig. 8a–c); stations 10–14 close to the entrance were, however, strongly dominated by dinoflagellates and haptophytes (red symbol; Fig. 8a–c) during survey 1. This well-mixed group displayed low values of σ_PSII (∼390–439 Ųquantum⁻¹) and high levels of F_v∕F_m (∼0.42–0.65), with F_v∕F_m approaching 0.65 at station 9 on 29 September 2017 (Fig. 8a). However, enhanced contributions of dinoflagellates and haptophytes around the entrance corresponded to a decline of F_v∕F_m (0.3–0.4) against an increase of σ_PSII (500–600 Ųquantum⁻¹) during survey 1. Furthermore, samples obtained from survey 2 at stations 1–9, stations 10–14, stations 16–18 and stations 19–24 were considered as well mixed (black), dinoflagellate–haptophyte-dominated (red), cyanobacteria-dominated (blue) and cryptophyte–chlorophyte-dominated (green), respectively. Stations 16–17 dominated by cyanobacteria (blue triangles; Fig. 8d) showed high level of F_v∕F_m (0.5–0.6) and relatively low values of σ_PSII (300–400 Ųquantum⁻¹). The F_v∕F_m and σ_PSII of cryptophyte–chlorophyte-dominated stations showed a moderate level of F_v∕F_m (0.4–0.5) and σ_PSII (580–680 Ųquantum⁻¹). More importantly, tight positive relationships existed between measurements of F_v∕F_m and Chl a ∕ TP (R²∼0.31–0.63; Fig. 8b and e). On the other hand, σ_PSII values were positively correlated with the PSC ∕ Chl a with R²∼0.6 (Fig. 8c and f). The PSC ∕ Chl a of cyanobacteria-dominated group (blue symbols) and the well-mixed group (brown symbols) were relatively low. The highest PSC ∕ Chl a and lowest Chl a ∕ TP values were observed for the dinoflagellate–haptophyte-dominated group, corresponding to the lowest σ_PSII and the highest F_v∕F_m. In addition, the cryptophyte–chlorophyte-dominated group had high levels of PSC/TChl a (∼0.18–0.26) and a slightly higher Chl a ∕ TP level compared to the dinoflagellate–haptophyte-dominated group. Overall, well-mixed groups with high proportion of large-size phytoplankton (e.g., diatoms and chlorophytes) showed higher Chl a ∕ TP levels along with larger F_v∕F_m and smaller σ_PSII values than those stations with a high fraction of dinoflagellates and picophytoplankton (e.g., haptophytes and prochlorophytes)(Fig. 8c and f).

Figure 8(a, d) σ_PSII vs. F_v∕F_m; (b, e) F_v∕F_m versus Chl a ∕ TP; and (c, f) σ_PSII versus PSC ∕ Chl a on 29 September and 29–30 October 2017, respectively. The data points identified by dominant taxa with black, red, green and blue symbols denoting well-mixed, dinoflagellate–haptophyte-dominated, cryptophyte–chlorophyte-dominated and cyanobacteria-dominated groups, respectively.

Download

Figure 9(a) Relationship between the percentage of the fluorescence yield of CDOM measured by FIRe against HPLC-measured Chl a concentration. The dashed circle corresponds to data acquired in the area adjacent to the GOM. (b) Comparisons between R_{rs_insitu} and R_{rs_OLCI} at band 9 (673 nm) and band 11 (709 nm). The dashed circles correspond to stations 1, 3 and 4 located close to the river mouth and are considered as outliers. (c) Exponential relationships between HPLC-measured Chl a concentrations and R_{rs_insitu} band ratio (673 nm ∕ 709 nm) in GB on September 29 (R²=0.89), 29 October (R²=0.93) and 30 October (R²=0.97). Red, green and blue lines and symbols indicate data sets obtained on 29 September, 29 October and 30 October 2017, respectively.

Download

3.2 Satellite observations of phytoplankton pigments

3.2.1 An OLCI Chl a algorithm and its validation

Blue to green band ratio algorithms have been widely used to study Chl a in the open ocean and shelf waters (D'Sa et al., 2006; Blondeau-Patissier et al., 2014); however, these bands generally fail in estuarine waters due to strong blue absorption by the high levels of CDOM and suspended particulate matter, especially after flooding events associated with hurricanes (D'Sa et al., 2011, 2018; Joshi and D'Sa, 2018). The percentage contribution by CDOM fluorescence (blank) to maximum fluorescence yield (F_m) obtained from in situ FIRe (Fig. 9a) demonstrated that Chl a fluorescence was strongly influenced by high amounts of CDOM fluorescence in GB, especially during the first survey (29 September 2017) when the bay was under strong floodwater influence (red triangles; Fig. 9a). The CDOM fluorescence signal constituted ∼25 % in the region adjacent to the nGOM (stations 12–14), between 25 % and 50 % in the upper GB and up to ∼65 % in Trinity Bay, which implies that blue and even green bands are highly contaminated by CDOM and might not be the most suitable bands for estimating Chl a in GB. However, an increase in peak height near 700 nm and its shift towards a longer wavelength (Fig. 3) can be used as a proxy to estimate Chl a concentration (Gitelson, 1992).

The C2RCC atmospheric-corrected R_{rs_OLCI} at each of the sampling sites were further compared with R_{rs_insitu} (Fig. 3) at phytoplankton red absorption (∼673 nm) and Chl a fluorescence (∼700 nm) bands (Fig. 9b). The C2RCC performed overall better for the second survey on 29–30 October 2017 (green and blue symbols; Fig. 9b) than the first survey on 29 September 2017 (red triangles; Fig. 9b) when stations 1, 3 and 4 (circled triangles; Fig. 3c) adjacent to the Trinity River mouth were included; these stations were the last sampling sites in the afternoon (∼ 04:30 pm) and under somewhat cloudy conditions. The time difference between the satellite pass and in situ measurements, differences in sky conditions and in shallow water depths also likely introduced more errors at these locations. The R² between R_{rs_OLCI} and R_{rs_insitu} at red and near infra-red (NIR) bands was 0.89 when the data from stations 3 and 4 were excluded, suggesting good usability of these two bands for Chl a empirical algorithms in GB. Thus, the higher the Chl a concentration, the stronger the red light absorption, resulting in higher reflectance at 709 nm; consequently, negative correlations were observed between red ∕ NIR band ratio and Chl a. The ratio of red (∼673 nm) ∕ NIR (709 nm) reflectance bands from in situ measurements were overall highly correlated with HPLC-measured Chl a with R²∼0.96, 0.94 and 0.98 on 29 September, 29 and 30 October 2017, respectively (Fig. 9c). The Sentinel-3A OLCI Chl a maps (Fig. 10a–c) were generated for all data based on the relationship between Chl a and the red ∕ NIR band ratio as

$$\begin{array}{ll}{\displaystyle }\mathrm{Chl}a\left(\mathrm{\mu }\mathrm{g}{\mathrm{L}}^{-\mathrm{1}}\right)& {\displaystyle }=\mathrm{216.38}\times \exp (-\mathrm{2.399}\\ \text{(14)}& {\displaystyle }& {\displaystyle }\times {\displaystyle \frac{R_{\mathrm{rs}}\left(\mathrm{673}\right)}{R_{\mathrm{rs}}\left(\mathrm{709}\right)}})\text{[All data]},\end{array}$$

The OLCI-derived Chl a (Fig. 10a–c) showed a good spatial agreement with Chl a_HPLC (Fig. 10d–f). In addition, a comparison of this algorithm with that of Gilerson et al. (2010) revealed slightly better performance (not shown) inside of GB and especially in the area adjacent to the shelf.

Figure 10Chl a concentrations generated based on an in situ band ratio (R_rs673 ∕ R_rs709) algorithm with (a), (b) and (c) representing Chl a distribution on 29 September, 29 and 30 October 2017, respectively. Panels (d), (e) and (f) show the validation between HPLC-measured Chl a and OLCI-derived Chl a on 29 September, 29 and 30 October 2017, respectively.

Download

The Chl a concentration on 29 September 2017 was overall higher than on 29–30 October 2017 throughout the bay. East Bay displayed very high Chl a concentrations, with the highest value (>30 µg L⁻¹) observed on 29 September 2017 (Fig. 10a). The narrow shape and shallow topography of East Bay results in relatively higher water residence time (Rayson et al., 2016); thus, the reduced exchange with shelf waters likely leaves the East Bay vulnerable to eutrophication. The average Chl a concentrations on 29–30 October 2017 were ∼15 µg L⁻¹ along the transect (station 1–11) and ∼4–6 µg L⁻¹ (station 12–14) close the entrance of GB. In addition, the Chl a concentration adjacent to San Jacinto River mouth (>16 µg L⁻¹) was higher than that in Trinity Bay, which might suggest that the San Jacinto inflow had higher nutrient concentrations than Trinity Bay as also previously reported (Quigg et al., 2010). Furthermore, the OLCI-Chl a maps on 29 and 30 October 2017 show extremely high Chl a concentrations in a narrow area adjacent to the San Jacinto River mouth, with Chl a approaching ∼40 µg L⁻¹ at station 19 (Fig. 10c).

3.2.2 Long-term Chl a observations in comparison with the Hurricane Harvey event

OLCI-derived Chl a maps between August 2016 and November 2017 (Fig. 11a₁–a₁₅) and time series of averaged Chl a in the areas of Trinity Bay, East Bay and adjacent to the nGOM (Fig. 11b) revealed regionally different responses to freshwater discharge from the San Jacinto and the Trinity Rivers (Fig. 11b). Due to the relatively much higher discharge from the Trinity River, the spatial distribution of Chl a in Trinity Bay (Fig. 11) generally indicates its greater influence than the San Jacinto River. During the winter and spring of 2017, phytoplankton Chl a peaks of ∼32 µg L⁻¹ were observed in Trinity Bay (Fig. 11b) after high inflows from both rivers (Fig. 11a₅–a₈) and then decreased to ∼20 µg L⁻¹ in summer (July and August 2017). Generally, overall lower Chl a values (∼10 µg L⁻¹) were observed between September and December 2016 compared to 2017 in the absence of significant meteorological and hydrological events (Fig. 11a₁–a₄). However, with the East Bay less directly affected by river discharge, Chl a levels remained fairly constant in the range of ∼18–24 µg L⁻¹ before the hurricane. In contrast, extremely high river discharge (∼3300 m³ s⁻¹) induced by Hurricane Harvey in late August 2017, elevated the Chl a concentrations in both the Trinity and East Bay to higher levels as observed on 14 September 2017 (∼30–35 µg L⁻¹; Fig. 11a₁₁) compared to the mean state of the fall season in 2016. Chl a then continuously decreased through September and October 2017 in Trinity Bay and East Bay, and were relatively low (≤10 µg L⁻¹) in November 2017 under no additional pulses of river discharge. Chl a levels adjacent to the entrance of GB, which exhibited much lower values year round than that of the Trinity Bay and East Bay, also showed a slightly positive response to the enhanced river discharge and the hurricane-induced flooding events. In addition, Chl a was always observed in low concentrations along the Houston Ship Channel.

Figure 11(a₁₋₁₅) OLCI-derived Chl a shown for the period of 31 August 2016–25 November 2017. (b) Trinity River discharge at Romayor, Texas (USGS 08066500; black line), and the west flank of the San Jacinto River (USGS 08067650; blue line); the green, red and gray lines and symbols represent the mean of Chl a at stations 1–7 in Trinity Bay, at stations 17–18 in East Bay and at stations 12–14 close to the entrance of GB corresponding to 43 cloud-free Sentinel-3A OLCI images (colored symbols; dated symbols correspond to images a₁₋₁₅).

Download

3.2.3 Reconstruction of total pigment absorption spectra from OLCI-derived Chl a

The reconstructed a_pig(λ) based on the third order function of Chl a_HPLC (gray lines; Fig. 12a and b) agreed well with the spectrophotometrically measured a_pig(λ) (black lines; Fig. 12a and b) during both surveys (R²=0.86; Fig. 12c). The R² for modeled versus measured a_pig(λ) are between 0.76 and 1.00 from 400 to 700 nm with averaged R² of whole spectra reaching ∼0.82 on 29 September 2017 and ∼0.89 on 29–30 October 2017, respectively. The vector coefficients $C=[C_{\mathrm{3}},C_{\mathrm{2}},C_{\mathrm{1}},C_{\mathrm{0}}]$ obtained from Eq. (10) were further applied to Eq. (11) to generate a_{pig_OLCI}(λ) based on OLCI-derived Chl a images on 6 July (Fig. 11a₉), 29 September (Fig. 10a), 29–30 October (Fig. 10b–c) and 25 November 2017 (Fig. 11a₁₅), respectively; these contained 259×224 pixels in each image. The a_{pig_OLCI}(λ) at each pixel was retrieved at 1 nm intervals, and thus 301 images of a_{pig_OLCI}(λ) representing each wavelength were obtained over GB.

Figure 12Spectrophotometrically measured and multiple-regression fitted a_pig(λ) spectra acquired on (a) 29 September and (b) 29–30 October 2017 in GB. Gray and black lines represent modeled and measured results, respectively. (c) Comparison between modeled and spectrophotometrically measured a_pig(λ) for all data with color representing wavelength.

Download

3.2.4 Accuracy of phytoplankton pigment retrievals from Sentinel-3A OLCI

The reconstructed a_{pig_OLCI}(λ) was spectrally decomposed into 16 individual pigment spectra at each pixel based on Eq. (13). A comparison of all data between HPLC-measured pigments and NNLS algorithm inverted pigments shows that R² ranged from a low of 0.40 for diatoxanthin to 0.96 for Chl a, and RMSE was in the range of 0.103–0.584 (Table 4). The NNLS-modeled Chl a also correlated well with OLCI-derived Chl a (R²=0.98; Fig. 13a), with each exhibiting similar quantitative and spatial patterns. For the other 15 simultaneously simulated pigments, R² of only seven pigments were greater than 0.650 (Table 4). In addition, the resulting RMSEs were less than 0.3 for most of pigments, except zeaxanthin, violaxanthin, diatoxanthin and diadinoxanthin. Further, for those pigments with relatively lower RMSE, their slopes were very close to 1 and y intercepts approached to 0. Five NNLS-derived versus HPLC-measured diagnostic pigments including alloxanthin, Chl b, zeaxanthin, fucoxanthin and peridinin are shown in Fig. 13. The R² between NNLS-derived and HPLC-measured pigments for surveys 1 and 2 was highest for alloxanthin (0.91; Fig. 13b). For the other pigments R² was 0.854 for Chl b (Fig. 13c), 0.689 for zeaxanthin (Fig. 13d), 0.645 for fucoxanthin (Fig. 13e) and 0.566 for peridinin (Fig. 13f).

Table 4Statistical results between HPLC-measured and NNLS-modeled pigments.

Download Print Version | Download XLSX

Figure 13Sentinel-3A OLCI-derived pigment concentrations against HPLC-measured pigment concentrations in Galveston Bay: (a) Chl a, (b) alloxanthin, (c) Chl b, (d) zeaxanthin, (e) fucoxanthin and (f) peridinin. Red and green symbols indicate data sets obtained on 29 September and 29–30 October 2017, respectively. The dashed circle includes the station located in algal bloom area.

Download

3.3 Spatiotemporal variations of diagnostic pigments

Flooding due to Hurricane Harvey not only enhanced Chl a but also affected the phytoplankton pigments composition. NNLS-retrieved pigment maps for July, September, October and November 2017 including those of alloxanthin, Chl b, zeaxanthin, fucoxanthin and peridinin (Fig. 14) show different levels of variations before and after the hurricane event. Alloxanthin, which is unique to cryptophytes (Wright and Jeffrey, 2006), exhibited the same spatial distribution patterns (Fig. 14a₁–e₁) with Chl a. Alloxanthin was especially low (∼0.5 µg L⁻¹, Fig. 14a₁) in the major basin area on 6 July 2017 before the hurricane and slightly elevated (∼0.7 µg L⁻¹, Fig. 14b₁) in September and October 2017 after the hurricane passage. Furthermore, extremely high alloxanthin (∼3.5 µg L⁻¹, Fig. 14c₁–d₁) was observed adjacent to San Jacinto River mouth on 29–30 October 2017, which coincided with the high %Chl a of cryptophytes at stations 19 and 23 (Fig. 6b). The bloom with high concentration of alloxanthin on 29 October 2017 (∼3.5 µg L⁻¹; Fig. 14c₁) then extended to a broader area on 30 October 2017 (Fig. 14d₁).

Figure 14Sentinel-3A OLCI-derived maps of diagnostic pigment concentrations for Galveston Bay. Simulated (a1–e1) alloxanthin, (a2–e2) Chl b, (a3–e3) zeaxanthin, (a4–e4) fucoxanthin and (a5–e5) peridinin concentrations. Panels (a), (b), (c), (d) and (e) represent columns (maps for 6 July, 29 September, 29–30 October and 25 November 2017), respectively and panels 1–5 represent rows (pigments). Panels (f), (g), (h) and (l) are the corresponding IFCB data for 6 July, 29 September, 29–30 October and 25 November 2017, respectively. Note that IFCB pictures of freshwater species including chlorophytes and cyanobacteria that appeared on 20–30 September 2017 have been zoomed in for better clarity.

Download

Chl b is abundant in the group of chlorophytes (green algae) (Hirata et al., 2011) and the spatial distributions of Chl b (Fig. 14a₂–e₂) also showed strong correlations with Chl a on 6 July 2017, 29 September, 29–30 October and 25 November 2017. The NNLS-derived Chl b exhibited overall low values (∼0.5–2 µg L⁻¹; Fig. 14a₂) before the hurricane and showed obvious elevation throughout the bay after the hurricane, eventually decreasing to pre-hurricane level by 25 November 2017. Furthermore, Chl b concentrations observed on 29 September 2017 were higher than that on 29–30 October 2017, which corresponded to a decline of the chlorophyte percentage derived from the IOP inversion algorithm (Fig. 6). More importantly, images obtained from IFCB at the entrance to GB also detected a freshwater Chlorophyte species (Pediastrum duplex; Fig. 14g) on 29 September 2017. However, this species was rarely observed in IFCB images for the other dates (Fig. 14a₁ and Fig. 14c₁–e₂). In addition, Chl b concentrations approached ∼2.8 µg L⁻¹ in the bloom area and the corresponding green discoloration of water was also observed during the field survey on 30 October 2017.

Zeaxanthin is known as taxa-specific pigment for prokaryotes (cyanobacteria) (Moisan et al., 2017; Dorado et al., 2015). NNLS-derived zeaxanthin maps (Fig. 14a₃–e₃) displays significantly different patterns to Chl a concentrations, exhibiting low concentrations in the areas where the Chl a concentrations were high. For example, zeaxanthin was especially low in the bloom area on 29–30 October 2017, which agreed well with low %Chl a of cyanobacteria at stations 19 and 23 (Fig. 6), thus indicating that this localized algal bloom event was not associated with cyanobacteria. In addition, the zeaxanthin concentration was high ∼3.0 µg L⁻¹ (Fig. 14a₃) in both GB and shelf waters on July 6 July 2017 before the hurricane event. Later, zeaxanthin increased slightly on 29 September 2017 (Fig. 14b₃) with IFCB data detecting N₂-fixing cyanobacteria (Anabaena spp.; Fig. 14g) and remained elevated on 29–30 October 2017 (Fig. 14b₃–c₃). Zeaxanthin eventually decreased to very low values (∼1.2 µg L⁻¹; Fig. 14e₃) on 25 November 2017.

Fucoxanthin is a major carotenoid found in diatoms (Hirata et al., 2011; Moisan et al., 2017) and the NNLS-derived fucoxanthin maps (Fig. 14a₄–e₄) show highly similar distribution patterns with Chl a. Maps of fucoxanthin show low concentrations on 6 July 2017 (∼1.5 µg L⁻¹; Fig. 11a₄), and display a large increase on 29 September 2017 (∼1.6–3.0 µg L⁻¹; Fig. 11b₄). The diatoms group detected by the IFCB were dominated by marine species before the hurricane but subsequently shifted to freshwater species (e.g., Pleurosigma; Fig. 14g) and then back to marine species after October 2017. Overall, fucoxanthin concentrations in GB were relatively higher during survey 1, which corresponded to the higher %Chl a of diatoms (Fig. 6) compared to survey 2. Although, fucoxanthin decreased to low values on 25 November 2017 (∼1.6 µg L⁻¹; Fig. 11e₄), it accounted for a higher fraction of phytoplankton diagnostic pigments compared to other dates in July, September and October 2017.

Peridinin, a primary biomarker pigment for certain dinoflagellates (Örnólfsdóttir et al., 2003), also displayed significantly distinct patterns in comparison to Chl a (Fig. 14a₅–e₅). On 6 July 2017, peridinin was ∼0.24–0.36 µg L⁻¹, accounting for a high proportion of the diagnostic pigments; meanwhile, diversity of marine dinoflagellate species observed by the IFCB at this time was also high (Fig. 14f). However, peridinin decreased (∼0.001–0.05 µg L⁻¹) after the hurricane, with freshwater dinoflagellate species (Ceratium hirundinella; Fig. 14g) detected by the IFCB on 29 September 2017. In addition, maps of peridinin during both surveys (Fig. 14b₅–d₅) present a higher concentration (∼0.3 µg L⁻¹) in higher salinity waters adjacent to the bay entrance, which agreed well with the increasing fraction of dinoflagellate at stations 10–14 detected from IOP inversion model (Fig. 6). In contrast, peridinin was found in low concentrations in both GB and shelf waters on 25 November 2017 (Fig. 14e₅), with dinoflagellate species rarely observed by the IFCB (Fig. 14l).

4.1 Performance of the semi-analytical IOP inversion algorithm

The residuals between R_{rs_insitu} and R_{rs_mod} on 29 September and 29–30 October 2017 are negative in the blue (400–450 nm) and red (610–630 nm) spectral range at most stations, whilst keeping positive ∼700 nm, which could be attributed to a number of factors. First, the underestimation near 700 nm by the IOP inversion model is possibly induced by the absence of a fluorescence component in the IOP inversion model; thus, R_{rs_insitu} containing fluorescence signals were generally higher than R_{rs_mod} near 700 nm. Second, in the range of 610–630 nm, the absorption was overestimated at most of the stations; in this spectral range, the shape of spectra was strongly modulated by cyanobacteria absorption. Thus, this overestimation at ∼620 nm is likely introduced by the input absorption spectrum (eigenvector) for cyanobacteria since all of the $a_{\mathrm{phi}}^{*}\left(\mathit{\lambda }\right)$ inputs are general absorption spectral shapes for different phytoplankton groups. However, the spectra of $a_{\mathrm{phi}}^{*}\left(\mathit{\lambda }\right)$ can vary in magnitude and shape associated with package effects under different environmental conditions (e.g., nutrient, light and temperature) even for the same species (Bricaud et al., 2004). More detailed absorption spectra of phytoplankton under different conditions (e.g., high or low light and replete or poor nutrients) could improve the performance of the IOP algorithm. Furthermore, the role of scattering might be another key factor to explain differences between R_{rs_insitu} and R_{rs_mod} for the whole spectra. The quantity and composition of suspended materials including phytoplankton, sediment and minerals will collectively determine b_bp (λ) in both shape and magnitude. However, the input eigenvector of b_bp (λ) in the present study was not divided into detailed sub-constituents and was a sum spectrum-based on a power-law function (Table 2). In reality, b_bp (λ) spectra are not smooth and regular, and thus the b_bp (λ) values of phytoplankton and sediment might introduce errors to the whole spectrum due to their own scattering characteristics.

4.2 Distributions of NNLS-retrieved phytoplankton pigments from Sentinel-3A OLCI

The NNLS-inversion algorithm showed relatively higher R² for those pigments that better correlated with HPLC-measured Chl a (e.g., Chl b and alloxanthin), which was reasonably consistent with Moisan et al. (2017). This outcome could potentially be attributed to the fact that the NNLS-pigment inversion algorithm was developed based on the relationship between HPLC-measured Chl a and spectrophotometer-measured a_pig(λ). For instance, pigments that were relatively poorly correlated with HPLC-measured Chl a, such as fucoxanthin, diatoxanthin and diadinoxanthin on 29–30 October 2017, the OLCI-derived concentrations in the cryptophyte–chlorophyte algal bloom area showed higher concentrations than those of HPLC measurements (e.g., values in gray circle; Fig. 13e), thus resulting in lower R². However in previous studies (Moisan et al., 2017; Pan et al., 2010), relatively higher R² versus lower RMSE were observed between satellite-derived and HPLC-measured fucoxanthin compared to this study. This was likely because in addition to being based on long-term measurements, fucoxanthin is one of the most abundant diatom biomarker pigments in coastal waters and correlated very well with Chl a in their study area (United States northeast coast). In contrast, the cryptophyte–chlorophyte algal bloom area with extremely high Chl a appeared to disturb the correlations between Chl a and fucoxanthin in this study. Also, pigments with apparent high values in algal bloom areas, such as Chl b, Chl c, alloxanthin, lutein, showed higher R² with RMSE less than 0.3. Thus, in situ measurements of Chl a and a_pig(λ) in waters with stronger gradients in magnitude and greater variations in phytoplankton community structures could potentially increase the challenge of applying NNLS-inversion algorithms in optically complex estuarine waters. Further, the highly dynamic estuarine environment could contribute as well to additional uncertainties in the validation of inverted pigments due to variations such as turbulence, turbidity or light field that are likely to occur during the time interval between in situ and Sentinel-3 OLCI (∼4 h) observations. HPLC measurements also cannot detect extremely low pigment concentrations; for example, HPLC-measured peridinin concentrations were 0.001 µg L⁻¹ at several stations, however, OLCI-derived peridinin showed higher and variable values at these stations (data in gray circles; Fig. 13f), thus this could increase the RMSE of the NNLS-inverted peridinin. It was also found that the slopes of all pigments were smaller than 1, which demonstrate that NNLS-inverted pigments were relatively smaller than HPLC measurements, especially for those stations located in the algal bloom area; this could most likely be attributed to the underestimation of Chl a values by the Sentinel-3 OLCI empirical algorithms in the algal bloom area. Algal bloom dominated by the cryptophytes group, which is also known to cause red tides worldwide, to some degree, could increase red reflectance and thus increase ratio values of red ∕ NIR and decrease estimated Chl a values. Therefore, reliable estimates of satellite-detected Chl a is crucial for the accuracy of retrieved pigments. The goal of the empirical Chl a algorithm for Sentinel-3A OLCI is to obtain a more accurate estimation of surface Chl a concentration, which is better for retrieving other accessory pigments. However, the primary limitation of Chl a empirical algorithms in this study was that the derived relationships between red ∕ NIR and Chl a in GB may only be valid within a specific time period due to temporally limited field observations versus highly dynamic estuarine environments. Therefore, a Chl a empirical algorithm that is more broadly applicable over a longer time period will largely improve the accuracy of retrieved pigments over a series of remote-sensing images and can be more useful for spatiotemporal studies of phytoplankton functional diversity. More importantly, the highly similar absorption spectra of many carotenoids are another key issue limiting the accuracy of spectral decomposition techniques. Although the 16 input pigment spectra used in this study were selected from Thrane et al., 2015), which were correctly identified from unknown phytoplankton community structure with low error rate reported from Monte Carlo tests, the potential effects of aliasing spectra of some pigment pairs (e.g., fucoxanthin vs. peridinin, diadinoxanthin vs. lutein, β-carotene vs. zeaxanthin) could still be a factor. Thus, the reported errors or R² for the retrieved total carotenoids in Thrane et al. (2015) were apparently lower or higher, respectively, than those of modeled total chlorophylls, which showed consistency with this study. Although the predicted pigments showed a range of R² and RMSE with known uncertainties, all are within the acceptable range and could be useful for studying the spatiotemporal responses of PFTs to environmental variations, especially in such optically complex estuaries.

The derived maps of phytoplankton diagnostic pigments appeared to be reasonably correlated with HPLC-measured diagnostic pigments and showed overall agreement with extracted phytoplankton taxonomic compositions detected from the IOP inversion algorithm. The retrieved diatom-specific fucoxanthin maps, however, show high concentrations compared to other pigments adjacent to the entrance (Fig. 13b₄ and c₄), which contradicts the diatom %Chl a calculated from the IOP inversion algorithm. As the Chl a fraction of diatoms was relatively uniform at stations 12–14 (Fig. 6b) Nair et al.(2008) concluded that fucoxanthin can occur in other phytoplankton types (e.g., raphidophyte and haptophyte). Fucoxanthin and/or fucoxanthin derivatives such as 19^′-hexanoyloxyfucoxanthin can also replace peridinin as the major carotenoid in some dinoflagellates (e.g., Karenia brevis; Jeffrey and Vest, 1997). The elevated contributions from groups of dinoflagellates, haptophytes and prochlorophytes adjacent to the entrance (stations 10–14; Fig. 6b) along with high concentrations of fucoxanthin likely suggest the presence of elevated fractions of haptophytes and dinoflagellates, and further implies that fucoxanthin is an ambiguous marker pigment for diatoms. This could also explain the poor correlation between inverted %Chl a and %DP observed for the groups of diatoms and haptophytes (Fig. 4g and l). These results also further suggest the inherent limitations of using a DP-type comparison between major biomarker pigments and phytoplankton groups because the major assumption for DP-type methods is that the diagnostic pigment of distinct phytoplankton groups are uncorrelated to each other. This assumption is invalid in that concentrations of major biomarker pigments are significantly correlated with each other and also may vary in time and space under some external environmental stress (e.g., temperature, salinity, mixing, light and nutrient) (Latasa and Bidigare, 1998).

4.3 Response of phytoplankton taxa to environmental conditions

Previous studies showed diatoms to be the most abundant taxa in GB, and they tend to be more dominant during winter and spring, corresponding to periods of high freshwater discharge and nutrient-replete conditions (Dorado et al., 2015; Örnólfsdóttir et al., 2004a). Transitions from chain-forming diatoms such as Chaetoceros and rod-like diatoms pre-flood to small cells, such as Thalassiosira and small pennate diatoms were generally observed during high-river discharge periods (Anglès et al., 2015; Lee, 2017). In contrast, cyanobacteria were the most abundant species during the warmer months (June–August) when river discharge was relatively low (Örnólfsdóttir et al., 2004b). Further, phytoplankton groups in GB responded differentially both taxonomically and spatially to the freshening events due to their contrasting nutrient requirements and specific growth characteristics. For instance, most phytoplankton taxa (e.g., diatom, chlorophyte and cryptophyte) can be positively stimulated by fresh inflows due to their relatively rapid growth rate (Paerl et al., 2003). However, Roelke et al. (2013) also documented that cyanobacteria and haptophytes in the upper GB were not sensitive to nutrient-rich waters from both rivers due to the extra nutrients obtained from N₂-fixation abilities and mixotrophic characteristics, respectively. In the lower part of GB, dinoflagellates and cyanobacteria are known to be more dominant during the low-river discharge due to their preference for higher phosphorus (P) compared to some other groups, and to low turbulence (Lee, 2017), and thus are generally inversely related to the fresh inflows (Lee, 2017; Roelke et al., 2013).

Perturbations following Hurricane Harvey affected the phytoplankton taxonomic composition with alterations in phytoplankton community structure observed as the GB system transitioned from marine to freshwater then to marine system (Figs. 6 and 14). Higher fractions of zeaxanthin and peridinin and the presence of large and slow-growing marine dinoflagellates detected by the IFCB pre-hurricane (6 July 2017) indicate that both cyanobacteria and dinoflagellates were the main groups of the phytoplankton community during summer, and were likely associated with warmer temperature and lower river flow (Lee, 2017). Later, massive Chl a observed in September 2017 and the decline of Chl a to background state in October 2017 were likely associated with the hurricane-induced high-river discharge and the resulting variations in nutrient concentration and composition. Higher fractions of diatoms and chlorophytes accompanied by increasing fucoxanthin and Chl b on 29 September 2017 to some extent agreed well with measurements of Steichen et al. (2018) in the two weeks following Hurricane Harvey that freshwater species (diatoms, green algae and cyanobacteria) appeared immediately following the flooding event. The greater abundance of diatoms and chlorophytes during survey 1 in comparison to survey 2 were likely due to their rapid growth rates, enhanced nutrient uptake rates, and tolerance of low salinity and high turbulence under high nutrient loading conditions following the freshwater inflows (Roy et al., 2013; Santschi, 1995). Therefore, it is not surprising that Chl b concentrations showed very low values in July and November, 2017, when river discharge was correspondingly low. Cyanobacteria, which normally prefer low salinity conditions, also showed specific responses to this flood event. On 29 September 2017, zeaxanthin slightly increased compared to summer season in July 2017. The decline of diatoms and chlorophytes versus slightly increased cyanobacteria levels observed on 29–30 October 2017 could be attributed to the relatively slow growth rates of cyanobacteria compared to that of chlorophytes and diatoms (Paerl et al., 2003); cyanobacteria appeared to have lagged behind these groups in terms of responding to enhanced freshwater discharge when longer residence times were again restored. In contrast, the presence of green algae and cyanobacteria could as well be explained by the clarity and turbidity gradient of water. Quigg et al. (2010) reported that when turbidity was relatively high, chlorophytes dominated over cyanobacteria with biomass ratio of chlorophyte/cyanobacteria greater than two, which supported our observations that chlorophyte dropped off whilst cyanobacteria increased during survey 2 on 29–30 October 2017. In addition the highest cyanobacteria percentage in East Bay suggests that calm and stratified waters may accelerate cyanobacteria growth as the buoyancy regulation mechanism of cyanobacteria is possibly restricted by the water mixing (Roy et al., 2013). The peridinin concentration, which initially decreased in September and then increased in the lower GB on 29–30 October 2017, suggests that dinoflagellates show overall preference for high-salinity waters. Furthermore, previous IFCB observations from Biological and Chemical Oceanography Data Management Office (BCO-DMO) showed that algal blooms after hurricanes in the nGOM were initially dominated by diatoms, and subsequently transitioned to blooms of dinoflagellates, likely associated with nutrient ratios and chemical forms of nutrients supplied by the flood waters and rainfall (Heisler et al., 2008). In addition, high concentrations of peridinin observed along the Houston Ship Channel might provide evidence that the ballast water addition from shipping vessels likely promotes harmful species of dinoflagellates (Steichen et al., 2015). Finally, low concentrations of all pigments on 25 November 2017 with relatively higher fractions of fucoxanthin compared to previous dates (Fig. 14), indicate the major role of marine diatoms at that time and further confirms that diatoms can be found under a wide range of inflows in GB.

The localized cryptophyte–chlorophyte bloom that occurred ∼60 d after Hurricane Harvey on 29–30 October 2017 was captured by both satellite and in situ measurements. This bloom might not be associated with the flooding events of Hurricane Harvey, and could be linked to nutrient-rich runoff flowing into GB, reflecting the sensitivity and rapid response of the phytoplankton community to nutrient input in GB. In shallow and turbid estuaries, human activities are altering the environment and causing phytoplankton changes in diversity and biomass to occur more frequently. Dugdale et al. (2012) reported that variations of the phytoplankton community in the San Francisco estuary could be attributed to anthropogenically elevated concentration of ammonium, which restrains the uptake of nitrate, thus reducing the growth and reproduction of larger diatoms and shifting towards smaller species (e.g., cryptophytes and green flagellates). Furthermore, “pink oyster” events related to alloxanthin of cryptophytes in GB occurred more frequently from September through October in recent years (Paerl et al., 2003). The eastern side of the Houston Ship Channel in the mid-bay region was reported as the area most heavily impacted by the intense “pink oyster” events. Previous studies and present observations both suggest that this cryptophyte–chlorophyte-dominated bloom could be promoted by the nutrient-driven eutrophication from the Houston Ship Channel, urbanization and industrialization along the upper San Jacinto River complex.

4.4 Photo-physiological state of natural phytoplankton community

In this study, the CDOM-corrected F_v∕F_m and σ_PSII likely represented a composite of both phytoplankton taxonomy and physiological stress (e.g., nutrient and mixing). Typically, lowest N and P concentrations were measured closest to the nGOM (Quigg et al., 2009). Phytoplankton communities living close to nGOM were usually in poor nutrient conditions and would have been expected to maximize their light harvesting (increase in σ_PSII) due to nutrient stress. Simultaneously, phytoplankton cells might experience a decline of functional proportion of reaction centers of PSII (RCII), which means a decrease in F_v∕F_m. The observed low levels of F_v∕F_m and Chl a ∕ TP versus high values of σ_PSII and AP ∕ TP adjacent to the nGOM showed agreement with previous studies that the fraction of carotenoids is higher for nutrient-poor cultures (Schitüter et al., 1997; Holmboe et al., 1999). In contrast, phytoplankton in well-mixed waters (stations 7–9) might experience abundant nutrients due to the resuspension of the cyclonic gyre around Smith Point; as such, their photosynthetic machinery were likely healthier. Aiken et al. (2004) documented that the Chl a ∕ TP ratio was relatively higher when plants were in good growing conditions, which is similar to the observations in this study that phytoplankton have a higher fraction of Chl a accompanying higher rate of photosynthetic efficiency (F_v∕F_m) under nutrient replete conditions. Overall, the spatial pattern of F_v∕F_m and σ_PSII in GB could be mainly attributed to physiological stress of nutrient and hydrodynamic conditions since the light availability (PAR) during the sampling period did not spatially vary significantly at the surface. Furthermore, FIRe measurements (F_v∕F_m and σ_PSII) also presented a taxonomic signal superimposed upon environmental factors. Each cluster with different dominant taxa (well-mixed group, chlorophyte–cryptophyte, cyanobacteria and dinoflagellate–haptophyte) displayed different physiological characteristics. The taxonomic sequence of eukaryotic groups from high F_v∕F_m, low σ_PSII to low F_v∕F_m, high σ_PSII in the present observations showed potential effects of phytoplankton cell size corresponding to diatoms, chlorophytes, cryptophytes, dinoflagellates and haptophytes. The prokaryote (cyanobacteria) had relatively high values of F_v∕F_m and low values of σ_PSII; this agreed with F_v∕F_m for some species of nitrogen-fixing cyanobacteria that can range from 0.6 to 0.65 (Berman-Frank et al., 2007). Yet, it is difficult to separate the contributions from environmental factors and taxonomic variations to the changes of FIRe fluorescence signals since all these parameters are interrelated. Different phytoplankton groups and sizes will display distinct physiological traits (F_v∕F_m and σ_PSII) when experiencing considerable environmental pressures. Thus, effects of physiological stress on F_v∕F_m and σ_PSII variations for natural samples can only be determined when taxonomic composition can be excluded as a contributor (Suggett et al., 2009).