2.1 Physical, sediment and biogeochemical models

Our model covered the entire GoM (Fig. 1a) and was built on the Coupled Ocean–Atmosphere–Wave–Sediment Transport (COAWST) modeling system (Warner et al., 2008, 2010). COAWST is an open-source model platform that consists of three numerical models, namely the Weather Research and Forecasting (WRF) model (Skamarock et al., 2008), the Regional Ocean Modeling System (ROMS; Shchepetkin and McWilliams, 2005; Haidvogel et al., 2008) and the Simulating WAves Nearshore (SWAN) model (Booij et al., 1999). The Community Sediment Transport Modeling System (CSTMS) is included in ROMS to simulate sediment transport, stratigraphy and geomorphology. The model coupling toolkit (MCT; Jacob et al., 2005) enables the interaction among these three models. The details of the model setup and validation of the three-way coupled hydrodynamic-sediment transport model (WRF–ROMS–SWAN–CSTMS) were described in Zang et al. (2018), where four types of sediment (two cohesive and two noncohesive) were defined with different grain diameters and settling velocities. There were 40 sediment layers on the sea floor with a total thickness of 1 m to resolve sediment bed erosion and deposition. The driving force of sediment resuspension was determined by bottom shear stress induced by wave and current. Readers are referred to Zang et al. (2018) for a detailed hydrodynamic and sediment validation.

Figure 1Model domains applied in this study. The entire panel (a) is the WRF model domain (6 km resolution) overlaid with water depth (colored shading). The black box represents the model grid used by ROMS and SWAN, with a 5 km resolution. The black box with a dashed line (27–31^∘ N, 94–86^∘ W) covers the northern Gulf of Mexico (nGoM). More details on the nGoM are shown in (b). The thick purple and red lines indicate the locations of 50 m isobath transect and transect D (Rabalais et al., 2001), respectively.

Given the importance of diatom in the phytoplankton community in the nGoM (Zhao and Quigg, 2014), it is necessary to have both nitrogen and silicon cycles in the model. The biogeochemical model in this study was largely built on the North Pacific Ecosystem Model for Understanding Regional Oceanography (NEMURO; Kishi et al., 2007), which incorporated both nitrogen and silicon flows. There were 11 state variables included in the model, namely nitrate, ammonium, two types of phytoplankton (small and large), three types of zooplankton (microzooplankton, mesozooplankton and predatory zooplankton), particulate and dissolved nitrogen, particulate silica, and silicic acid concentration. River nutrient discharge during the hurricane was retrieved from the United States Geological Survey (USGS) Water Data for the Nation website (http://nwis.waterdata.usgs.gov; station no. 07374000). The growth of phytoplankton was driven by water temperature, light availability and nutrient concentration. Instantaneous remineralization of particulate organic nitrogen at the bottom was estimated following Fennel et al. (2006). Our model did not include phosphate because its limitation on primary production in the nGoM was mainly between May and July (Laurent et al., 2012; Laurent and Fennel, 2014). We incorporated two types of chlorophyll corresponding to the large and small phytoplankton tracers, respectively. Following Fennel et al. (2006), chlorophyll dynamics was derived from a phytoplankton equation by multiplying the ratio of chlorophyll to phytoplankton biomass. To obtain an ideal parameterization set and a stable initial condition for the biogeochemical variables, we first conducted a 20 yr (1993–2012) coupled physical–biogeochemical simulation using only the ROMS model, where WRF and SWAN were disabled to achieve a feasible computation load (step 1 in Fig. 2). The atmospheric forcing was provided by the 6 h, 38 km horizontal resolution Climate Forecast System Reanalysis (CFSR; Saha et al., 2010, 2011; http://cfs.ncep.noaa.gov). The physical setup of the 20 year simulation was the same as Zang et al. (2019). The biogeochemical parameterizations (Table S1 in the Supplement) were largely adapted after a recent GoM biogeochemical modeling study by Gomez et al. (2018). Since this study focused on the response of the biogeochemical process to hurricane events, details of the 20 year simulation setup and model observation comparison were provided in the Supplement. Once validated, the biogeochemical variables were extracted from the 20 year model on 30 August 2008 as the initial condition for this Gustav simulation (Step 2 in Fig. 2).

Figure 2Flow chart of long-term (20 years) and hurricane (11 d) simulations. In step 1 we only run ocean (ROMS) and biogeochemical (NEMURO) models, which provide initial inputs for the next step. Step 2 couples ocean (ROMS), wave (SWAN), atmosphere (WRF), sediment (CSTMS) and new biogeochemical (NEMURO) models with the new sediment-induced light attenuation term.

Download

The light available for photosynthesis (I) is estimated using the following equation (McSweeney et al., 2017):

where I₀ is the light intensity at the surface layer, and Z is the water depth. par is the fraction of light available for photosynthesis (specified as 0.43). α_w and α_chl are the light attenuation coefficients of sea water and chlorophyll, respectively. PSn and PLn represent concentrations of small and large phytoplankton. Compared with the original biogeochemical model, we added a new sediment-induced light attenuation term in this equation. α_sed is the light attenuation coefficient due to suspended sediment, and SSC is total suspended sediment concentration in the respective layer. We performed a benchmark run (α_sed=0.059; McSweeney et al., 2017) to represent the scenarios with sediment-induced light attenuation. The simulation period was from 30 August to 10 September 2008.

2.2 Sensitivity tests

High turbidity in the Mississippi River delta due to fluvial sediment discharge and resuspension suggested the vital role of sediment in the underwater optical environment. To quantitatively evaluate the importance of suspended sediment in light attenuation, we conducted a sensitivity test (test 1) without sediment-induced light attenuation (α_sed=0). Since the physical properties of a sediment particle (e.g., size, shape, roughness and color) determine its light attenuation efficiency (Baker and Lavelle, 1984; Storlazzi et al., 2015), a wide range of α_sed (0.025–0.075) has been reported in previous studies (e.g., Pennock, 1985; Van Duin et al., 2001; Arndt et al., 2007; McSweeney et al., 2017). Here, we increased or decreased the benchmark α_sed (0.059) by 20 % and 40 % to examine the sensitivity of primary production to sediment-induced light attenuation (tests 2–5). The rest of the model setup was the same between the benchmark run and sensitivity tests (tests 1–5). The deviation due to the chaotic nature of turbulence was not considered in this study.

4.1 Temporal variability of biogeochemical variables

To examine the temporal variation in biogeochemical variables during the passage of Gustav, we plotted the time series of spatially averaged net primary production (NPP; growth of phytoplankton minus the respiratory losses), surface chlorophyll concentration, surface NO₃ concentration, SSC, downward solar shortwave radiation and sea surface temperature (SST) over the nGoM inner shelf (Fig. 5; < 50 m water depth). NPP exhibited strong diel variation, and the peaks were strongly correlated with shortwave radiation maximum (Fig. 5a and e). Such a diel cycle could also be found in chlorophyll concentration but with a 3 to 4 h delay (Fig. 5b). Before the arrival of Gustav, daily averaged NPP was around 0.05 g C m⁻² h⁻¹, and the differences in NPP and chlorophyll concentration between the benchmark run and test 1 were minor (Fig. 5a and b).

Following Gustav's landfall along coastal Louisiana at 16:00 Coordinated Universal Time (UTC) on 1 September, surface SSC increased to 3.8 kg m⁻³ because of strong seabed resuspension (Fig. 5d). Daily averaged NPP reduced to 0.03 $\mathrm{g}\mathrm{C}({\mathrm{m}}^{\mathrm{2}}{\mathrm{h}}^{-\mathrm{1}}{)}^{-\mathrm{1}}$ in test 1. Once sediment-induced light attenuation was included, daily averaged NPP further declined to 0.01 $\mathrm{g}\mathrm{C}({\mathrm{m}}^{\mathrm{2}}{\mathrm{h}}^{-\mathrm{1}}{)}^{-\mathrm{1}}$, suggesting that light availability severely limited short-term productivity on the inner shelf. Chlorophyll concentrations in the benchmark run and test 1 were reduced by 40 % as Gustav approached. Hurricane-related surface cooling, together with decreased light (Fig. 5e and f), contributed to the reductions in both chlorophyll and NPP.

Figure 4Simulated 5 d composite (5–9 September) of surface chlorophyll concentration after Gustav compared to the corresponding SeaWiFS-derived surface chlorophyll results over the nGoM inner shelf (h < 50 m) for model results based on (a) benchmark (α_sed=0.059) and (b) test 1 (α_sed=0) runs.

Download

Figure 5Time series of spatially averaged (inner shelf; water depth < 50 m) net primary production (a), surface chlorophyll concentration (b), surface NO₃ concentration (c), surface suspended sediment concentration (d), solar shortwave radiation (e) and sea surface temperature (f). In (a–c), blue represents the benchmark run (α_sed=0.059) and red represents test 1 (α_sed=0). Dots in (a) are the daily averaged net primary production. The black dashed line shows the time of the Gustav landfall.

Download

The difference in daily averaged NPP between the benchmark run and test 1 maximized on 2 September due to light limitation modulated by resuspended sediment (Fig. 5a and d). On 3 September, daily averaged NPP of test 1 recovered to 0.04 $\mathrm{g}\mathrm{C}({\mathrm{m}}^{\mathrm{2}}{\mathrm{h}}^{-\mathrm{1}}{)}^{-\mathrm{1}}$ and remained steady through the end of our simulation (Fig. 5a). For the benchmark run, however, the recovery of NPP was much slower; daily averaged NPP was lower than that of test 1 until 7 September when most of the suspended sediment settled back onto the seabed. The NO₃ concentration went up gradually between 2 and 7 September in the benchmark run (Fig. 5c) as the nutrient consumption was constrained by the decline in photosynthetic activity. Accumulated NO₃, together with the improved optical environment due to low SSC, resulted in higher NPP and algal bloom after 7 September (Fig. 5a and b).

4.2 Vertical structure of biogeochemical variables

We extracted concentrations of chlorophyll, NO₃, sediment and water density anomaly along transect D in Rabalais et al. (2001; see Fig. 1b for transect D location) at three time points (31 August, 2 September and 10 September) to represent pre-, during and posthurricane stages, respectively (Figs. 6 and 7). Before the approach of Gustav, offshore water was well stratified (Fig. 7d). Chlorophyll concentration decreased seaward from 5 to 0.3 mg m⁻³ (Fig. 6a and d). Sediment-induced light attenuation did not alter the vertical structure of the chlorophyll and NO₃ much (Fig. 6a, d, g and j) owing to low SSC in the water column (Fig. 7a). On 2 September, strong vertical mixing increased the SSC to more than 1 kg m⁻³ over the entire water column (Fig. 7b and e). Chlorophyll concentration in waters < 40 m in the benchmark run was ∼ 4 mg m⁻³, lower than that in test 1 due to sediment-induced light attenuation (Fig. 6b and e). Higher NO₃ concentration in the benchmark run was a result of the weakened primary production and nutrient consumption (Fig. 6h and k). The most striking differences in chlorophyll and NO₃ between the two simulations were in water shallower than 20 m.

Figure 6Model-simulated chlorophyll and NO₃ along transect D on 31 August (first column), 2 September (second column) and 10 September (third column). (a–f) Chlorophyll concentration of the test 1 and benchmark run, respectively (note that the color scale is different from Fig. 3). (g–l) NO₃ concentration of test 1 and benchmark run, respectively.

Download

Figure 7(a–c) Model-simulated suspended sediment concentration (SSC) and (d–f) water density anomaly along transect D on 31 August (a, d), 2 September (b, e) and 10 September (c, f), respectively.

Download

In test 1, chlorophyll concentration during the posthurricane stage was lower than that of the prehurricane stage (Fig. 6a and c), in contrast to the condition captured by satellite imagery (Fig. 3b and c). The benchmark run, however, successfully reproduced the magnitude and seaward extension of the posthurricane bloom (Fig. 6f). High chlorophyll concentration (> 1 mg m⁻³; Fig. 6f) with low NO₃ (Fig. 6l) was simulated in the top 20 m of the water column where stratification partially recovered (Fig. 7f) and sediment concentration was low after the passage of Gustav (Fig. 7c). At water deeper than 20 m, the chlorophyll concentration dropped drastically to less than 0.1 mg m⁻³, while the NO₃ concentration further increased to > 1 mmol m⁻³. The synchronized high turbidity and low chlorophyll concentration implied that, 9 d after Gustav's landfall, the primary production in deeper water could still be constrained by light availability. A similar vertical structure (high SSC and low chlorophyll at the bottom) was also simulated in the Delaware estuary, where near-bottom productivity was constrained by the estuarine turbidity maximum (McSweeney et al., 2017). Such a stratified water column with high and/or low productivity at the surface and/or bottom is generally favorable for bottom oxygen depletion. The elevated surface phytoplankton growth following the hurricane could thus result in increased particulate organic matter (POM) whose remineralization contributes to bottom water hypoxia (Wiseman et al., 1997). Meanwhile, the posthurricane stratification recovery in the summer and fall seasons would have likely prevented oxygen ventilation to the bottom. The high respiration rate caused by resuspended POM could further lower the oxygen level (Bianucci et al., 2018). McCarthy et al. (2013) reported a post-Gustav respiration peak associated with organic matter resuspension in the bottom boundary layer. A recent numerical model study also simulated a substantial increase in near-bottom oxygen consumption due to resuspended POM remineralization during moderate resuspension events (Moriarty et al., 2018). These past studies and the new finding of this study suggest that particulate matter (both organic and inorganic) dynamics might substantially contribute to bottom oxygen depletion and hypoxia development following a hurricane passage. More in situ observations of oxygen dynamics in the bottom boundary layer are needed.

4.3 The posthurricane offshore bloom

Posthurricane blooms have been widely observed in the mid- and low-latitude oceans (Davis and Yan, 2004; Miller et al., 2006; Pan et al., 2017; D'Sa et al., 2019). A bloom in the open ocean was usually isolated and patchy, and its formation was mainly related to nutrients supplied from deep waters via vertical mixing (Pan et al., 2017; Walker et al., 2005). The mechanism of the bloom formation on the outer shelf, however, was more complex due to possible impacts from the inner shelf water. Strong post-Gustav cross-shelf transport has been reported by previous studies (Korobkin et al., 2009; Zang et al., 2018). The seaward dispersal of higher nutrient and chlorophyll coastal waters could have potentially contributed to the outer shelf bloom, but their respective contributions remained unclear. To quantify the cross-shore exported nutrient and chlorophyll, we estimated depth-integrated offshore (seaward) NO₃ and chlorophyll flux along the 50 m isobath transect (Fig. 1b; Table 1). Compared with test 1 (NO₃ – 7.35 $\mathrm{mmol}\mathrm{N}(\mathrm{m}{\mathrm{s}}^{-\mathrm{1}}{)}^{-\mathrm{1}}$; chlorophyll – 66.88 $\mathrm{mg}(\mathrm{m}{\mathrm{s}}^{-\mathrm{1}}{)}^{-\mathrm{1}}$), the benchmark run estimated a higher NO₃ flux (38.71 $\mathrm{mmol}\mathrm{N}(\mathrm{m}{\mathrm{s}}^{-\mathrm{1}}{)}^{-\mathrm{1}}$) and a lower chlorophyll flux (43.10 $\mathrm{mg}(\mathrm{m}{\mathrm{s}}^{-\mathrm{1}}{)}^{-\mathrm{1}}$). The differences in NO₃ and chlorophyll fluxes between the two simulations could be explained by nutrient accumulation and NPP reduction on the inner shelf associated with resuspended sediment (Fig. 5a and c). Given the better offshore bloom intensity reproduced by the benchmark run (Fig. 3d and e), we conclude that the posthurricane offshore bloom was mainly triggered by nutrients exported from the inner shelf water.

Table 1Offshore fluxes of NO₃ and chlorophyll along the 50 m isobath transect (see location in Fig. 1b).

Download Print Version | Download XLSX

4.4 Sensitivity to sediment light attenuation coefficient (α_sed)

A wide range of particle physical properties (e.g., size, shape, roughness and color) influence sediment light attenuation efficiency, which contributes to the difficulty in the parameterization of α_sed over a large region such as the nGoM (Baker and Lavelle, 1984; Storlazzi et al., 2015). To examine the sensitivity of primary production to sediment light attenuation efficiency, the results of sensitivity tests with different α_sed (tests 2–5) were compared against the benchmark run.

Ahead of Gustav's landfall, the difference in primary production between the benchmark run and sensitivity tests was limited (Fig. 8a), which suggested that the nGoM ecosystem was mainly limited by nutrients rather than light (Fennel et al., 2011). A total of 2 d after the landfall (1–3 September), high SSC suppressed photosynthesis in the entire water column, which overwhelmed the response associated with different α_sed settings. As such, primary production was not sensitive to α_sed from 1 to 3 September, although the nGoM ecosystem was also light limited. After 3 September, the differences in primary production and NO₃ concentration increased among the sensitivity tests through 8 September (Fig. 8). Primary production became more sensitive to α_sed than SSC, which largely decreased due to settling (Fig. 5d). In the last 2 d of our simulation, the primary production differences reduced again to prehurricane conditions as the nGoM ecosystem shifted back to a nutrient-limited system.

Figure 8Comparison of spatially averaged (inner shelf; water depth < 50 m) net primary production (a) and NO₃ concentration (b) between benchmark run (blue) and sensitivity tests with different α_sed. Note: test 2 – cyan; test 3 – orange; test 4 – black; and test 5 – magenta. The black dashed line shows the time of the Gustav landfall.

Download

In general, the influence of α_sed is significant when underwater light is limited by sediment and SSC was moderately high. The optical environment over the muddy inner Louisiana shelf is dominated by CDOM and chlorophyll under normal conditions (D'Sa and Miller, 2003). During energetic events (e.g., hurricanes and cold fronts), however, high concentrations of resuspended sediment become the most important light absorber. Given the high frequency of cold fronts in winter (once every 3–7 d) and energetic hurricanes in summer (Keim et al., 2007; Walker and Hammack, 2000), it is reasonable to speculate that the ecosystem along coastal Louisiana would be sensitive to α_sed, not only on an event scale but also on seasonal to annual scales. The role of long-term sediment dynamics in water clarity and marine ecology has been reported in other regions (Capuzzo et al., 2015; Dupont and Aksnes, 2013; Wilson and Heath, 2019). To prove this hypothesis in the nGoM, we need a long-term biogeochemical simulation that explicitly includes sediment-induced light attenuation effects in the future.

4.5 Model uncertainties

The optical environment over the muddy Louisiana shelf is dominated by phytoplankton, suspended sediment, CDOM and detritus particles (Le et al., 2014). The model presented in this study only includes the light attenuation due to the former two constituents, and the potential influence from CDOM and detritus warrants future study. Light attenuation due to CDOM was simply parameterized using salinity in a previous model study (Justić and Wang, 2014), yet few biogeochemical models incorporate dissolved and/or detritus-induced light attenuation. In the nGoM, CDOM plays an indispensable role in modulating the optical properties of inner shelf waters (D'Sa and Miller, 2003; D'Sa et al., 2018); thus, including CDOM-induced light attenuation would likely lower the threshold of sediment resuspension above which the nGoM ecosystem would be light limited. To estimate the importance of CDOM-induced light attenuation in the biogeochemical models, a long-term CDOM climatology is desired in the future.

We use SeaWiFS-derived chlorophyll concentrations to evaluate model performance. However, deriving high-quality chlorophyll data during hurricanes is still a challenge because (1) the presence of thick clouds limits the availability and quality of satellite images (Huang et al., 2011), (2) the uncertainty of chlorophyll estimation can be amplified by strong CDOM absorption (D'Sa et al., 2006; D'Sa and Miller, 2003), and (3) conducting chlorophyll measurements during a hurricane to calibrate bio-optical algorithms is limited by cost and safety. Given the rapid change in and a wide range of sediment and chlorophyll concentrations after hurricanes, the algorithms based on observations under normal conditions might incur a bias. To achieve high-quality satellite-derived chlorophyll data, it is essential to optimize an algorithm based on field observations during hurricane events.

In this study we simplified α_sed as a constant over the entire GoM. When water is highly turbid, the availability of light for photosynthesis could be more related to sediment concentration rather than α_sed (McSweeney et al., 2017). Thus, using a constant to represent the sediment light attenuation coefficient when sediment concentration is high should not introduce considerable bias. The optical characteristics of sediment particles, however, could greatly modify light availability underwater when SSC is relatively low (Storlazzi et al., 2015). Our sensitivity tests also suggest the importance of α_sed in photosynthesis and primary production, as resuspended sediment settles back on the sea floor. Therefore, it is necessary to develop a spatially explicit α_sed to better parameterize the sediment's impact on light attenuation in future work.

Organic matter remineralization in sediment can dramatically increase nutrient concentration in the bottom boundary layer during strong resuspension (Couceiro et al., 2013). Field measurements after hurricanes Gustav and Ike suggested that the resuspension can expose the organic material in sediment to a more favorable environment for respiration (McCarthy et al., 2013). Nevertheless, so far most biogeochemical models either neglect or simply parameterize this process (Chai et al., 2007; Fennel et al., 2006; Kishi et al., 2007). Moriarty et al. (2018) developed a particulate organic matter resuspension model and found that the remineralization intensity increased by an order of magnitude during moderate resuspension events in the nGoM. Given the strong storm-driven resuspension during hurricanes, nutrient dynamics can be modified greatly by remineralization after the storm passage as well. Thus, incorporating organic matter resuspension and remineralization, in conjunction with the light attenuation effects addressed in this study, will help to improve our understanding of a hurricane's impact on the biogeochemical cycling in shelf waters.

Our biogeochemical model includes freshwater and terrestrial nutrient input via a river channel. Du et al. (2019) estimated the freshwater budget during hurricane Harvey and found that surface runoff and groundwater accounted for ∼ 34 % of the total freshwater load during the hurricane. Although our understanding of the nutrient flux associated with these two types of freshwater inputs is still limited, excluding surface runoff and groundwater flux in the model implies our underestimation of terrestrial nutrient discharge from the land. Coupling groundwater and hydrology models with an ocean model is a feasible way of achieving a comprehensive assessment of a hurricane's impact on the coastal and shelf ecosystem. In addition, water heating due to light absorption can also impact the ecosystem (Cahill et al., 2008; Mobley et al., 2015), but it has yet to be considered in our model.