A 3-D unstructured-grid hydrodynamic model for the
northern Gulf of Mexico was developed, with a hybrid
The northern Gulf of Mexico (GoM) is characterized by complicated shelf and coastal processes including multiple river plumes with varying spatial scales, a highly energetic deep current due to steep slopes, upwelling in response to alongshore wind, and mesoscale eddies derived from the Loop Current in the Gulf Stream (Oey et al., 2005; Dukhovskoy et al., 2009; Dzwonkowski et al., 2015; Barkan et al., 2017). Freshwater from the Mississippi–Atchafalaya River (MAR) basin introduces excess nutrients and terminates amidst one of the United States' most productive fishery regions and the location of the largest zone of hypoxia in the western Atlantic Ocean (Rabalais et al., 1996, 2002; Bianchi et al., 2010). The physical, biological, and ecological processes in the region have been attracting increasing attention, given its sensitive response to large-scale climate variation, accelerated sea-level rise, and extensive anthropogenic interventions (Justić et al., 1996; Rabalais et al., 2007).
Understanding the interaction and coupling between regional-scale ocean dynamics and local-scale estuarine processes is of great interest. Many observational (in situ and satellite) (e.g., Cochrane and Kelly, 1986; DiMarco et al., 2000; Chu et al., 2005) and numerical modeling (e.g., Zavala-Hidalgo et al., 2003, 2006; Hetland and Dimarco, 2008; Fennel et al., 2011; Gierach et al., 2013; Huang et al., 2013) studies have been conducted for the shelf of the GoM. Hetland and Dimarco (2008) configured a hydrodynamic model based on the Regional Ocean Modelling System (ROMS; Shchepetkin and McWilliams, 2005) for the Texas–Louisiana shelf, which has been used for subsequent physical and/or biological studies (Fennel et al., 2011; Laurent et al., 2012; Rong et al., 2014). Zhang et al. (2012) extended the model domain westward to cover the entire Texas coast. Wang and Justić (2009) applied the Finite-Volume Coast Ocean Model (FVCOM; Chen et al., 2006) over a similar domain to that of Hetland and Dimarco (2008). Lehrter et al. (2013) applied the Navy Coastal Ocean Model (NCOM; Martin, 2000) over the inner Louisiana shelf with a focus on Mississippi River plumes. In addition, there were modeling studies for larger domains such as the entire GoM (Oey and Lee, 2002; Wang et al., 2003; Zavala-Hidalgo et al., 2003). For example, Zavala-Hidalgo (2003) used the NCOM to investigate the seasonally varying shelf circulation in the western shelf of the GoM. Bracco et al. (2016) used the ROMS to examine the mesoscale and sub-mesoscale circulation in the northern GoM.
Other hydrodynamic modeling studies focused on specific estuarine systems such as Galveston Bay (Rayson et al., 2015; Rego and Li, 2010; Sebastian et al., 2014), Mobile Bay (Kim and Park, 2012; Du et al., 2018a), and Choctawhatchee Bay (Kuitenbrouwer et al., 2018). These models tend to have smaller domains, including the target estuary and the inner shelf just outside the estuary. The dynamics in these coastal bays are affected by both large-scale shelf conditions and localized small-scale geometric and bathymetric features such as narrow but deep ship channels, seaward-extending jetties, and offshore sandbars, which are typically on the order of 10 to 100 m. Including both the estuarine and shelf processes and their interactions is critically important for a more comprehensive understanding of regional physical oceanography in the northern GoM. For this purpose, cross-scale models with unstructured grids have become an attractive option.
The hydrodynamic conditions (e.g., salinity, stratification, and vertical mixing) over the Louisiana shelf are known to be dominated by the influence of MAR plumes (Lehrter et al., 2013; Rong et al., 2014; Androulidakis et al., 2015). However, their effect on the salinity on the Texas shelf has not been well documented. Measurements at Port Aransas (600 km to the west of Atchafalaya River) show an evident seasonal cycle, with higher salinity during the summer and lower salinity during the winter (Bauer, 2002). Is this seasonality related to the seasonal variation of the MAR discharge and/or to the seasonality of the shelf transport? A broader question may be how the MAR discharge affects the salinity along the Texas coast. Furthermore, it is also important to understand the temporal and spatial scales with which the salinity at or near the mouth of an estuarine system respond to river plumes from neighboring river systems. For example, how long will it take for the salinity at the Texas coast to respond to a pulse of freshwater input from the MAR? This timescale in comparison to the timescales of estuarine processes (e.g., recovery timescale from storm disturbance) will allow one to determine whether the remote influence of neighboring major rivers is necessary to consider.
Here, we present a model for the northern GoM with a domain including all the major estuaries, as well as the shelf, and a fine-resolution grid for local estuaries to resolve small-scale bathymetric or geometric features such as ship channels and dikes. Using Galveston Bay as an example, we highlight the flexibility and capability of the model to simulate both estuarine and shelf dynamics. We demonstrate the importance of the interactions among estuaries and the shelf by investigating the remote influence of the MAR discharge on the hydrodynamics along the Texas coast.
We employed the Semi-implicit Cross-scale Hydroscience Integrated System Model (SCHISM; Zhang et al., 2015, 2016), an open-source community-supported modeling system derived from the early SELFE model (Zhang and Baptista, 2008). SCHISM uses a highly efficient semi-implicit finite-element and finite-volume method with a Eulerian–Lagrangian algorithm to solve the turbulence-averaged Navier–Stokes equations under the hydrostatic approximation. It uses the generic length-scale model of Umlauf and Burchard (2003) with the stability function of Kantha and Clayson (1994) for turbulence closure. One of the major advantages of the model is that it has the capability of employing a very flexible vertical grid system, robustly and faithfully resolving the complex topography in estuarine and oceanic systems without any smoothing (Zhang et al., 2016; Stanev et al., 2017; Du et al., 2018b; Ye et al., 2018). A more detailed description of the SCHISM, including the governing equations, horizontal and vertical grids, numerical solution methods, and boundary conditions, can be found in Zhang et al. (2015, 2016).
The model domain and the horizontal grid, with the upper
panels showing zoomed-in views of selected coastal systems. Locations of major
river inputs are indicated with red dots, with the associated mean river
discharges (m
The model domain covers the Texas, Louisiana, Mississippi, and Alabama
coasts, including the shelf as well as major estuaries (e.g., Mobile Bay,
Mississippi River, Atchafalaya River, Sabine Lake, Galveston Bay, Matagorda
Bay, and Corpus Christi Bay) (Fig. 1). The domain also includes part of the
deep ocean to set the open boundary far away from the shelf to avoid
imposing boundary conditions at topographically complex locations. The
horizontal grid contains 142 972 surface elements (triangular and
quadrangular), with the resolution ranging from 10 km in the open ocean to
2.5 km on average on the shelf (shallower than 200 m) to 40 m at the Houston
Ship Channel, a narrow but deep channel along the longitudinal axis of
Galveston Bay. The fine grid for the ship channel is carefully aligned with
the channel orientation in order to accurately simulate the salt intrusion
process (Ye et al., 2018). Vertically, a hybrid
The bathymetry used in the model is based on the coastal relief model (3 arcsec resolution;
Bathymetry in the model domain
showing zoomed-in views
The model was validated for the 2-year conditions in 2007–2008 and was
forced by the observed river discharge, reanalysis atmospheric forcing, and
open boundary conditions from global HYCOM output. Daily freshwater inputs
from United States Geological Survey (USGS) gauging stations were specified at 15 river boundaries (Fig. 1). For the Mississippi River, the largest in the study area, river
discharge at Baton Rouge, LA (USGS 07374000), was used. For the Atchafalaya
River, the second largest, the discharge data at the upper river station
(USGS 07381490 at Simmesport, LA) were used, but the data before 2009 at this
station are not available. However, we found a significant linear
relationship between this station and the one near the river mouth (USGS
07381600 at Morgan City, LA) with a 2 d time lag (
Reanalyzed 0.25
To investigate the remote influence of the MAR discharge, we conducted three numerical experiments that use the same model configuration as in the realistic 2007–2008 model run except for freshwater discharge, wind forcing, initial salinity condition, and salinity boundary condition. To isolate the influence of the MAR discharge, we considered freshwater discharges (constant long-term means) only for the Mississippi River, Atchafalaya River, and Galveston Bay, with no discharge from other coastal systems. To examine the effect of seasonal wind, we chose the January 2008 and July 2008 winds as representative of winter and summer winds, respectively. The January wind was dominated by northeast–east wind and expected to induce a stronger downcoast (from Louisiana toward Texas) longshore current compared to the predominantly south wind in July (Fig. S1). The initial salinity condition is set to 36 psu throughout the entire domain and for all vertical layers. Salinity at the ocean boundary is set to 36 psu throughout the simulation period.
Differences among the three experiment settings include the following: (1) experiment
Jan-G includes only the river discharges into Galveston Bay
(259 m
The model results for 2007–2008 were compared with observations for water
level at seven NOAA tidal gauge stations, salinity at four Texas Water
Development Board (TWDB) stations, temperature at three NOAA stations, and
current velocity at two Texas Automated Buoy System (TABS) buoys (see Fig. 2
for station locations). Comparisons were made for both total and subtidal
(48 h low-pass-filtered) components. For quantitative assessment of the
model performance, two indexes were used, model skill (Willmott, 1981) and
mean absolute error (MAE):
Model–observation comparisons were made for water level at stations
along the coast and inside Galveston Bay. Manning's friction coefficient,
which is converted to the bottom drag coefficient for the 3-D simulation in
the model, was used as a calibration parameter. The model results with a
spatially uniform Manning's coefficient of 0.016 m
Subtidal surface elevation comparison between the model (red line) and observations (black line) at NOAA tidal gauge stations (see Fig. 2 for their locations).
Error estimates for model–data comparison for 2007–2008.
The model reasonably reproduces the observed variation in salinity at stations inside Galveston Bay (Fig. 4 and Table 1). The MAEs are no larger than 3 psu and the model skills range between 0.81–0.93 and 0.75–0.93 for the total and subtidal components, respectively. It is important to note that the salinity at the bay mouth under normal (i.e., non-flooding) conditions is sensitive to the longshore transport of low-salinity water from neighboring estuaries, such as the nearby Sabine–Neches River, Atchafalaya River, and Mississippi River. Successful simulation of salinity at the bay mouth requires an accurate simulation of not only the bay-wide transport, but also the longshore transport. Errors in the modeled salinity at the bay mouth can propagate to the upper bay. For example, salinity during days 60–100 is overestimated at the mouth (station BOLI) and this error propagated into the middle bay station (station MIDG) (Fig. 4). Discrepancies as large as 10 psu are not likely caused by inaccurate discharge from the Trinity River, as this river has a very limited influence on the salinity on the shelf (further discussed in Sect. 4.3). Unfortunately, with no data available for the vertical salinity profile, the model performance for vertical mass transport cannot be evaluated. However, accurate simulation of the observed salinity at the mid-bay station provides alternative evidence supporting the model's validity in horizontal mass transport and salt intrusion.
Salinity comparison between the model (red line) and observations (black cross) at four TWDB stations (see Fig. 2 for their locations).
The model also captures the sharp change in salinity during Hurricane Ike
(around day 620). The salinity at the upper bay (Fig. 4b) decreased from 26 psu to 0 within 2 d, which was caused by a pulse of freshwater
discharge from Lake Houston (see reservoir storage at USGS 08072000). In
addition, the model reproduces the spatial difference well in the amplitude
of the tidal signal in salinity. Salinity in Trinity Bay (Fig. 4a) shows a very
weak tidal signal, while salinity at the bay mouth (Fig. 4d) has a much
stronger tidal signal. Galveston Bay, in general, has micro-tidal ranges
with a mean tidal range of 0.3 m at the mid-bay station (Eagle Point in Fig. 2). The tidal signal, however, becomes stronger at the narrow bay mouth (2.5 km wide), with the tidal current being as strong as 1 m s
The modeled salinity was also compared to the observed salinity structure over the Texas–Louisiana shelf using the data from a shelf-wide summer survey in July 2008 as an example (Fig. 5). Both the horizontal and vertical structures of salinity on the shelf are well reproduced by the model, with an MAE over 65 stations of 1 and 2 psu for the surface and bottom salinity, respectively. Data and the model consistently show a relatively shallow halocline at section A (west of Mississippi Delta) and a deeper halocline at section F (off Atchafalaya Bay). The upper layer off Atchafalaya Bay was nearly well mixed, which is also reproduced by the model, although the model somewhat underestimates the bottom salinity at section F. In addition, the model also shows that there was little tidal variability of the vertical salinity profile on the shelf (e.g., stations F4 and A7 in Fig. 5), which can be attributed to the small tidal range in the northern GoM.
Salinity distribution at the Texas–Louisiana shelf from
the shelf-wide survey on 22–27 July 2018: comparison of
The model reproduces the observed temperatures well at three NOAA stations
located from the Galveston Bay mouth to the upper bay (Fig. 6). Both the
seasonal and diurnal cycles are well captured, with MAEs of about
1
Temperature comparison between the model (red line) and observations (black line) at three NOAA stations (see Fig. 2 for their locations).
The model performance in reproducing temperature over the Texas–Louisiana shelf was further examined with satellite data for sea surface temperature (SST). Seasonality of the SST extracted from MODIS over the northern GoM is overall reproduced well (Fig. 7). It is worth noting that the model also reproduces the relatively low temperatures on the southern Texas coast during summer, which is a well-known upwelling zone during the summertime when upcoast (from Texas toward Louisiana) winds drive an offshore surface transport (Zavala-Hidalgo et al., 2003).
Temperature comparison (monthly average) between the model (left panels) and MODIS satellite data (right panels) for selected months in 2008.
The shelf current plays a key role in transporting low-salinity water
originating from MAR, and it can be affected by not only the wind field, but
also the mesoscale eddies in the northern GoM. One of the important features
of the Texas–Louisiana shelf is the quasi-annual pattern of the shelf current,
which is predominantly westward most of the time except during summer
(Cochrane and Kelly, 1986; Li et al., 1997; Cho et al., 1998). The prominent
downcoast shelf current is driven by along-shelf wind and enhanced by the
MAR discharge (Oey, 1995; Li et al., 1997; Nowlin et al., 2005). Under
summer wind that usually has an upcoast component, the nearshore current is
reversed to the upcoast direction (Li et al., 1997). Such seasonality also
occurred during 2007–2008. The model reproduces the observed subtidal
component of the surface longshore current at two TABS buoy stations outside
Galveston Bay, buoy B (
Comparison of the subtidal east–west surface shelf current between the model (red line) and observations (black line) at two TABS buoys (see Fig. 2 for their locations).
The conditions in Texas coastal waters are impacted by several remote sources, including mesoscale eddies (Oey et al., 2005; Ohlmann and Niiler, 2005), longshore transport of low-salinity water from major rivers (Li et al., 1997; Nowlin et al., 2005), and Ekman transport induced by longshore wind and the resulting upwelling–downwelling (Li et al., 1997; Zhang et al., 2012). Here, based on the realistic model results and numerical experiments, we discuss the remote influence of major river discharge and shelf dynamics on the longshore transport, salinity, stratification, and vertical mixing at the Texas coast, as well as the water exchange between the coastal ocean and local coastal system.
The strength and direction of the shelf current are sensitive to the wind field.
Comparison of the model results on day 150 (31 May 2007) and day 160
(10 June 2007) clearly shows the different distribution of lower-salinity water
along the coast in response to wind field and the resulting shelf current
(Fig. 9). The river discharge differences between these two days are
negligible, and thus the differences in lower-salinity water distribution can
be mainly attributed to the differences in shelf current. Day 150 was
characterized by a significant downcoast shelf current in the inner shelf,
with a current speed exceeding 0.5 m s
Regulated by the shelf current, salinity distribution over the shelf
exhibits evident seasonality. The model results show that a narrow band of
lower-salinity water persisted from Louisiana to the western Texas inner
shelf during January–May 2008 (Fig. 10). The salinity at the Galveston Bay
mouth decreased by about 10 psu from January to May, which can be attributed
to the increasing Mississippi discharge from January to May in 2008
(Mississippi discharge data at
Comparison of the observed wind field and the modeled
surface residual current and surface salinity on day 150 (31 May 2007) and
day 160 (10 June 2017). The filled colors indicate the daily mean wind
speed
The modeled monthly mean surface salinity in 2008, with the grey contour lines denoting depth contours of 50, 100, 150, and 200 m.
Longshore transport plays a key role in redistributing freshwater
from the estuarine bays along the shelf. The results from three numerical
experiments show that, under the January wind, the downcoast longshore
transport among four selected cross-shelf sections varies little. The longshore transport is enhanced by the MAR discharge (long-term
mean) by 10 %–14 % (
Downcoast longshore transport at four selected cross-shelf sections for three numerical experiments with constant long-term mean river discharges: river discharges into Galveston Bay only with January 2018 wind (Jan-G) and the MAR discharge as well as discharges into Galveston Bay with January 2018 wind (Jan-GAM) or July 2018 wind (Jul-GAM).
The influence of the MAR discharge on shelf salinity also depends on the wind condition and the resulting shelf current. Surface salinity maps averaged over days 250–300 show distinctly different spatial patterns of the lower-salinity water under different wind conditions (Fig. 12). The patterns are similar to the results from the 2007–2008 realistic run (Fig. 10). Under the winter wind, lower-salinity water is trapped nearshore by the shelf current, forming a narrow band along the coast. Under the summer wind, on the other hand, water on the Texas shelf is replenished by saltier water originating from the southwest, leading to a tongue-shaped saltier-water intrusion toward the lower-salinity water over the Louisiana shelf. Consequently, salinity is higher on the Texas shelf and lower on the Louisiana shelf when compared to that under the winter wind.
Surface salinity distributions averaged over days 250–300 from three numerical experiments. Grey contour lines denote depth contours for 50, 100, 150, and 200 m.
Numerical experiments reveal different time and spatial scales with which the surface salinity in Texas coastal water responds to the MAR discharge (Fig. 13). At the Galveston Bay mouth, the salinity begins to decrease from about day 25 in response to the MAR discharge and continues to decrease until around day 100 when it reaches a quasi-steady state. The MAR discharge (long-term mean) reduces the salinity by about 10 psu under the January wind but only by 5–6 psu under the July wind. Further south at the Port Aransas mouth, the response time doubles to about 50 d, with the MAR discharge reducing the salinity by about 6 psu under the January wind. Salinity changes little in response to discharges from Galveston Bay and the MAR discharge under the July wind. As the influence from Galveston Bay is very limited at the Aransas Bay mouth even under a downcoast wind, it is reasonable to assume the influence will be even smaller under an upcoast wind.
Subtidal surface salinity at the mouth of
Vertical profiles of salinity along a section from the Trinity Bay, along
the Houston Ship Channel and the adjoining shelf, show that the MAR discharge
increases salinity stratification on the shelf (Fig. 14). The lower-salinity
water along the coastline increases the cross-shelf baroclinic pressure
gradient, leading to a stronger stratification. There is a distinctive difference
between Jan-GAM and Jul-GAM. A stronger stratification on the inner shelf appears
under the July wind, with the bottom-surface salinity difference as large as
4 psu. Vertical mixing on the inner Texas shelf is weakened due to the MAR
discharge, particularly under the July wind. The vertical diffusivities are 1
or 2 orders of magnitude smaller than those under the January wind. Under
the July wind, the stratification along the ship channel becomes stronger,
probably because of higher salinity near the bay mouth and/or a weaker
wind in July with a mean speed of 4.79 m s
Salinity
Salinity change due to remote river input and a shift in the wind field
affects the estuarine dynamics, such as estuarine circulation, salt flux,
and estuarine–coastal exchange. We examined the change in exchange flow and
salinity at the Galveston Bay mouth due to remote river influence and
a different shelf current. Following Lerzak et al. (2006), we calculated the
tidally averaged and cross-sectionally varying components (
Vertical profiles of exchange flow (
The influence of the MAR discharge on the dynamics of Galveston Bay was
further examined with total exchange flow (TEF) using the isohaline
framework method proposed by MacCready (2011), which was found to be a
precise way to quantify landward salt transport (Chen et al., 2012). In
this method, the tidally averaged volume flux of water with salinity greater
than
Table 2 lists the values of
Total exchange flow (
An unstructured-grid hydrodynamic model with a hybrid vertical grid was developed and validated for water level, current velocity, salinity, and temperature for Galveston Bay as well as over the shelf in the northern GoM. The good model performance, particularly in terms of salinity (vertically and horizontally), is at least in part attributable to the inclusion of multiple river plumes along the coastline as well as the interaction between estuaries and the shelf. This model provides a good platform that can be used for other purposes in future studies. Its flexibility in the horizontal and vertical grids allows for refinement in any region of interest without a penalty in the time step (due to the semi-implicit scheme). For example, it would be relatively easy to adapt the model by refining the grid inside any target bay, e.g., Corpus Christi Bay.
The 2007–2008 model run reveals the seasonally varying influence of the MAR discharge on the Texas shelf. Three numerical experiments were carried out to examine the extent to which the major rivers in the region influence local coastal bay systems in Texas. The MAR discharge has a great influence on the salinity regime along the Texas coast and its influence depends on the wind-controlled shelf circulation. Winter wind drives a stronger downcoast longshore transport with its magnitude at least 1 order larger than that under summer wind. The MAR discharge (long-term mean) enhances the downcoast transport by 10 %–14 % under winter wind, and it lowers the salinity by up to 10 psu at the mouth of Galveston Bay and 6 psu at the mouth of Port Aransas. Vertical mixing is also sensitive to wind forcing. Summer wind tends to displace low-salinity water further offshore, while the winter wind constrains the low-salinity water to a narrow band against the shoreline. As a result, the stratification is stronger and vertical mixing is weaker over the shelf during summer. The lower-salinity condition on the Texas shelf decreases the longitudinal salinity gradient at the bay mouth, leading to a weakened estuarine circulation and weaker salt exchange.
This study demonstrates the necessity of including the remote influence of the MAR discharge for modeling Texas coastal systems, particularly for processes associated with relatively long timescales (e.g., months). Receiving relatively small freshwater discharge and being limited by narrow outlets and small tidal ranges, the estuarine bay systems along the Texas coast, e.g., Galveston Bay, Aransas Bay, and Corpse Christi Bay, are characterized by relatively slow water exchange and long flushing times. In this study, we show that the exchange flow plays an important role for water renewal and that the exchange flow varies greatly depending on the wind field and the resulting shelf current. Modulation by the MAR discharge, when coupled with downcoast wind conditions, could have a great influence on the dynamics of estuaries along the Texas coast.
All the observational data used for model validation are available online.
Salinity data are extracted from TDWB
(
The supplement related to this article is available online at:
JD and KP led the effort for model development, data analysis, and preparation of the paper. JS, YJZ, FY, and ZW provided guidelines for the model configuration in terms of forcings and boundary conditions. XY assisted in the visualization of modeled and observed data. NNR provided the shelf-wide survey data for the model validation. All authors were involved in writing the paper.
The authors declare that they have no conflict of interest.
The numerical simulation was performed on the high-performance computer cluster at the College of William and Mary.
This study was partially supported by the Texas Coastal Management Program, the Texas General Land Office, and NOAA through CMP contract no. 19-040-000-B074.
This paper was edited by Eric J. M. Delhez and reviewed by Ivica Janeković and one anonymous referee.