Impacts of climate change on estuarine stratification and implications for hypoxia within a shallow subtropical system

Abstract

Vertical density stratification often plays an important role in the formation and expansion of coastal hypoxic zones through its effect on near-bed circulation and vertical oxygen flux. However, the impact of future climate change on estuarine circulation is widely unknown. Here, we developed and calibrated a three-dimensional hydrodynamic model for Pensacola Bay, a shallow subtropical estuary in the northeastern Gulf of Mexico. Model simulations based on years 2013–2017 were applied to examine changes in salinity, temperature, and density under future climate scenarios, including increased radiative forcing (IR) and temperature (T), increased freshwater discharge (D), sea level rise (SLR), and wind intensification (W). Simulations showed that the impacts of climate change on modeled state variables varied over time with external forcing conditions. The model demonstrated the potential for sea level rise and increased freshwater discharge to episodically increase vertical density gradients in the Bay. However, increased wind forcing destabilized vertical gradients, at times reducing the spatial extent and duration of stable stratification. For time periods with low freshwater discharge, moderate increases in wind speed (10%) can destabilize density gradients strengthened by increased discharge (10%) and sea level rise (0.48 m). In contrast, destruction of strong density gradients that form near the mid-Bay channel following flood events requires stronger wind forcing. These results highlight the importance of considering natural variability in freshwater and wind forcing, as well as local phenomena that are generally unresolved by global climate models.

1. Introduction

Global climate change fueled by anthropogenic activities will affect biogeochemical and physical processes within coastal ecosystems. There is some concern that these changes will contribute to the expansion of coastal hypoxic zones, which have been increasingly observed since the 1960s (Diaz and Rosenburg, 2008; Rabalais et al., 2010; Altieri and Gedan, 2015). Coastal hypoxia is associated with a range of lethal and non-lethal aquatic species effects (Vaquer-Sunyer and Duarte, 2008, 2011), including reduced growth rates (Chabot and Dutil, 1999; McNatt and Rice, 2004; Hrycik et al., 2017) as well as behavioral and physiological alterations (Mattiasen et al., 2020). While excess nutrient loading drives biological processes that deplete oxygen (Diaz and Rosenburg, 2008), hypoxia is often modulated by physical processes, such as vertical density stratification, freshwater discharge, and tidal and wind mixing (Scully, 2010, 2013; Xia and Jiang, 2015). The effects of climate change on estuarine hydrodynamics will therefore have important implications for surface water quality (Whitehead et al., 2009; Robins et al., 2016; Tian et al., 2021).

Evaluating the effects of climate change is often difficult in part due to the range of processes impacted, including increased sea surface temperature, altered freshwater discharge patterns, and sea level rise. These factors are often downscaled from global climate models and are associated with a range of confidences at varying temporal and spatial scales (IPCC, 2013). Downscaled climate data are often at low spatiotemporal resolutions that may not be accurate to assess the impacts of climate change and climatic variability on an estuarine system at regional or local scales. Furthermore, climate impacts may have discordant effects on biogeochemical and physical processes that regulate water quality (Robins et al., 2016). For example, warmer sea temperatures decrease oxygen saturation, increase settling rates of organic matter (Bach et al., 2012), and increase benthic metabolism (Caffrey, 2004; Vaquer-Sunyer et al., 2012). On the other hand, sea level rise may increase the volume and intrusion length of cooler salt water (Hong and Shen, 2012; Liu and Liu, 2014; Robins et al., 2014; Krvavica and Ružić, 2020; Khojasteh et al., 2021), which could mitigate biogeochemical effects associated with warming. This is particularly true in shallow estuaries, where 1 m of sea level rise would increase salinity intrusion length by greater than 25% (Prandle and Lane, 2015). Sea level rise and increased freshwater discharge will also likely strengthen vertical density gradients, which may weaken vertical oxygen exchange allowing for the development of near-bed hypoxia (Hong and Shen, 2012).

While the above observations are generally true for a single climate factor, synergistic or discordant effects of multiple climate factors are not well understood and may vary depending on site-specific conditions. For example, while the role of wind in breaking down density stratification and increasing vertical oxygen flux has been widely recognized (Simpson et al., 1990; Scully et al., 2005; Lin et al., 2008), future changes in wind dynamics and interannual variability are often not considered when evaluating the effects of climate change on estuarine circulation. In shallow stratified estuaries, strong winds can eliminate near-bed hypoxia over relatively short time scales of a few days (Xia and Jiang, 2015; Duvall et al., 2022). Strengthening of local land-sea temperature gradients in the future could intensify sea breeze circulation patterns (Lebassi et al., 2009; Liu et al., 2015; Pazandeh Masouleh et al., 2019), weakening vertical density gradients and mitigating the development of seasonal hypoxia. Downscaling global climate factors without considering high-frequency processes such as sea breeze circulation may lead to inaccurate predictions of coastal ecosystem response (Fagundes et al., 2020).

Here, we developed and calibrated a three-dimensional hydrodynamic model for Pensacola Bay, an estuary in the northeastern Gulf of Mexico. The model was used to evaluate the effects of climate change on salinity, temperature, and density gradients over a five-year simulation period (2013–2017). We considered the independent and combined effects of changes in external forcing, including increased irradiance, increased atmosphere and river temperatures, increased freshwater discharge, sea level rise, and strengthened cross-shore winds. The findings presented here focus on changes in vertical density stratification near the tidal channel. Uncertainty regarding climate change factors and implications for hypoxia in Pensacola Bay are discussed.

2. Methods

2.1. Study site

Pensacola Bay system is a shallow (mean depth = 3 m), subtropical estuarine complex in northwest Florida (Fig. 1a). The system drains a 18,000 km² watershed and is comprised of four interconnected bays: Escambia Bay, Blackwater Bay, East Bay, and Pensacola Bay proper. Discharge from the Escambia River accounts for 80% of all freshwater inputs (Murrell and Lores, 2004; Hagy and Murrell, 2007) and empties into Escambia Bay and Pensacola Bay proper on the western side of the system (Fig. 1b and c). Tides in the Bay are microtidal (mean tidal range = 0.37 m), and drive exchanges with the Gulf of Mexico through a narrow, deep channel that extends from the mouth of Pensacola Bay to the middle reach of Escambia Bay near P5 (Fig. 1c). Tidal exchanges with Big Lagoon to the west and Santa Rosa Sound to the east are generally small (Hagy and Murrell, 2007). Strong vertical density gradients form in deeper areas near the channel during the late spring and early summer following elevated discharge from the Escambia River.

Fig. 1.

(a) Location of study site in northwest Florida, USA. (b) Model domain including the Pensacola Bay system and portions of Santa Rosa Sound (SRS) as well as northern Gulf of Mexico. Pensacola Bay system is comprised of Pensacola Bay proper (PB), Escambia Bay (ES), Blackwater Bay (BB), and East Bay (EA). Red bounding box in (a) and (b) corresponds to (c) the location of observational stations used to calibrate the model in 2016 and 2017. The location of NOAA 42012 (Table 1) is not shown and is located approximately 20 km west of the model domain. Unlabeled rivers are smaller tributaries that account for approximately 1% of mean Escambia River discharge. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

The magnitude of springtime freshwater discharge and sea breeze intensity influence interannual variations in the spatial extent and duration of seasonal hypoxia, which is commonly observed near the channel bottom including at P5 (Hagy and Murrell, 2007; Duvall et al., 2022). For the model period (2013–2017), mean Escambia River discharge from April–May was 242 ± 190 m³ s⁻¹ (USGS, 2020, Fig. 2a). Discharge from the Escambia River exceeded 945 m³ s⁻¹ during a flood event in April 2014 (Fig. 2b). Following the flood, stable stratification ($\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}>10\text{kg}{\mathrm{m}}^{-3}$) observed near the channel was resistant to vertical mixing, leading to the development of a hypoxic zone that persisted for several months (Fig. 2d, f; Duvall et al., 2022). Higher dissolved oxygen concentrations were observed in 2016 due to drier conditions that resulted in weaker stratification in the middle reach of Escambia Bay.

Fig. 2.

Distribution of April–May freshwater discharge (m³ s⁻¹) from the Escambia River for (a) 2013–2017 and (b) 2014 and 2016 only. (c, d) Distribution of $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ (kg m⁻³) and (e, f) bottom dissolved oxygen (DO, mg L⁻¹) in the middle reach of Escambia Bay during June–July for same years shown in (a) and (b). (g) Distribution of power spectral density (PSD; m² s⁻²/hr⁻¹) at period = 24 h of the cross-shore wind vector during April–August for years 1980–2020. Vertical lines show mean PSD during the model period (2013–2017) and when cross-shore wind speeds increase by 10, 30, and 50%.

Cross-shore winds destabilize vertical density gradients and reoxygenate bottom waters in the late spring and summer, mitigating the formation of hypoxia (Duvall et al., 2022). A peak in the power spectral density (PSD; Welch, 1967) of the cross-shore wind vector at period = 24 h (h) is consistent with sea breeze circulation, whereby southerly winds develop in the late afternoon due to differential heating of air masses. Prior studies have shown that the amplitude of the sea breeze circulation is maximized at 30° latitude due to the inertial rotation of the surface wind vector being in resonance with the diurnal heating cycle (Rotunno, 1983; Yan and Anthes, 1987). The magnitude of summertime sea breeze can be estimated from the PSD at 24 h and shows that winds were generally weaker during the model period compared to previous years (Fig. 2g).

2.2. Model setup

We developed a three-dimensional hydrodynamic model using the Environmental Fluid Dynamics Code (EFDC; DSI, 2020) originally developed by Hamrick (1992) and described in (e.g., Jin et al., 2000; Ji et al., 2007). EFDC models have been widely applied to study shallow estuaries, such as Pearl River Estuary (Wei et al., 2016; Wang and Hong, 2021), Indian River Lagoon (Rosario-Llantín and Zarillo, 2021) and adjacent Perdido Bay (Xia et al., 2011; Xia and Jiang, 2015). EFDC solves turbulent-averaged equations of motion for an incompressible, variable density fluid. A curvilinear model grid was used to represent Pensacola Bay system as well as Santa Rosa Sound and northern Gulf of Mexico (Fig. 1b). The grid consisted of 3299 surface grid cells ranging in size from 0.12 to 1.43 km in the horizontal direction (mean area = 0.35 km²) and 5 vertical sigma layers of equal thickness. Values for hydrodynamic model parameters are given in Supplementary Table 1. The model was also used to compute age of water (DSI, 2020), which represents the total elapsed time for a water particle to be transported to a particular grid cell from a model boundary.

Water surface elevation measured in Pensacola Bay (NOAA PCLF1; Table 1) was applied as tidal forcing at the southern boundary of the model domain in the Gulf of Mexico (NOAA, 2020). Sea surface temperature measured in the Gulf of Mexico (NOAA 42012) and constant salinity (36 ppt) were used as forcing conditions at this boundary. The model domain included twelve freshwater inflow boundaries. Freshwater inputs from the three largest tributaries (Escambia River, Yellow River, and Blackwater River) were based on timeseries of measured discharge (USGS, 2020). Timeseries for nine smaller tributaries were estimated as a fraction of the discharge from Escambia or Yellow Rivers based on simulated inflow using a Loading Simulation Program C (LSPC) watershed model (Shen et al., 2005). In general, mean discharge from each small tributary is approximately 1% of mean Escambia River discharge and does not significantly affect circulation patterns in Pensacola Bay. Sea surface temperature measured in Pensacola Bay (NOAA PCLF1) and constant salinity (0 ppt) were used as forcing conditions at freshwater boundaries.

Table 1.

Summary of data sources for model forcing and implemented changes to forcing data for climate change model runs.

PARAMETER		DATA STATION	LOCATION	CLIMATE CHANGE MODEL
				RUN	IMPLEMENTATION
SOLAR IRRADIANCE		ERA5	Pensacola, FL 30.350 N, 87.250 W	IR	Increased short wave radiation by 6 W m−2
AIR TEMP		ERA5	Pensacola, FL 30.350 N, 87.250 W	T	Increased air temperature by 3 °C
WATER TEMP	Freshwater	NOAA PCLF1	Pensacola, FL 30.404 N, 87.211 W	T	Increased freshwater temperature by 1.7 °C
	Gulf	NOAA 42012	Orange Beach, FL 30.060 N, 87.548 W
RIVERDISCHARGE	Escambia	USGS 02376033	Molino, FL 30.670 N, 87.267 W	D	Increased river discharge by 10%
	Yellow	USGS 02369600	Milton, FL 30.569 N, 86.924 W	D	Increased river discharge by 10%
	Blackwater	USGS 02370000	Baker, FL 30.833 N, 86.735 W	D	Increased river discharge by 10%
WATER LEVEL		NOAA PCLF1	Pensacola, FL 30.404 N, 87.211 W	SLR	Increased sea surface elevation by 0.48 m.
WIND		NOAA PCLF1	Pensacola, FL 30.404 N, 87.211 W	${\mathbf{W}}_{\mathbf{10},\mathbf{30},\mathbf{50}}$	Increased wind speed by 10, 30, and 50%

Meteorological boundary conditions included wind, solar radiation, air temperature, rainfall, evaporation, humidity, cloud cover, and atmospheric pressure. Wind speed and direction measured in Pensacola Bay (NOAA PCLF1) were applied as the wind forcing. Missing wind values were interpolated from the European Center for Medium-Range Weather Forecasts (ECMWF) fifth generation climate reanalysis (ERA5; C3S, 2020) for the subregion closest to the study area (−87.2500, 30.3500). Spatially averaged winds from the ERA5 subregion closest to Pensacola Bay showed good agreement with winds measured at NOAA PCLF1 (Supplementary Fig. 1). All other meteorological conditions were based on ERA5.

2.2.1. Model-data comparison

The Pensacola Bay hydrodynamic model was calibrated and verified using observations from 2016 to 2017. The dataset consisted of continuous timeseries and vertical profile measurements. Timeseries observations allowed for assessment of model performance as a function of frequency, while profile measurements provided greater spatial coverage and were used to verify along-estuary gradients in modeled state variables.

Timeseries of temperature and salinity were collected above the seabed using WET Labs water quality monitors (WQMs). WQMs burst sampled every 30 min for 1 min at 1 Hz sampling frequency. WQMs were deployed in the middle reach of Escambia Bay at P5, P5M and P5E from June–August 2016 and 2017 (Fig. 1c). P5 is situated at the northern end of the mid-Bay channel and is influenced at times by both riverine and marine processes. In 2016, an additional buoy-mounted WQM was deployed at P5 approximately 0.5 m beneath the water surface. This allowed us to calibrate the surface to bottom density gradient ($\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$) as a measure of vertical stratification (Hagy and Murrell, 2007).

Profiles of conductivity, temperature, and depth (CTD) were collected with a SBE 25Plus Sealogger at seven stations along a transect from the upper reach of Escambia Bay to Pensacola Bay proper (Fig. 1c). Stations were surveyed during 39 cruises between January 2016 and October 2017. A total of 193 vertical profiles were collected. Not all stations were surveyed during each cruise and some stations were surveyed more than once on a given date.

To assess model performance, observed $\left(O\right)$ and modeled $\left(M\right)$ sea surface elevation, temperature, and salinity data were used to compute the root-mean-square error ($RMSE$), bias $\left(B\right)$, and Pearson correlation coefficient $\left(r\right)$ as:

$$RMSE=\sqrt{\frac{\sum {\left(O_i-M_i\right)}^2}{n}}$$ (1)

$$B=\frac{1}{n}\sum \left(M_i-O_i\right)$$ (2)

$$r=\frac{\sum \left(O_i-\overline{O}\right)\left(M_i-\overline{M}\right)}{\sqrt{\sum {\left(O_i-\overline{O}\right)}^2\sum {\left(M_i-\overline{M}\right)}^2}}$$ (3)

where $O_i$ is the observed value and $M_i$ is the corresponding modeled value, and $\overline{O}$ and $\overline{M}$ are the means of $n$ observed or modeled values, respectively.

Continuous timeseries measurements allowed for assessment of model performance in the time-frequency domain using the magnitude-squared wavelet coherence. The wavelet coherence varies between 0 (no relationship) and 1 (perfect relationship) and provides a local measure of correlation between observed and modeled variables, or between two modeled variables. Therefore, the wavelet coherence can be used to determine frequencies at which the model is capturing observed variability, or to assess model-model agreement in the time-frequency domain. The magnitude-squared wavelet coherence, $C_n^2$, is computed as

$$C_n^2(s)=\frac{{|S(s^{-1}{W}_n^{MO}(s))|}^2}{S\left(s^{-1}{|W_n^M\left(s\right)|}^2\right)•S\left(s^{-1}{|W_n^O(s)|}^2\right)}$$ (4)

where $S\left(W\right)$ is a smoothing operator,

$$S\left(W\right)=S_s\left(S_t\left(W_n\left(s\right)\right)\right)$$ (5)

(Torrence and Webster, 1999). Here, $s$ is scale and $t$ is time. $W_n^{\mathit{MO}}$ is the cross-wavelet transform and $W_n^M$ and $W_n^O$ are the continuous wavelet transforms of modeled and observed timeseries. The smoothing operator often varies with wavelet (Torrence and Webster, 1999), taken here to be a Morlet wavelet. Values influenced by edge effects (i.e., cone of influence) were excluded from our analyses (Torrence and Compo, 1998). MATLAB codes were provided by Grinsted et al. (2014).

2.3. Climate change simulations

Modifications were made to the present-day model to simulate climate driven changes in radiative forcing (IR), atmospheric and freshwater temperature (T), freshwater discharge (D), sea level rise (SLR), and wind (W), presented in Table 1. Following the Intergovernmental Panel on Climate Change (IPCC) intermediate warming scenario (representative concentration pathway (RCP) 6.0), we assumed an increase in radiative forcing of 6 W m⁻². Under the RCP6.0 scenario, regional air temperature is predicted to rise by 3 °C by 2100 (IPCC, 2013), which would increase the temperature of riverine freshwater discharge by 1.7 °C by 2100 (Morrill et al., 2005; Lehrter et al., 2017). While there is high uncertainty regarding future changes in freshwater discharge (Biasutti et al., 2012; IPCC, 2013), we assumed an increase of 10% by 2100 based on projected mean hydrologic change across North America (Sperna Weiland et al., 2012). Under the RCP6.0 scenario, global mean sea level is projected to rise 0.48 m by 2081–2100 (IPCC, 2013). Previous estimates of sea level rise for Pensacola Bay (Devkota et al., 2013) agree with global mean projections. For the SLR model, increased magnitude of 0.48 m was added to the water level forcing at the southern open boundary following the approach of other recent modeling studies (Hong and Shen, 2012; Wang and Hong, 2021). This approach does not consider changes in tidal range due to geomorphic evolution of the bed (Palmer et al., 2019) or shoreline protection structures (Lee et al., 2017). To evaluate the response of modeled state variables to increased wind forcing, wind speed was increased by 10% to represent projected change in 10 m wind forcing for 2081–2100 (${\mathrm{W}}_{10}$; McInnes et al., 2011; Palmer et al., 2019). We also considered wind speed increases of 30% (${\mathrm{W}}_{30}$) and 50% (${\mathrm{W}}_{50}$) given the interannual variability in sea breeze circulation (Fig. 2g), as well as the uncertainty associated with future changes in local wind patterns. Although these increases exceed current climate projections, a 50% increase in the magnitude of cross-shore winds observed during the model period (2013–2017) still falls within the range of summertime wind conditions observed over the past 40 years (1980–2020; Fig. 2g).

To assess the effects of climate change on modeled state variables, results from IR, T, D, SLR, and W models were independently compared to output from the calibrated model for 2013–2017 representing present-day conditions. The combined effects of T, D, SLR, and W scenarios ($\mathrm{T}+\mathrm{D}+\text{SLR}+\mathrm{W}$) were also considered.

3. Results

3.1. Model validation

Modeled sea surface elevation, temperature, salinity, as well as the bottom to surface density gradient ($\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$) were compared to timeseries (WQM) and vertical profile (CTD) measurements from Pensacola Bay. Timeseries of observed and modeled sea surface elevation at NOAA PCLF1 are highly correlated ($r=0.99$, p < .001), as shown in Fig. 3. For bottom salinity, the range of $r$ values for WQM timeseries collected at P5, P5M, and P5E was 0.59–0.77 (p < .001) and $r=0.65$, p < .001 for CTD measurements at P5 (Table 2; Fig. 4a and b). Average $\mathit{RMSE}$ and $B$ for bottom salinity timeseries were 4.49 ppt and 0.52 ppt, respectively. The range of $r$ values for bottom temperature timeseries was 0.69–0.90 (p < .001) and $r=0.96$, p < .001 for CTD measurements at P5 (Table 2; Fig. 4c and d). Average $\mathit{RMSE}$ and $B$ were less than 1 °C for bottom temperature timeseries. Surface salinity and temperature timeseries were only recorded at P5 in 2016. $r$ values were 0.61 and 0.67 (p < .001), respectively, much lower than correlation values for the surface layer obtained from CTD measurements at P5 ($r=0.86$ and 0.98, p < .001). One reason for the discrepancy between modeled and observed temperature and salinity timeseries is WQM measurements represent point observations, while model output is a spatially averaged quantity. Furthermore, the WQM was deployed approximately 1 m above the seafloor, which was close to the boundary between model layer 1 (i.e., bottom layer) and layer 2. Overall, the model was able to simulate along-estuary gradients in depth-averaged salinity and temperature (Fig. 4e–h). Across all CTD stations, average $B$ in salinity and temperature was 3.69 ± 0.56 ppt and 0.12 ± 0.24 °C, respectively.

Fig. 3.

Observed and modeled water level (m) at NOAA PCLF1 from May–September (a, c) 2016 and (b, d) 2017.

Table 2.

Model performance metrics for root-mean-square error ($\mathit{RMSE}$), bias ($B$), and correlation coefficient ($r$, p < .001) for timeseries (WQM) and profile (CTD) measurements. Performance metrics were computed using WQM data collected at P5, P5M, and P5E from June–August 2016 and 2017, as well as CTD casts collected at P5 from January 2016 through October 2017.

		Year	Station	Bottom Layer			Surface Layer
				RMSE	B	r	RMSE	B	r
Temp (°C)	WQM	2016	P5	1.07	−0.52	0.70	1.24	0.65	0.67
			P5M	1.04	−0.68	0.73
			P5E	0.94	−0.46	0.69
		2017	P5	0.87	−0.21	0.86
			P5M	0.78	−0.38	0.90
			P5E	0.99	−0.56	0.83
	CTD	2016–2017	P5	1.14	0.15	0.96	1.12	0.18	0.98
Salinity (ppt)	WQM	2016	P5	3.89	2.26	0.59	3.23	2.56	0.61
			P5M	3.97	2.72	0.77
			P5E	4.27	2.27	0.70
		2017	P5	4.34	−1.30	0.70
			P5M	4.57	−1.42	0.69
			P5E	5.88	−1.44	0.65
	CTD	2016–2017	P5	4.25	1.09	0.65	4.75	2.82	0.86
Density (kg m−3)	WQM	2016	P5	3.14	1.86	0.53	2.27	1.68	0.63
			P5M	3.18	2.25	0.73
			P5E	3.36	1.84	0.65
		2017	P5	3.36	−0.91	0.64
			P5M	3.47	−0.95	0.65
			P5E	4.46	−0.91	0.62
	CTD	2016–2017	P5	3.33	0.78	0.53	3.59	2.07	0.85

Fig. 4.

Observed and modeled timeseries of surface and bottom salinity (ppt) and temperature (°C) at P5 from June–August (a, c) 2016 and (b, d) 2017. Observed and modeled depth-averaged salinity (e, f) and temperature (g, h) at all CTD stations during April–October 2016 and 2017. Red open circle designates CTD station nearest to P5. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

We also assessed the model’s ability to simulate $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$, a measure of density stratification critical to establishing near-bed hypoxia in the channel. The correlation between $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ computed from WQM timeseries at P5 and modeled $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ was $r=0.65$, p < .001 (Fig. 5a). The correlation between observed and modeled $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ varied as a function of frequency, which has important implications for water quality modeling and simulating observed intermittency in benthic dissolved oxygen (Duvall et al., 2022). The time-average wavelet coherence demonstrates that the correlation between the model and WQM observations exceeded 0.8 at weekly timescales and longer (Fig. 5b). Lower coherence at hourly to daily timescales is in part because model output is a spatially averaged quantity. The analyses presented in this paper focus on changes in stratification at longer, sub-tidal timescales, which is well-described by the model. Overall, the model was able to simulate along-estuary gradients in $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ (Fig. 5c and d). Across all CTD stations, average $B$ in $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ was −0.01 ± 1.38 kg m⁻³.

Fig. 5.

(a) Timeseries of observed (black) and modeled (red) $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ at P5 from June–August 2016. (b) Time-averaged magnitude-squared wavelet coherence for observed and modeled data in (a). (c) $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ measured by CTD profiles and (d) modeled $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ during April–October 2016 and 2017. Red open circle designates CTD station nearest to P5. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

3.2. Single climate factor models

We used the calibrated present-day model to evaluate changes in density stratification ($\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$) under varying external forcing conditions for each climate change scenario (IR, T, D, SLR, and W; Table 1). Under low discharge and weak wind conditions, density stratification for all future climate models was similar to the present-day model (Fig. 6). The largest excursions from present-day $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ were observed for D, SLR, and W models and occurred during times of elevated discharge or high winds (Fig. 6). For example, before the April 2014 flood event (e.g., March 27, 2014), $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ at P5 was similar for present-day and climate change models (Fig. 7a,d,g,j). During the flood (e.g., April 17, 2014) $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ at P5 increased with SLR due to increased bottom salinity (Fig. 7h). On the other hand, $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ decreased with D because increased discharge from the Escambia River pushed the salt front further seaward (Fig. 7e). During wind events (e.g., April 29, 2014), mid-estuary stratification can be broken down by strong southerly winds that enhance vertical mixing (Fig. 7l).

Fig. 6.

(a) Modeled timeseries of $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ (kg m⁻³) at P5 for present-day model as well as climate change models (D, SLR, ${\mathrm{W}}_{10}$). (b) Escambia River discharge (m³ s⁻¹) in present-day model and increased discharge model (D). (c) Water level (m) for present-day model and sea level rise model (SLR). (d) North – south (NS) wind vector (m s⁻¹) for present-day model and increased (10%) wind speed model (${\mathrm{W}}_{10}$). Positive values indicate cross-shore winds from the south. Vertical dashed lines correspond to timestep of along-estuary salinity profiles shown in Fig. 7.

Fig. 7.

Profiles of salinity (ppt) along the CTD transect from Escambia to Pensacola Bay. Profiles are shown for (a–c) present-day, (d–f) Increased discharge (D), (g–i) Sea level rise (SLR), and (j–l) Increased wind (${\mathrm{W}}_{10}$) models. Profiles of salinity are shown for pre-flood (March 27, 2014), peak flood (April 17, 2014), and a post-flood wind event (April 29, 2014). Dashed line corresponds to location of P5.

For the 5-year model period, the distribution of $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ as a function of freshwater forcing shows similar trends (Fig. 8). The largest deviations from present-day $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ at P5 occurred during extreme flood events, defined as present-day discharge from the Escambia River exceeding the 95th percentile of the distribution (524 m³ s⁻¹). During these events, median $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ at P5 increased with SLR (13.1 kg m⁻³) and decreased with D (11.3 kg m⁻³) relative to the present-day model (11.9 kg m⁻³). When discharge was less than the 75th percentile of the distribution (206 m³ s⁻¹), $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ at P5 increased for D due to freshening of surface waters. The largest deviations from present-day $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ for the ${\mathrm{W}}_{10}$ and ${\mathrm{W}}_{50}$ models occurred on days when maximum southerly wind speed exceeded the 95th percentile of the distribution for present-day wind conditions (5.69 m s⁻¹; Fig. 9). On these days, median $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ at P5 was 9.62 kg m⁻³ for the present-day model and 8.88 and 6.02 kg m⁻³ for the ${\mathrm{W}}_{10}$ and ${\mathrm{W}}_{50}$ models, respectively.

Fig. 8.

(a) Histogram of 5-year discharge from Escambia River for 2013–2017. Dashed lines mark the 50th (121 m³ s⁻¹), 75th (206 m³ s⁻¹), and 95th (524 m³ s⁻¹) percentiles of the distribution. (b) Mean daily stratification, $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$, at P5 as a function of mean flow for the percentiles shown in (a). The bottom and top edges of the box mark the 25th and 75th percentiles of the distribution, respectively. Whiskers on each box show the minimum and maximum data values.

Fig. 9.

(a) Histogram of 5-year wind speed at NOAA PCLF1 station from 2013 to 2017. Dashed lines mark the 50th (2.30 m s⁻¹), 75th (3.59 m s⁻¹), and 95th (5.69 m s⁻¹) percentiles of the distribution. (b) Mean daily stratification, $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$, at P5 as a function of maximum speed of southerly winds (90–270°) for the percentiles shown in (a).

3.3. Combined climate change effects

Considered together ($\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{10}$), our simulations show that along-estuary distributions of salinity and $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ varied under climate change compared to current conditions. Fig. 10 shows changes in depth-averaged salinity and $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ for a two-week period in May 2016 (May 1–14, 2016). In the absence of elevated Escambia River discharge (>150 m³ s⁻¹), the largest increase in depth-averaged salinity occurred north of P5 (Fig. 10e, g). For the $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{10}$ model, salinity generally increased north of P5 independent of tidal amplitude or wind conditions. This was likely due to increased saltwater intrusion associated with sea level rise. Mean $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ increased only slightly by 0.01 kg m⁻³ for the $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{10}$ model (Fig. 10f, h). For the $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{50}$ model, mean $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ during this two-week period would decrease by 1.19 kg m⁻³ due to increased vertical mixing associated with strong southerly winds.

Fig. 10.

(a–b) North – south (NS) and east – west (EW) wind vectors (m s⁻¹) for ${\mathrm{W}}_{10}$ scenario. Positive values indicate winds from the south and west. (c–d) Discharge (m³ s⁻¹) from Escambia River for D scenario and water level (m) for SLR scenario. (e) Depth – averaged salinity (ppt) and (f) $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ (kg m⁻³) as a function of time and distance along the CTD transect from Escambia to Pensacola Bay for $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{10}$ model. (g–h) Difference between $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{10}$ and present-day models for depth-averaged salinity and $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$. Red circles correspond to location of CTD stations along the transect. Dashed line marks location of P5. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Time-averaged spatial patterns of salinity, temperature, and $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ show that climate change would push cooler saltwater from the Gulf of Mexico further up-estuary (Fig. 11). For $\mathrm{T}+\mathrm{D}+\text{SLR}+\mathrm{W}$ models, mean depth-averaged salinity in the Bay during the 2014 flood period (March 15 – June 30) increased by 1.26–1.49 ppt despite increased freshwater discharge from the Escambia River (Fig. 11a, d, g, j). Furthermore, mean depth-averaged temperature in Pensacola Bay increased by 0.38 °C for $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{10}$ model and decreased by 0.01 and 0.35 °C for $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{30}$ and $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{50}$ models, respectively (Fig. 11b, e, h, k). For the temperature only (T) model, depth-averaged temperature for the entire Bay increased by 0.81 °C. Thus, warming was mitigated in the $\mathrm{T}+\mathrm{D}+\text{SLR}+\mathrm{W}$ models by sea level rise and strengthened by southerly winds that pushed cooler saltwater from the Gulf of Mexico into Pensacola Bay. The total area of stratification exceeding 16 kg m⁻³ (i.e., the 95th percentile of the distribution of $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ for present-day conditions during the 2014 flood period) increased by 50% (+13 km²) and 11% (+3 km²) for the $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{10}$ and $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{30}$ models, respectively, and decreased by 33% (−8.5 km²) for the $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{50}$ model relative to present-day (Fig. 11c, f, i, l; Table 3). Thus, stronger wind forcing under the ${\mathrm{W}}_{50}$ model was able to breakdown increased stratification due to D and SLR during the 2014 flood period. The relative effect of changes in wind forcing depends on the strength of vertical density gradients. For example, during June–August 2016, the area where $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ exceeded 12 kg m⁻³ only increased by +1.16 km² under the $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{10}$ model relative to the present-day model and was 0 km² for $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{30}$ and $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{50}$ models (Table 3). Therefore, when freshwater inputs were low, a subtle 10% increase in wind speed was largely able to compensate for small increases in stratification due to D and SLR. This is consistent with Figs. 6–8 that show smaller changes in $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ for D and SLR models when Escambia River discharge is low.

Fig. 11.

(a, d, g, j) Depth-averaged salinity (ppt), (b, e, h, k) depth-averaged temperature (°C), and (c, f, i, l) $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ during 2014 flood event (March 15 – June 30) for present-day (a–c), $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{10}$ (d–f), $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{30}$ (g–i), and $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{50}$ (j–l) models. Black contour line marks $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}=16\text{kg}{\mathrm{m}}^{-3}$, the 95th percentile of the distribution of $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ within Pensacola Bay for the period shown.

Table 3.

Area (km²) of Pensacola Bay that exceeds a given $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ (kg m⁻³) during spring 2014, summer 2016, and summer 2017 for present-day and $\mathrm{T}+\mathrm{D}+\text{SLR}+\mathrm{W}$ models. Percent of the total Bay area (350 km²) is also reported for present-day and $\mathrm{T}+\mathrm{D}+\text{SLR}+\mathrm{W}$ models in parenthesis. Analysis does not include grid cells within Gulf of Mexico or Santa Rosa Sound.

DATE	MODEL	Vertical Density Stratification $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ (kg m−3)
		5	10	12	15	16	17	18
Mar 15 – Jun 30, 2014	present-day	232 (66%)	121 (34%)	92.9 (27%)	43.3 (12%)	26.1 (7.5%)	2.97 (0.8%)	0.00
	T	236	123	96.4	45.5	30.0	6.53	0.00
	D	226	118	93.4	47.0	32.4	6.88	0.43
	SLR	280	144	108	53.6	34.6	10.8	0.00
	${\mathrm{W}}_{10}$	222	115	87.9	40.6	19.8	0.77	0.00
	$\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{10}$	270 (77%)	136 (39%)	107 (30%)	56.8 (16%)	39.1 (11%)	17.5 (5.0%)	0.00
	$\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{30}$	253 (72%)	125 (36%)	96.8 (28%)	44.4 (13%)	29.1 (8.3%)	8.31 (2.4%)	0.00
	$\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{50}$	228 (65%)	111 (32%)	84.6 (24%)	38.4 (11%)	17.6 (5.0%)	0.34 (0.1%)	0.00
Jun 1 – Aug 31, 2016	present-day	190 (54%)	46.3 (13%)	2.04 (0.6%)	0.00	0.00	0.00	0.00
	T	191	51.2	2.61	0.00	0.00	0.00	0.00
	D	192	57.6	8.38	0.00	0.00	0.00	0.00
	SLR	213	53.4	0.00	0.00	0.00	0.00	0.00
	${\mathrm{W}}_{10}$	178	36.0	0.00	0.00	0.00	0.00	0.00
	$\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{10}$	206 (59%)	58.3 (17%)	3.20 (1.0%)	0.00	0.00	0.00	0.00
	$\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{30}$	187 (53%)	32.5 (9.3%)	0.00	0.00	0.00	0.00	0.00
	$\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{50}$	167 (48%)	13.9 (4.0%)	0.00	0.00	0.00	0.00	0.00
Jun 1 – Aug 31, 2017	present-day	215 (62%)	108 (31%)	77.5 (22%)	6.64 (1.9%)	0.00	0.00	0.00
	T	218	110	79.1	7.90	0.00	0.00	0.00
	D	212	107	78.6	11.6	0.00	0.00	0.00
	SLR	260	123	86.5	10.4	0.00	0.00	0.00
	${\mathrm{W}}_{10}$	196	97.7	67.6	1.01	0.00	0.00	0.00
	$\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{10}$	247 (71%)	118 (34%)	84.8 (24%)	11.7 (3.3%)	0.00	0.00	0.00
	$\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{30}$	217 (62%)	101 (29%)	65.9 (19%)	0.68 (0.2%)	0.00	0.00	0.00
	$\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{50}$	182 (52%)	85.1 (24%)	46.6 (13%)	0.00	0.00	0.00	0.00

Time-averaged magnitude-squared wavelet coherence provides a measure of correlation between present-day and $\mathrm{T}+\mathrm{D}+\text{SLR}+\mathrm{W}$ models and allows us to evaluate timescales impacted by climate change (Fig. 12). In general, coherence with the present-day model decreases as wind speed increases. For $\mathrm{T}+\mathrm{D}+\text{SLR}+\mathrm{W}$ models, state variables at P5 show higher coherence at longer timescales, which suggests that the impacts of climate change on state variables are episodic on the order of hours to days. At daily timescales (24 h) bottom salinity and temperature had lower coherence (0.75–0.93 and 0.59–0.83, respectively) relative to surface variables (>0.83). This was likely due to the effects of SLR and W on bottom layer temperature and salinity. Mean coherence for $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ is 0.58–0.79 and 0.84–0.96 at timescales less than and greater than 24 h, respectively. This finding is supported by results from single climate factor models showing the intermittent influence of D, SLR, and W on $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ at P5 (Figs. 6–9).

Fig. 12.

Time-averaged magnitude-squared wavelet coherence between present-day and (a) $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{10}$, (b) $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{30}$ and (c) $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{50}$ models at P5. Each spectrum represents the average of five summertime periods (May–September 2013–2017). Dashed line marks period = 24 h.

4. Discussion

Two main findings regarding the effects of climate change on estuarine stratification emerged from model simulations. First, impacts of climate change on modeled state variables vary with forcing and are frequency dependent. Uncertainties associated with climate change projections are especially relevant in evaluating the individual and combined impacts of climate-related forcings. This is particularly true for wind forcing, which will likely play an important role in modulating the spatial extent and duration of strong density gradients in and around the channel of Pensacola Bay. Intensification of coastal winds due to climate change has only more recently been considered in estuarine modeling studies (Palmer et al., 2019; Polli et al., 2021). Second, changing climate-related forcings may induce discordant effects on biological and physical controls over the formation and persistence of hypoxia in Pensacola Bay. Future expansion of coastal hypoxia may be mitigated by reduced density stratification and intrusion of cooler seawater into the Bay. These points are further elaborated below.

4.1. Uncertainty associated with climate change simulations

Overall, our results show that natural variability in physical forcing mechanisms is a large source of uncertainty regarding future changes in density stratification. As shown by others, natural variability can dominate uncertainty associated with estimates of regional climate change, particularly at shorter timescales (Kjellström et al., 2011). Natural variability may also be a large source of uncertainty for future hypoxia projections, as recently noted by Meier et al. (2021) in the Baltic Sea. Using an ecosystem model and dynamic downscaling approach, Meier et al. (2021) showed that the greatest source of uncertainty in future projections of hypoxic area was natural variability, which exceeded uncertainty associated with differences in climate models and RCPs. In addition to the uncertainty associated with changes in density stratification, future changes in hypoxia in Pensacola Bay may also depend on unknown changes in land use, unknown impacts of climate change on nutrient loadings, and unknown bioavailable fractions of nutrient inputs (Meier et al., 2019). Uncertainty associated with projected changes in sea level, freshwater discharge, wind, and temperature considered in this study are discussed below.

Combined climate factor simulations showed that sea level rise would increase salinity and at times decrease temperature despite increased atmospheric warming and freshwater discharge. This is consistent with increased saltwater intrusion, which strengthens vertical density gradients in Escambia Bay and Pensacola Bay proper when sea level rise coincides with flooding and weak wind forcing. Changes in density stratification partly depend on the magnitude of sea level rise, modeled here to be 0.48 m based on IPCC global mean projections for the RCP6.0 scenario. Global projections of sea level rise vary with warming scenario considered, and do not account for regional variability. More recent observation-based extrapolations suggest mean sea level along the Eastern Gulf will rise by 0.48 m by the year 2050 (relative to 2000), exceeding global averages largely due to higher rates of land subsidence (Sweet et al., 2022). Saltwater intrusion and vertical density gradients would be expected to increase with increased sea level rise. We did not consider changes in tidal amplitude associated with sea level rise (Khojasteh et al., 2021) or shoreline hardening (Lee et al., 2017). While dynamical downscaling has revealed changes in tidal amplitude due to increased water depth in some estuaries (Howard et al., 2019; Jackson et al., 2022), hydrodynamic modeling by Passeri et al. (2016) demonstrated negligible changes in tidal amplitude for Pensacola Bay even if sea levels rise 2 m by 2100.

Elevated freshwater discharge increased or decreased mid-estuary density stratification depending on model period considered. Projections of climate induced changes in precipitation and streamflow vary among global climate models and across a wide range of spatial and temporal scales. Here, we considered a 10% increase in freshwater discharge, which is consistent with the mean projected change for North America from 2081 to 2100 under a moderate warming scenario using an ensemble of 12 global climate models (Sperna Weiland et al., 2012). A 10% increase is generally consistent with other studies showing intensification of the hydrological cycle in the northern Gulf of Mexico (Sinha et al., 2017; Lu et al., 2020), and is within the range of observed interannual variability. However, projected changes in streamflow vastly vary among models, with estimated changes for North America ranging from −15 to +65% among the 12 models included in Sperna Weiland et al. (2012). Even within a single river basin or geographic region, models may exhibit stark dissimilarities in relative change (e.g., Najjar et al., 2009; Hovenga et al., 2016), justifying the importance of multi-model ensembles. We did not consider seasonal shifts in the timing of peak streamflow, which could have important implications for summertime hypoxia. A multi-model ensemble for nearby Apalachicola River Basin showed decreased precipitation and streamflow under future climate scenarios for the months of April–June (Hovenga et al., 2016). If seasonal shifts are consistent across northwest Florida, decreased freshwater discharge in the late spring may limit stratification and the development of near-bed hypoxia, as was observed in 2016 (Table 3; Fig. 2).

Combined climate simulations revealed that intensified sea breeze circulation would be needed to breakdown density gradients strengthened by sea level rise and increased freshwater input. The magnitude and frequency of southerly winds that enhance vertical mixing has increased from 1980 to present. The total number of observations exceeding 8 m s⁻¹ during May–August increased by 123% in 2010–2019 compared to 1980–1989 (Supplementary Fig. 2). Overall, a linear fit showed that the speed of extreme southerly winds is increasing by 0.023 m s⁻¹ per year (Supplementary Fig. 3). These findings are proceeded by observational wind analyses from the Gulf of Mexico (Liu et al., 2015) and elsewhere (Pazandeh Masouleh et al., 2019) that suggest strengthening sea breeze with climate change. In northwest Florida, future changes in local sea breeze circulation may depend in part on changes in the extent and location of the Atlantic Warm Pool (Misra et al., 2011). Despite its importance, intensification of coastal winds due to climate change has only more recently been evaluated in hydrodynamic modeling studies (Palmer et al., 2019; Polli et al., 2021). Exclusion of climate-induced changes in wind forcing may lead to overestimation of the synergistic effects of sea level rise and increased streamflow on stratification and hypoxia.

Increased temperatures due to atmospheric warming and warmer riverine discharge had a limited impact on density stratification, a finding supported by previous studies that have demonstrated stratification in Pensacola Bay results from salinity rather than temperature gradients (Hagy and Murrell, 2007). Our temperature scenario (T) did not consider changes in temperature forcing at the southern boundary due to warming of open waters outside of the model domain. Previous modeling by Lehrter et al. (2017) demonstrated that a 3 °C increase in atmospheric temperature would result in a 1.1 °C increase in depth-averaged temperature in the northern Gulf of Mexico based on data from 2006. If temperature forcing at the southern boundary is increased by 1.1 °C as implemented in our sensitivity analysis, mean depth-averaged temperature in Pensacola Bay decreases by 0.18 °C (compared to −0.35 °C shown in Fig. 11) for the $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{50}$ model relative to present-day during March–June 2014. While this slightly dampened temperature change reflects the modified open water boundary condition, high inflow of warm freshwater during the 2014 flood elevated temperatures in the Bay, which may have minimized the impact of the changed boundary condition. It is also important to note that these results do not necessarily suggest that Pensacola Bay will cool in the future, as relative change depends in part on the warming scenario considered as well as the degree of tidal exchange between the Gulf of Mexico and Pensacola Bay. For a given warming scenario, average water column temperature in the northern Gulf will rise slower than temperatures in Pensacola Bay in part due to the reduced thermal capacity of shallow waters (Oczkowski et al., 2015). Thus, increased intrusion of cooler Gulf waters due to sea level rise or intensified southerly winds will act to cool the Bay relative to the temperature only (T) scenario, particularly near the mid-Bay channel prone to hypoxia formation.

4.2. Implications for coastal hypoxia

The impacts of climate change on estuarine stratification (Table 3) have important implications for the development of hypoxia in shallow, subtropical systems. To quantify potential changes in the frequency and duration of hypoxia near the mid-Bay channel, we compared timescales of biological production and respiration to vertical mixing at P5. Rates of total production and respiration in the lower water column and benthos near the channel were estimated to be 15.2 mmol m⁻² d⁻¹ and 29.6 mmol m⁻² d⁻¹, respectively (Murrell and Caffrey, 2005). Based on these estimates, the lower water column would become hypoxic (<62.5 mmol m⁻³) after approximately 12.4 days in the absence of advective or diffusive processes and assuming dissolved oxygen is at saturation (240 mmol m⁻³) at time = 0 days. Assuming bottom waters are reoxygenated when $\mathrm{\Delta }{\mathrm{\sigma }}_{\mathrm{t}}$ < 5 kg m⁻³, the average number of days between mixing events during May–August 2013–2017 at P5 was 54.0 ± 58.0, 15.0 ± 15.6, and 7.65 ± 3.91 days for $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{10}$, $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{30}$ and $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{50}$ models, respectively, compared to 51.2 ± 59.3 days for the present-day model. For $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{30}$ and $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{50}$ models, the bottom layer at P5 was reoxygenated before respiration induced hypoxia occurred during all years except 2014, wherein stable stratification persisted throughout the summer. For present-day and $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{10}$ models, timescales of physical mixing were similar. Given the observed relationship between water column stratification and benthic hypoxia near the channel (Hagy and Murrell, 2007; Duvall et al., 2022), increased wind forcing is expected to decrease the frequency and duration of hypoxic events. For present-day and $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{10}$ models, the average number of days between mixing events is less than 12.4 days (i.e., amount of time needed for hypoxia to develop due to net respiration) for 89 and 88% of the Bay, respectively, for the three time periods in Table 3. Therefore, much of the bay does not exhibit persistent stratification and the impacts of climate change on near-bed hypoxia will likely be concentrated near the channel.

Age of water estimates support field observations showing worsened hypoxia during periods of strengthened stratification, such as summer 2014 (Fig. 2). The age of water at P5 in the bottom model layer increased by 15.4 days from June 1 to August 31 in 2014 compared to an increase of 7.93 days in 2016. Mean magnitude of the difference in water age between surface and bottom layers was 10.4 days in 2014 compared to 5.14 days in 2016, which is consistent with model results and field observations showing stronger density stratification in 2014 (Fig. 2). Previous studies have demonstrated lower phytoplankton biomass (Murrell and Caffrey, 2005; Wan et al., 2013) and decreased hypoxic extent (Hagy and Murrell, 2007) when transport time scales are shorter.

During March–June 2014, average temperature in the Bay increased by 0.38 °C for $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{10}$ model and decreased by 0.01 and 0.35 °C for $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{30}$ and $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{50}$ models, while salinity increased by 1.26–1.49 ppt. Changes in temperature and salinity for $\mathrm{T}+\mathrm{D}+\text{SLR}+\mathrm{W}$ models relative to present-day resulted in negligible differences in depth-averaged and bottom layer oxygen saturation. This is markedly different from other systems like the Chesapeake Bay, where decreasing oxygen solubility due to warming may trigger a large decline in benthic oxygen concentrations (Ni et al., 2019). Changes in temperature and salinity could theoretically alter particle settling velocities, which may worsen hypoxia by trapping additional organic matter in the system and increasing benthic respiration (Smith and Hollibaugh, 1993; Jarvis et al., 2020). However, median settling velocity for a 10 μm polystyrene bead (density = 1052 kg m⁻³; Bach et al., 2012) would decrease from 21.2 m d⁻¹ for the present-day model to 20.9 m d⁻¹ for the $\mathrm{T}+\mathrm{D}+\text{SLR}+{\mathrm{W}}_{50}$ model, or by approximately 1.5%. Thus, modeled changes in temperature and salinity would likely result in little change in particle settling for much of the system.

5. Conclusions

A three-dimensional hydrodynamic model was developed for Pensacola Bay to evaluate the effects of climate change on spatial and temporal patterns of salinity, temperature, and density. Our results demonstrate that changes in response variables to multiple climate change factors may differ from changes due to a single factor, thus it is important to evaluate multiple factors concurrently. As noted by others, downscaling global climate change projections to coastal ecosystems without considering local phenomena such as sea breeze circulation, as well as the timing and magnitude of spring discharge, may not accurately predict local ecosystem response. Our results showed that the effects of climate change vary across space and time, in part due to observed variability in external forcings during the model period that included both wet and dry conditions. Therefore, it is important to consider local ecosystem complexity as well as natural variability when evaluating potential impacts of global climate change.

Data availability

Data available upon publication. Data DOI:10.23719/1524280