Red Mangrove forest of the southeastern Florida Everglades

Authorship statement: The conception and design of the study and the acquisition of data 13 were conducted by JAH, EC-M, and LL-W. Data analyses were performed by JAH and EC-M. 14 Interpretation of the data was carried out by JAH, EC-M, LL-W, TT, and CB. The manuscript 15 was drafted by JAH with the help of EC-M. All authors have contributed to the final version of 16 the manuscript and have approved its submission to the journal. 17

INTRODUCTION wet season, and only 25% during the dry season (Duever et al. 1994). Analysis of long-term 2 4 0 (110-year) rainfall trends for South Florida has shown that the annual hydrologic regime can be 2 4 1 divided into two seasons: a wet season from May to October and a dry season from November 2 4 2 to April (Abiy et al. 2019). For the 2019 calendar year, temperature and relative humidity data 2 4 3 were collected from an eddy covariance flux tower installed at TS/Ph-7 and operational since 2 4 4 December 2016. Rainfall data were collected from a nearby meteorological station (station 2 4 5 name: "Taylor_River_at_mouth") managed by the US Geological Survey as a part of the 2 4 6 Everglades Depth Estimation Network (https://sofia.usgs.gov/eden).

4 7
Experimental Design 2 4 8 island habitats (center and edge), and season (wet and dry), as well as for the interaction 3 4 1 between these effects and season, which was used as the repeated measure. For the repeated 3 4 2 measures ANOVA, islands were nested within locations (fringe vs. interior) and treated as 3 4 3 experimental units. All effects were considered fixed, except for when testing for significant 3 4 4 differences in habitat, which included location as a random effect to account for the nested 3 4 5 structure of the sampling scheme. One-way ANOVAs were used to test for differences in soil 3 4 6 surface elevation among locations and habitats, and the interaction between them. Two-way 3 4 7 ANOVAs were carried out for all leaf functional traits and nutrient concentrations, where 3 4 8 comparisons were made across all habitat (i.e., edge and center) and season (i.e., wet and dry) 3 4 9 combinations. Tukey HSD post-hoc tests were used to identify significant pairwise comparisons 3 5 0 when ANOVAs indicated statistical differences. Repeated measures ANOVAs were performed 3 5 1 using PROC MIXED (SAS Institute, Cary, NC, USA) and the one-way and two-way ANOVAs 3 5 2 were performed in R v3.5.1 (R Core Team 2018).

5 3
We constructed linear mixed-effects models (with a Gaussian error distribution and 3 5 4 identity link function) to address our research questions. Island habitat and season were 3 5 5 included as fixed effects in the models to address questions 1 and 2, respectively, with water 3 5 6 levels and porewater salinity being also included as the continuous covariates to parse out their 3 5 7 marginal effects. We couple inference from these models to leaf nutrient analyses and our 3 5 8 measurements of the hydrological environment to inform about nutrient and water use of R. 3 5 9 mangle (question 3). Prior to model fitting, response variables were confirmed to meet the 3 6 0 assumptions of data normality. Four separate models were constructed, one for each of four 3 6 1 gas exchange variables of interest, A net , g sw , c i , and wue. For each model, fixed effects for 3 6 2 season (wet and dry), habitat (center and edge), porewater salinity and water level were 3 6 3 considered, including interaction terms for water level and porewater salinity with season. In all 3 6 4 models, random intercept terms were considered for location (i.e., fringe vs. interior), islands, 3 6 5 and islands nested within location. Random slopes were explored but determined to not 3 6 6 improve model fits. The best-fitting models were determined via stepwise model comparison 3 6 7 using AIC based on backward selecting random effects then backward selecting fixed effects, 3 6 8 as implemented with the 'lmerStep' function in the lmerTest R package (Kuznetsova et al. 2017). 3 6 9 The best fitting models included a random intercept term for islands, which helped remove 3 7 0 variability in the data because of the sampling design. Random effects for location (i.e. fringe vs. 3 7 1 interior) were insignificant, signifying that most of the random variance in the gas exchange data 3 7 2 was among islands, which we consider as the experimental unit in all mixed-effects models. 3 7 3 The mixed effect models were fit using restricted maximum likelihood estimates via the lme4 R 3 7 4 package (Bates et al. 2015). Models were evaluated using model predicting, tabling and 3 7 5 plotting functions from the sjPlot R package (Lüdecke 2018). All analyses were complete in R 3 7 6 v3.5.1 (R Core Team 2018). 3 7 7

7 8
Mangrove island micro-elevational differences and ecohydrology 3 7 9 Soil surface elevation significantly declined from mangrove island center to edge 3 8 0 habitats from -0.14 ± 0.1 m at mangrove island centers to -0.4 ± 0.02 m at mangrove island 3 8 1 edges, a mean difference of about 30 cm (F 1,20 = 108.42, p < .001; Table 1). Water levels 3 8 2 relative to the soil surface were significantly higher in the edge than in center habitats (F 1,178 = 3 8 3 178.33, p < .001), measuring on average 36.9 ± 1.4 cm in edge habitats, and 12.8 ± 1.2 cm in 3 8 4 mangrove island centers (Table 2). We recorded water levels of 0 cm (i.e., non-inundated 3 8 5 habitats) in 10% of our measurements, and those were exclusive to mangrove island centers 3 8 6 during the dry season. There was a significant effect of season (F 1,178 = 11.11, p < .001) on 3 8 7 water levels, where they increased from 17.05 ± 1.5 cm in the dry season to 30.4 ± 1.6 cm in 3 8 8 the wet season ( Figure S1).

8 9
Continuous water level data recorded at the fringe and interior mangrove zones 3 9 0 indicated a similar flooding trends between locations, with lower water levels during the dry 3 9 1 season and higher water levels in the wet season, up to 40-47 cm above the soil surface in both 3 9 2 locations ( Figure S2). Water levels at the interior mangrove forest always remained higher than 3 9 3 those registered in the fringe mangrove zone ( Figure S2). Porewater salinity was significantly 3 9 4 different between habitats (F 1,178 = 91.45, p < .001) and seasons (F 1,178 = 17.87, p < .001), with 3 9 5 lower salinity values in the center (21.5 ± 0.3) of the islands relative to the edge (25.1 ± 0.3) 3 9 6 habitats, and slightly lower porewater salinity during the dry season (22.5 ± 0.4) than in the wet 3 9 7 season (24.1 ± 0.3; Table 2, Figure S1). There was no significant interaction (F 1,178 = 0.26, 3 9 8 p > .05) between island habitats and seasons, indicating that the variation in porewater salinity 3 9 9 between habitats is independent of seasonality (Table 2). Surface water salinity was not 4 0 0 significantly different among center and edge habitats (F 1,163 = 2.36, p > .05), but increased 4 0 1 significantly from the dry to the wet season (F 1,163 = 8.97, p < .01, Table 2). There was also a 4 0 2 significant habitat-season interaction for surface water salinity, but a Tukey post-hoc HSD test 4 0 3 indicated that only island center habitats in the dry season were different from all other pairwise 4 0 4 comparisons (Table 2). 4 0 5 Rates of leaf gas exchange and their relationships to the hydrological environment 4 0 6 A net measurements ranged from 0.1 to 15.1 µmol m -2 s -1 , with 90% of the observations 4 0 7 recorded between 2 and 14 µmol m -2 s -1 (see Figure S3). g sw values were low, ranging from 4 0 8 <0.01 to 0.72 and averaging 0.1 mmol mol -1 (see Figure S3). Associated c i values ranged from 4 0 9 40 to 377 and averaged 242 µmol mol -1 , with 98% of them being greater than 150 µmol mol -1 . 4 1 0 Lastly, measured rates of wue varied between >0.01 and 0.21 mmol CO 2 mol H 2 O -1 , being 4 1 1 normally distributed about a mean value of 0.09 mmol mol -1 . 4 1 2 1 4 The linear mixed-effects model for A net included fixed effects for island habitat, porewater 4 1 3 salinity, water level, season, and an interaction term for water level with season (Figure S4 &  4  1  4  Table S4). There was substantial variation in A net rates among leaves (σ 2 of about 6 µmol m -2 s -4 1 5 1 ), and the random variation among islands was about 0.02 µmol m -2 s -1 (see Table S4). All 4 1 6 fixed effects were statistically significant (p < .05), except the interaction term, which was 4 1 7 marginally significant (p = .05) but greatly improved model fit. Mangrove edge habitats reduced 4 1 8 A net by over 2.5 µmol m -2 s -1 relative to mangrove island centers ( Figure 3). Seasonality had a 4 1 9 comparable negative effect, leading to an average decrease in A net of just over 2 µmol m -2 s -1 in 4 2 0 the wet season relative to the dry season ( Figure 3). After accounting for variation in the data 4 2 1 because of habitat and season, the marginal effects of water level and porewater salinity were 4 2 2 positive, albeit weak, leading to increases in A net of <0.1 µmol m -2 s -1 per cm increase in water 4 2 3 level ( Figure 4) or per ppt increase in porewater salinity ( Figure 5). Therefore, as water levels 4 2 4 increased, A net increased, with increases being consistent across habitats ( Figure 4); a similar 4 2 5 pattern was observed in relation to soil porewater salinity, although the magnitude of increase in 4 2 6 A net was smaller ( Figure 5). These relationships of A net with water level variability were 4 2 7 consistent across seasons, although rates of A net were depressed during the wet season ( Figure  4  2  8 3). The mixed-effects model for A net fit satisfactorily for these types of linear mixed effects 4 2 9 models modeling leaf-gas exchange data using environmental predictors, explaining 24% of the 4 3 0 variation in the data, 22% of which was explained by ecohydrological data (i.e., fixed effects) 4 3 1 (Table S4). 4 3 2 g sw was modeled using an identical mixed-effects model to that of A net (Figure S5 &  4  3  3  Table S5). Generally, rates of g sw were low, with 98% of g sw measurements being <0.2 mmol 4 3 4 mol -1 . Random variance in g sw among islands was negligible, being <0.01 mmol mol -1 . Leaf g sw 4 3 5 in edge habitats was statistically lower than that of mangrove island centers (p < .001), being 4 3 6 depressed by about 0.02 mmol mol -1 (Figure 3). Water levels did not affect rates of g sw (p > .05, 4 3 7 Figure 4, Table S4), and soil porewater salinity had a marginal effect (p = .07) on g sw , where 4 3 8 conductance increased slightly at high salinities, after accounting for the effects of other 4 3 9 environmental variables in the model ( Figure 5). The effect of season on rates of g sw was 4 4 0 significant in the model, with the wet season leading to a 0.05 mmol mol -1 decrease in 4 4 1 conductance ( Figure 3) and the interaction between water levels and season being statistically 4 4 2 significant ( Figure 4). Overall, the mixed-effects model for g sw did not fit the data as well as the 4 4 3 model for A net , in that the model only explained about 12% of the variability in the data, with 9% 4 4 4 of its explanatory power coming from the environmental predictors (Table S5).

5
Although the model selection approach was the same as the other mixed-effects models, 4 4 6 the best-fitting model for c i was different from the models for A net and g sw . The model did not 4 4 7 include a fixed effect for soil porewater salinity (which was dropped out of the model in the 4 4 8 model selection procedure) but included all the same fixed effects as the models for A net and g sw , 4 4 9 which were all statistically significant (p < .001), and a random intercept term for islands ( Table  4 5 0 S6). The fixed effect for porewater salinity was excluded from the best-fitting model because of 4 5 1 limited variation in porewater salinity in the data set, and because the other factors (e.g., season) 4 5 2 explained most of the variation in c i over time. Island edge habitats had consistently higher c i 4 5 3 values than mangrove island centers, being about 27 µmol mol -1 greater (19 to 35, 95% CI; 4 5 4 Figure 3). The marginal effect of season alone was similar in magnitude to that of habitat; the 4 5 5 wet season led to a decrease in c i of 24 µmol mol -1 (14 to 34, 95% CI) relative to the dry season 4 5 6 ( Figure 3, Table S6). Water levels, by themselves (again, the marginal effect), led to a slight 4 5 7 decrease in c i but had a positive interaction with season, indicating that the relative decrease in 4 5 8 c i due to increasing water levels was suppressed during the wet season ( Figure 4). The random 4 5 9 intercept term in the model (for islands) explained a considerable amount of variation in the data 4 6 0 (σ 2 = 128 µmol mol -1 , with ߬ ௦ ௗ = 66 µmol mol -1 ). The mixed-effect model for c i fit the poorest 4 6 1 of all four models, explaining just under 12% of the variance in c i , about 9% of which was 4 6 2 explained by data from the hydrological environment (Table S6).

6 3
Lastly, we modeled wue using a similar mixed-effects model to that of g sw . As with the 4 6 4 other linear mixed-effects models, the model included islands as a random effect. In the model 4 6 5 for wue, all fixed effects were statistically significant (p < .001); however, the fixed effects were 4 6 6 more subtle in magnitude. Similar to the model for c i , porewater salinity was not included in the 4 6 7 best-fitting model. wue values were normally distributed about a mean value of 0.09 µmol mol -1 , 4 6 8 with 83% of the data having values between 0.05 and 0.15 µmol mol -1 . Mangrove island edge 4 6 9 habitats had lower wue by 0.01 µmol mol -1 than island centers ( Figure 3). The marginal effect of 4 7 0 water level, although being statistically significant in the model, was negligible; however, the wet 4 7 1 season caused an increase in wue by 0.02 µmol mol -1 relative to the dry season, with the 4 7 2 interaction between water level and season being slightly negative ( Figure 3, Figure 4). 4 7 3 Random variation in wue structured across the eight mangrove islands measured was 4 7 4 minuscule, being < 0.01 µmol mol -1 . The model fit for the wue mixed-effects model was 4 7 5 comparable to and slightly better than the model for c i , with fixed effects explaining just over 12% 4 7 6 of the variance in the data, about 9% of which was explained using the environmental predictors 4 7 7 (Table S7).

1 9
Leaf form and leaf tissue nutrient concentrations 5 2 0 We consistently measured leaves at the second-most terminal leaf pair on the leaf 5 2 1 rosette and found a comparable range in SLA values, from about 29 to 40 cm 2 g -1 (Table 2). 5 2 2 SLA of leaves of R. mangle trees in full sun in coastal Belize ranged from 30.4 to 56.1 cm 2 g -1 , 5 2 3 increasing toward the terminal end of leaf rosettes (i.e., with decreasing leaf age) (Farnsworth 5 2 4 and Ellison 1996). There were no statistical differences in the SLA of any of the leaves we 5 2 5 collected, over time, among islands or between habitats. Moreover, all leaves had comparable 5 2 6 leaf dry masses and leaf water contents (Table 2). Thus, we have confidence that all of the 5 2 7 leaves this study measured for leaf gas exchange and used in nutrient analyses are functionally 5 2 8 comparable. Leaf carbon content was not different among leaves, measuring between 37.7 and 5 2 9 46.3 %, which is consistent with leaf carbon content of R. mangle leaves from trees in the 5 3 0 Guaratiba Reserve near Rio de Jainero, Brazil (Rodrigues et al. 2015), and potentially globally. 5 3 1 We measured leaf N concentrations between 8 and 10 mg g -1 , which were slightly 5 3 2 greater in the dry season than in the wet season and slightly greater in center than edge 5 3 3 mangrove island habitats. Differences in Leaf N were only statistically higher between the 5 3 4 center habitat in the dry season than in the edge habitat in the wet season (Table 3), pointing to 5 3 5 a potential habitat-season interaction on variation in leaf N. Leaf P contents were less than 0.63 5 3 6 mg g -1 , averaging 0.48 mg g -1 ; which is very low, considering that leaf P for tropical trees 5 3 7 averages about 1.4 mg g -1 in the TRY database (n = 2962; Kattge et al. 2020). Leaf P 5 3 8 concentrations were lower in the wet than the dry season and lower in the edge habitat than in 5 3 9 mangrove island centers ( Table 3), suggesting that freshwater inputs affect P sorption in the 5 4 0 soils, uptake by roots, and mangrove whole-plant and leaf P status, even at seasonal 5 4 1 timescales. Typically, leaf tissue N:P ratios >20 indicate P-limitation (Güsewell 2004), although 5 4 2 for wetland ecosystems, P-limitation may occur at N:P ratios close to 25 (Wassen et al. 1995). 5 4 3 We measured leaf N:P ratios, which all averaged >40, confirming strong R. mangle leaf tissue 5 4 4 P-limitation at TS/Ph-7. These results are also consistent with high N:P ratios (126)  condition of scrub mangroves in Taylor River.

4 7
Seasonal signals in R. mangle physiology with implications for ecosystem functioning 5 4 8 Because data were conveniently grouped by season to account for how precipitation, 5 4 9 temperature and other environmental drivers might affect R. mangle gas exchange at TS/Ph-7 5 5 0 throughout the calendar year, we must first discuss the effect of season before going on to 5 5 1 discuss the effects of habitat or water levels and salinity. A net varied over 2.5 µmol m -2 s -1 , and 5 5 2 g sw varied about 0.25 mol m -2 s -1 within habitats between the wet and dry seasons ( Figure 3). 5 5 3 Despite differences in A net and g sw between seasons, we found no statistical differences in c i 5 5 4 and wue between seasons, although there was some variation (Figure 3). This points to 5 5 5 habitat-specific optimization of the diffusion of CO 2 into (i.e., c i ) and the movement of water 5 5 6 vapor out of (i.e., wue) leaves ( water levels rose, and mangrove island centers experienced greater inundation levels (Figures 5 5 9 S1 & S2), resulting in decreased A net and g sw (Figure 3). A similar reduction in A net and g sw was 5 6 0 measured in mangrove island edge habitats during the wet season (Figures 3 & 4). Although 5 6 1 A net was depressed in the wet season, the effect of inundation levels on reducing A net was 5 6 2 consistent across seasons ( Figure 4). g sw showed a similar pattern to A net , being highest in 5 6 3 mangrove island centers during the dry season ( Figure 3). The effect of water levels on g sw , 5 6 4 however, resulted in increased g sw in the wet season, an effect which was tempered during the 5 6 5 dry season (Figure 4).

6 6
In the Florida Everglades, irradiance peaks in at the April and May (Barr et al. 2009), and 5 6 7 rainfall and temperature reach maxima in June, July, and August (Figure 1 A, B). Thus, 5 6 8 photosynthetic demand for water is likely highest at the end of the dry season in April and May. 5 6 9 During this time, we measured lower water levels and porewater salinity levels relative to the 5 7 0 wet season. Barr et al. (2009)   island centers in the wet and dry seasons, respectively. Indeed, photosynthetic transpiration 5 7 7 was highest in March and April ( Figure S3). Thus, photosynthetic demand for water is higher in 5 7 8 the dry season in mangrove island centers relative to edges or either habitat in the wet season. 5 7 9 The drying of the soils at slightly higher elevation island center habitats in this scrub mangrove 5 8 0 forests likely increases A net . Thus, the seasonal variation in hydrology, mainly reductions in 5 8 1 water levels and porewater salinity during the dry season, albeit coupled with an increase in 5 8 2 surface water salinity in this study ( The effect of mangrove island habitat on leaf gas exchange 5 8 9 Our results showed significant differences in soil elevation between mangrove island 5 9 0 habitats (Table 1), which had an apparent effect on R. mangle leaf gas exchange rates ( Figures  5  9  1 3, 4, 5). Changes in root biomass and productivity between the center and edge island habitats 5 9 2 drive the elevation gradient within islands because of differences in soil depth (Table 1). Overall, 5 9 3 higher total (0-90 cm depth) root biomass (5975 ± 1333 g m -2 ) and productivity (491 ± 64 g m -2 5 9 4 y -1 ) are observed in the center habitat compared to the edge (3379 ± 638 g m -2 and 323 ± 18 g 5 9 5 m -2 y -1 , respectively -Castañeda-Moya et al. 2011). With elevational differences of 5 9 6 approximately 0.3 meters between mangrove island center and edge habitats, we measured 5 9 7 clear differences in A net and g sw (Figure 3). A net was nearly 3 µmol m -2 s -1 greater at mangrove 5 9 8 island centers than edges, and g sw was >1 mol m -2 s -1 higher; these differences were attributable 5 9 9 to mangrove island habitat alone, after accounting for variation explained by water level, salinity 6 0 0 or seasonality (i.e., they are marginal differences). Associated c i concentrations were about 30 6 0 1 µmol mol -1 lower and wue was >0.01 mmol mol -1 greater at island centers than at island edges 6 0 2 (Figure 3). 6 0 3 Thus, these findings support our first hypothesis about the effect of habitat micro-6 0 4 elevation (center vs. edge) on A net , with overall greater leaf gas exchange rates at mangrove 6 0 5 island centers compared to their edges. Interestingly, the effect of habitat on R. mangle leaf gas 6 0 6 exchange rates was similar in magnitude to the effect of season (Figure 3). The magnitude of 6 0 7 variation in A net that we report in this study is slightly larger than the magnitude of variation 6 0 8 reported by Lin and Sternberg (1992) who found that A net varied up to 2 µmol m -2 s -1 between 6 0 9 scrub and fringe R. mangle trees in the nearby Florida keys. Furthermore, our A net 6 1 0 measurements with average values between 5.7 µmol m -2 s -1 (edge habitat, wet season) and 6 1 1 10.2 µmol m -2 s -1 (center habitat, dry season, Figure 3), are within the range of values reported 6 1 2 for R. mangle interior scrub (5.3 µmol m -2 s -1 ) and fringe (10 µmol m -2 s -1 ) mangroves along a 6 1 3 distinct zonation pattern in the intertidal zone in Twin Cays, Belize (Cheeseman and Lovelock 6 1 4 2004). These results demonstrate the effect that higher elevation center habitats at TS/Ph-7 6 1 5 have on alleviating inundation stress, which pervades scrub mangrove physiology, making trees 6 1 6 growing in center habitats in the dry season physiologically comparable to tall, fringe mangroves. 6 1 7 Certainly, the stress relief is short lived when water levels rise in the wet season (Table 2, 6 1 8 Figure S1), and rates of leaf gas exchange are depressed once more ( Figure 3). 6 1 9 The effect of water level and salinity on R. mangle leaf gas exchange 6 2 0 During 2019, the hydrological environment ( Figure 1C,D) at our study site was 6 2 1 seasonally dynamic (Table 1) and tended to mirror patterns in local rainfall ( Figure 1B 1992). Indeed, the difference in surface water and porewater salinity increased in the wet 6 2 7 season, with surface water salinities decreasing, despite a slight increase in porewater salinities 6 2 8 (Table 1). When data were grouped by season, edge habitats were slightly more saline (about 6 2 9 4 ppt on average) than mangrove edges (Table 1), and there were no clear differences between 6 3 0 fringe and interior scrub mangrove zones (Figures S1 & S2). 6 3 1 Rates of mangrove leaf gas exchange (i.e., A net and g sw ) typically decrease with 6 3 2 porewater salinity, especially along strong salinity gradients in the environment (i.e., 6 3 3 gradients >30 ppt, Ball 2009, Clough and Sim 1989, Lugo et al. 2007). Porewater salinity was 6 3 4 only included in the linear mixed-effects models for A net and g sw , and its effect was minimal, 6 3 5 reducing A net by <1 µmol m -2 s -1 per ppt increase in porewater salinity and not affecting g sw . Like 6 3 6 the effect of porewater salinity on A net , the effect of porewater salinity on g sw was small in 6 3 7 magnitude and consistent across seasons and mangrove island habitats ( Figure 5). The 6 3 8 minimal influence of porewater salinity on leaf gas exchange is likely due to the small seasonal 6 3 9 and spatial variations that we observed between fringe and interior mangrove zones. 6 4 0 Differences of salinity in the linear mixed effects models, using data from the 2019 calendar year are likely 6 5 0 broadly applicable in space and time across scrub R. mangle forests of the southwestern 6 5 1 Everglades. 6 5 2 The effects of inundation on R. mangle photosynthesis can be difficult to separate from 6 5 3 the effects of salinity, however the linear mixed modeling approach we used permitted doing so. 6 5 4 In typical greenhouse experiments where mangrove seedlings are grown, inundation alone has 6 5 5 little effect on several Rhizophora species photosynthesis or biomass production (Pezeshki et al. 6 5 6 1990b, Hoppe-Speer et al. 2011). Inundation may sometimes lead to increases in rates of leaf 6 5 7 gas exchange over the short term, and often interacts with salinity over time to reduce A net , g sw 6 5 8 and growth rates (e.g., Cardona-Olarte et al. 2013). Thus, water levels and flooding duration 6 5 9 are key drivers controlling A net in mangroves. For instance, findings from a long-term 6 6 0 greenhouse inundation study by Farnsworth and Ellison (1996) exemplify how short-term 6 6 1 responses of R. mangle to inundation differ from longer-term responses. Over several years, 6 6 2 high inundation levels led to steady declines in A net of up 25% for a given g sw and decreases in 6 6 3 growth rates. Results of the high water level (30-40 cm above soil surface) treatment were 6 6 4 similar to those of the low water level (10-15 cm) treatments, suggesting that R. mangle 6 6 5 physiology is optimized at inundation levels that reach just a few centimeters above the soil 6 6 6 surface at high water level (Ellison and Farnsworth 1997). Indeed, we found that the 6 6 7 intermittent flooding conditions of mangrove island centers that averaged water levels, or 10-15 6 6 8 cm above the soil surface, allowed greater A net and g sw than permanently flooded mangrove 6 6 9 island edges, which averaged water levels of 30-40 cm. Thus, the water level regime in center 6 7 0 habitats allows for mangrove soils to repeatedly flood and desiccate, which may help the 6 7 1 species to maintain optimal stem water potentials and g sw (Ball 2009, Reef andLovelock 2015 and Farnsworth 1997, Huang 2000). We measured depressed g sw during the wet season and in 6 7 6 inherent differences in the photosynthetic capacities of the leaves of mangrove trees in those 7 1 0 respective habitats. Such differences might be attributable to differences in the acclimation of 7 1 1 leaves to light intensity or because of inundation stress (e.g., lower rates of nutrient acquisition 7 1 2 or sapflow), and may result from an interaction between light and flooding to optimally tune 7 1 3 photosynthetic potential as a strategy of coping with inundation stress . 7 1 4 R. mangle water use at TS/Ph-7 7 1 5 Leaf carbon isotopic δ 13 C fractionation values reflect g sw integrated over leaf lifespan. 7 1 6 Rubisco, the photosynthetic enzyme, discriminates against the heaver δ 13 C (O'Leary 1988, 7 1 7 Farquhar et al. 1989), which occurs naturally in the atmosphere at roughly -8.5‰ and has been 7 1 8 trending more negative as the anthropogenic impact on the atmosphere intensifies 7 1 9 (Dlugokencky and Tans 2020). Thus, δ 13 C values more negative than -8.5‰ in leaf tissues 7 2 0 indicate a longer residence time of air (i.e., CO 2 ) in leaf intracellular air spaces, and thus lower 7 2 1 g sw (Marshall et al. 2007, Lambers et al. 2008. We found R. mangle leaf δ 13 C fractionation 7 2 2 values between -25 and -26‰, which were slightly more negative in the wet season relative to 7 2 3 the dry season. The seasonal differences in leaf δ 13 C fractionation were greater in magnitude 7 2 4 for island center habitats than for edge habitats, although not statistically significant ( Figure 6, 7 2 5 Table 2). Our We can compare instantaneous water use efficiency (wue) measured using our portable 7 3 8 gas exchange system with intrinsic water use efficiency from leaf carbon isotopes (WUE). Both 7 3 9 metrics were similar in range, with concordant indications that water-use efficiency in mangrove 7 4 0 island centers during the wet season is about 0.1 mmol mol -1 (Figure 3, Table 2). WUE was not 7 4 1 different between habitats during the wet season (Table 2), but wue was lower at edges than 7 4 2 their centers during the wet season (Figure 3). WUE increased in dry season relative to the wet 7 4 3 season, with differences between habitats emerging during the dry season (Table 2); whereas  7  4  4 wue was not different among seasons but showed consistent differences between mangrove 7 4 5 island center and edge habitats (Figure 3). However, the linear mixed-effects model for wue 7 4 6 attributed some of the variation in the data to fluctuations in water levels (Figure 4), showing that 7 4 7 wue increases when water levels are higher, and that this increase is more significant in the dry 7 4 8 season relative to the wet season. Therefore, water levels are related to R. mangle water use 7 4 leaf tissue were measured across habitats, with higher P resorption in mangrove island edge δ 15 N values integrate long-term processes of N sources because isotopic 8 0 0 fractionation against the heavier isotope (i.e., 15 N) occurs during N transformations and 8 0 1 interactions between biotic (e.g., mycorrhizal fungi, or bacteria) and biogeochemical (e.g., 8 0 2 nitrification, denitrification) nutrient cycling processes (Garten 1993). In our study site, patterns 8 0 3 of δ 15 N in R. mangle leaves differed drastically between mangrove habitats, with values around 8 0 4 -4 to -5‰ for mangrove edge habitats and between 0 and -1‰ for island centers, indicating 8 0 5 lower 15 N discrimination in island center habitats (Table 2, Figure 6A). These Patterns of foliar δ 15 N between mangrove ecotypes can be discerned using in situ leaf 8 1 8 nutrient content. Indeed, a direct relationship between 15 N discrimination and leaf N:P ratios of 8 1 9 R. mangle leaves previously reported for the six FCE mangrove sites, including our study site, 8 2 0 indicates that leaf N:P ratios accounted for 70% of the variability in 15 N discrimination (Mancera 8 2 1 Pineda et al. 2009 Mangrove trees can potentially adapt to nutrient shortage or localized nutrient 8 4 0 deficiencies in the soil by altering root system architecture and morphology and thus, patterns of 8 4 1 nutrient use. This plant strategy may maximize the efficiency of capturing limiting resources 8 4 2 essential for growth (e.g., N, P) from soil or surrounding open water areas in nutrient-poor 8 4 3 environments such as Taylor River, as proposed by the optimal plant allocation theory (Chapin 8 4 4 et al. 1987, Gleeson andTilman 1992). We observed a slight decrease in foliar δ 15 N during the 8 4 5 wet season ( Figure 6A, Table 3), as water levels and porewater salinity increased, suggesting 8 4 6 that N-acquisition by R. mangle via algal biofilms may be slightly greater in the dry season than 8 4 7 in the wet season. Our results contrast with those of Mancera-Pineda et al. (2009) habitats. Taking this into consideration, mangrove island centers potentially have even more 8 5 3 positive δ 15 N signatures than we found, illustrating that in the center of mangrove islands, N is 8 5 4 taken up by roots in inorganic soluble forms (e.g., porewater ammonium, nitrate) readily 8 5 5 available for plant uptake, and not biotically via root symbionts. Lastly, highly depleted (i.e., 8 5 6 negative) N isotopic values in leaf tissues is characteristic of tropical wetlands with P limitation 8 5 7 because P-limitation increases N fractionation, especially in flooded wetlands with limited P 8 5 8 pedogenesis (McKee et al. 2002, Troxler 2007, Medina et al. 2008, as is the case with the 8 5 9 scrub R. mangle forest at TS/Ph-7 where the main source of P is the ocean (Childers et al 2006). 8 6 0 Soil total P concentrations in top 10 cm of the peat soils at this site have measured 0.055 (± 8 6 1 0.01) mg cm -3 , with atomic N:P ratios of roughly 72 (± 2) (Mancera-Pineda et al 2009), which is 8 6 2 considerably lower than soils of most mangrove forests globally, but consistent with mangrove 8 6 3 forests in karstic environments (Rovai et al 2018). 8 6 4 CONCLUSIONS 8 6 5 In summary, habitat heterogeneity, resulting from micro-elevational differences in 8 6 6 mangrove tree locations on islands within the open water-mangrove island forest landscape, 8 6 7 drives variation in R. mangle leaf physiological performance. Mangrove island edge habitats 8 6 8 experience greater and more-prolonged inundation than island centers in a seasonal dynamic, 8 6 9 which leads to reductions in g sw , reduced A net, and lower photosynthetic wue. Conversely, 8 7 0 mangrove island center habitats are alleviated from inundation stress in the dry season, leading 8 7 1 to increases in A net and g sw . Interestingly, c i levels increase with increasing water levels 8 7 2 because inundation likely slows not only g sw , but the entire biochemical process of CO 2 8 7 3 assimilation, including mesophyll and lower level (i.e., cell wall, plasma membrane, cytosol) 8 7 4 conductance. Reductions in A net interact with the salinity of the water that inundates scrub R. 8 7 5 mangle trees, in theory, because g sw rates are low and primarily respond to water loss from 8 7 6 leaves rather than carbon gain. Additionally, differences in nutrient (i.e., N) acquisition and use 8 7 7 patterns among scrub R. mangle trees growing at island edges vs. centers affect leaf nutrient 8 7 8 status and photosynthetic potential. 8 7 9 Therefore, the interaction of inundation stress with mangrove island micro-elevational 8 8 0 habitat principally alters tree water and nutrient use dynamics, which appear to cascade to 8 8 1 affect leaf gas exchange rates through their effects on g sw . Predominantly, prolonged 8 8 2 inundation more than porewater salinity showed this effect in our measurements at TS/Ph-7 8 8 3 because the hydrological regime in Everglades mangrove forests is characterized by distinct 8 8 4 hydroperiod regimes across the coastal landscape, with long hydroperiods and minor 8 8 5 fluctuations in salinity throughout the year ( Figure S2). At the forest level, such physiological 8 8 6 differences in scrub mangrove functioning with habitat and the hydrological environment can 8 8 7 help to parameterize demographic and carbon flux models to forecast ecosystem trajectories in 8 8 8 response to the impacts of sea-level rise and saltwater intrusion in this coastal region. This is 8 8 9 particularly significant given the current freshwater restoration efforts in the Everglades and the 8 9 0 associated uncertainties of water management on the spatial distribution of mangrove forests in 8 9 1 the Everglades landscape with projected climate change scenarios.  Table 3. Leaf functional traits, carbon and nutrient contents and N:P ratios, nitrogen and 1 2 4 9 phosphorus resorption efficiencies and bulk isotopic signatures, and intrinsic intracellular CO 2 1 2 5 0 concentrations (c i ) and intrinsic water use-efficiency (WUE) (calculated from 13 C fractionation)