Tropical Coastal Wetlands Ameliorate Nitrogen Export During Floods

Wetlands can increase resilience to extreme climatic events and have a key role in protection and water quality improvement in coastal ecosystems. Studies in tropical coastal wetlands at a catchment scale are scarce, and most work has been undertaken on small, temperate wetlands. In this study, we tested whether natural coastal wetlands in a tropical catchment (Tully-Murray, Queensland, Australia) could ameliorate nitrogen (N) exported to the Great Barrier Reef during a flood event. We measured denitrification rates in different types of coastal wetlands (mangroves, saltmarshes, waterbodies with macrophytes, and floodplain wetlands dominated by Melaleuca spp.) to assess their potential contribution to N losses during the 6-day duration of a flood in March 2018. Denitrification potential was variable across the landscape, and we identified “hotspots” in sub-catchments with high NO−3-N concentrations (0.4–0.6 mg L−1) and large areas of wetlands (>800 ha, >40% of the sub-catchment). These hotspots can denitrify up to 10 t of NO−3-N per day during a flood. We used our measured denitrification rates to provide input parameters for a model that includes the main biogeochemical processes affecting N transformations within wetlands (nitrification, denitrification, plant uptake, sedimentation, anammox, and mineralization), and accounts for transport via the duration, depth, and flow of water. Model simulations of a sub-catchment of the Tully-Murray indicate that flood inundation of large areas of natural wetlands (>40% of the sub-catchment area) could potentially remove 70% of the incoming NO−3-N load in the first 24 h of the flood. The management and restoration of coastal tropical wetlands could play a critical role in sustaining the health of coastal ecosystems through water quality improvement.


INTRODUCTION
Coastal wetlands are one of the most valuable ecosystems on Earth, considered essential for climate change adaptation and mitigation (Duarte et al., 2013). They provide key ecosystem services, including food provisioning, carbon sequestration, flood protection, and improvements in water quality (Barbier et al., 2011). Coastal wetlands are effective at improving water quality through three processes that remove nitrogen (N): plant uptake, soil accretion, and denitrification, with the latter accounting for most of the removal (>90%, Adame et al., 2019b,c, Figure 1). Denitrification is the conversion of nitrate (NO − 3 ) to nitrous oxide (N 2 O) and finally to nitrogen gas (N 2 ) and is the major pathway for permanent N removal from an ecosystem (Kulkarni et al., 2008). More recently, anaerobic ammonium oxidation (anammox) has been highlighted as an additional pathway for N removal through the conversion of nitrite (NO 2 ) and ammonium (NH + 4 ) to N 2 . Anammox is usually restricted to estuarine wetlands with low oxygen concentrations and accounts for 3-10% of N removal (Burgin and Hamilton, 2007;Zhou et al., 2014). Overall, denitrification is likely to be the most important pathway for N removal in coastal wetlands (Oliveira-Fernandes et al., 2012;Xiao et al., 2018), and thus a key process for the improvement of water quality by wetlands.
The potential to remove N through denitrification is closely associated with soil carbon and water NO − 3 concentration in many temperate (Piña-Ochoa and Álvarez-Cobelas, 2006;Zhou et al., 2014) and tropical wetlands (Adame et al., 2019b). Wetlands have carbon-rich soils due to high primary productivity (Mitsch and Gosselink, 2015), but NO − 3 concentrations can vary widely. The amount of NO − 3 a wetland receives is usually dependent on anthropogenic inputs (e.g., sewage, fertilizer) from the catchment (Wells et al., 2016). Additionally, wetlands obtain NO − 3 in-situ from nitrification, the conversion of NH + 4 to NO − 3 , which, in some tropical wetlands, accounts for 30% of the NO − 3 that is denitrified (Adame et al., 2019b).
Wetlands can only improve water quality if they are hydrologically connected, i.e., if they are in contact with water long enough for denitrification to occur. Thus, hydrological connectivity is critical for the provision of ecosystem services through improvement of water quality. Many tropical regions have high hydrological variability, with frequent storms and intense rainfall events that connect large areas through flooding. The inundation of tropical catchments causes the delivery of large quantities of water and nutrients into the coastal zone (Junk et al., 1989;Junk and Furch, 1993;Wallace et al., 2008). In many tropical catchments, particularly during the dry season, fertilizers can accumulate as N in the soil; after a rainfall event, the highly mobile NO − 3 is flushed into coastal wetlands (Davis et al., 2016). The inundation of coastal wetlands by floodwaters can create biogeochemical "hot spots, " which may contribute disproportionately to nutrient transformations, altering annual carbon, and N fluxes in the coastal zone (McClain et al., 2003). Denitrification fluxes occurring during floods are not easily accounted for but could be highly important.
In the Great Barrier Reef region, in the Wet Tropics of Australia, increased N loading from the catchment to the coastal zone due to intensification of agricultural activity has been highlighted as a significant stressor to the health of the reef (Hughes et al., 2015). Intense rainfall events are increasing in this region and are predicted to continue to increase as a result of climate change (ABM, 2019: 1970, http://www.bom.gov.au/, Eccles et al., 2019. After intense rainfall, large areas of coastal wetlands are inundated, creating denitrification hotspots (Tomasek et al., 2019). Thus, coastal wetlands in tropical catchments could play a significant role in ameliorating the export of N into the coastal zone during flooding events.
In this study, we hypothesized that coastal wetlands could strongly influence N exports to the coastal zone. We tested the hypothesis in a catchment in north-east Australia that has intensive agricultural land use but also has large areas of coastal wetlands (Figures 2A,C). Rates of denitrification were measured in different types of coastal wetlands (mangroves, saltmarshes, waterbodies with macrophytes, and floodplain wetlands dominated by Melaleuca spp.) to assess their potential contribution to N losses during the 6-day duration of a flood in March 2018. We used the measured denitrification rates in a model to assess N losses during the flood. The model includes the main biogeochemical processes affecting N transformations within wetlands (nitrification, denitrification, anammox, plant uptake, sedimentation, and mineralisation) and accounts for transport via the duration, depth, and flow of water. We also tested for the presence of denitrification "hotspots" where large areas of wetland are flooded by water with relatively high NO − 3 concentrations.

Study Site
The Tully-Murray Rivers are located within the Wet Tropics region of north-east Australia. They drain a catchment of 279,200 ha and discharge into the Coral Sea and into the Great Barrier Reef (Figure 2). The climate is tropical with monthly mean temperatures ranging from 22 to 34 • C (ABM, 2019(ABM, : 1907(ABM, -2018 and a mean annual rainfall of 2,700 mm (ABM, 2019(ABM, : 1871(ABM, -2018. The rivers in this region are characterized by dry periods during winter months and sporadic overbank floods in summer (between January and May) that inundate the adjacent wetlands from one to 12 days at a time (Karim et al., 2012). Contributions to flow from groundwater are also important, mainly after rainfall events, but also during the dry season (Rasiah et al., 2003). Tides in the region are classified as mesotidal with maximum amplitudes of 4 m (Lucinda Station, ABM, 2019). The catchments of the Tully-Murray Rivers are characterized by alluvial fans in the lower slopes and an alluvial plain with swamps in wet areas close to the coast (Wilson and Baker, 1990). The soil types are typically Rudosol, Kandosol, and Sodosol on the alluvial fans, Dermosol, Kurosol, Vertosol, and Sodosol on the alluvial plain, Tenosol on the beach ridges, and Organosol in swamp areas (Isbell, 2002). The catchments have a total area of wetlands of 24,500 ha, defined according to the Ramsar convention as: "Areas of permanent or periodic/intermittent inundation with water that is static or flowing, fresh, brackish or salt." To be considered wetlands, they should support plants or animals adapted and dependent on wet conditions for at least part of their life cycle and have anaerobic, undrained soils that are saturated, flooded or ponded (WetlandInfo, 2019, https:// wetlandinfo.des.qld.gov.au). Of the total wetland area of the catchments, 10,910 ha are estuarine, 9,390 ha are palustrine, and 200 ha are non-riverine waterbodies. Estuarine wetlands are influenced by tidal inundation and include mangroves, salt flats, and saltmarshes. Palustrine wetlands are vegetated swamps that are non-riverine or non-channel systems with more than 30% cover of emergent vegetation of grass, sedges, herbs, palms, or trees (Melaleuca and Eucalyptus spp., WetlandInfo, 2019).
The catchments of the Tully-Murray Rivers are characterized by steep mountains covered with tropical rainforests and a floodplain dominated by intensive cultivation of sugarcane and banana, as well as cattle grazing and pine plantations. As a result of intensive agricultural land use, mean annual N loads from rivers draining the catchments have increased significantly since European settlement. Dissolved inorganic N (DIN = NO − 3 -N + NO 2 -N + NH + 4 -N) has increased by a factor of five (from 170 to 840 t yr −1 ) and particulate N by a factor of 27, from 14 to 380 t yr −1 in the Tully River and from 4 to 250 t yr −1 in the Murray River (Kroon et al., 2012). Current annual total nitrogen (TN) exports from the catchments are estimated at 2,320 t yr −1 , of which 1,310 t yr −1 are DIN (Kroon et al., 2012).
We measured denitrification rates at eight sites representative of coastal wetlands of these catchments. Previous studies have shown no significant differences in denitrification potential between the dry and wet seasons (Adame et al., 2019b). Thus, all measurements were conducted once at each site at the end of the wet season (May to June) in either 2017 or 2018. We measured two Melaleuca spp. palustrine wetlands (Girramay and Hull Heads National park), an ephemeral Melaleuca-Eucalyptus spp. palustrine wetland (Djiru National Park), two marshes (saltmarsh in the southern of Hinchinbrook Channel and a freshwater marsh), dominated by Eleocharis spp. in Girramay National Park), a mangrove forest dominated by Brugueira guimnorriza, and a coastal lagoon (Barrett's lagoon) with water lilies (Nymphaea spp.) and emergent grasses (Figure 2A). Denitrification rates for the two Melaleuca sites have previously been published in Adame et al. (2019b). The denitrification rates were used in conjunction with a map and hydrological characteristics of a flood that occurred on March 2018 to identify hotspots of denitrification. Finally, a numerical model was developed to estimate fluxes of different forms of N (NO − 3 -N, NH + 4 -N), dissolved organic nitrogen (DON), and particulate organic nitrogen (PON) within a sub-catchment that was identified as a "denitrification hotspot." Detailed information on the methodology is given below.

Wetland Characteristics
At each site, we measured physicochemical characteristics of water and soil, including water temperature, electrical conductivity (EC), and pH with a calibrated water quality meter (ProPlus, YSI meter, OH, USA). Soil samples were analyzed for bulk density (BD) by weighing a dry sample of soil of known volume. Soils were also analyzed for N (%) and organic carbon (OC%) (EA-IRMS, Serco System at Griffith University). Water samples were collected in triplicate, filtered through a 0.45 µm membrane filter, and stored frozen before being analyzed for nutrients within the next week (colorimetric analyses; Chemistry Center, Department of Industry, Innovation and Science, Brisbane, Australia). Detection limits (mg L −1 ) were: 0.002 for NH + 4 -N, and 0.001 for NO − x -N and PO − 4 -P.

Denitrification Rates
To measure denitrification rates, we used the isotope pairing technique (Nielsen, 1992;Steingruber et al., 2001), which consists of adding enriched NO − 3 -15 N to water overlying sediments at a saturating concentration and estimating denitrification rates from 15 N-N 2 gas production. At each site, we collected intact sediment cores of ∼8 cm-depth inside Perspex tubes (4.8 cm internal diameter × 30 cm long) that were capped at the bottom with a rubber bung, filled with water collected from each site, and left to equilibrate overnight. The cores were sampled within each wetland in a location that had soil and vegetation representative of the site. The cores were taken 50 cm apart from each other and included small roots and leaf litter. For each experiment, we collected 12 sediment cores to allow three cores to be sampled at time 0, 2, and 5 h. For each experiment, we ran a blank sample of distilled water.
Nutrients were measured in three cores per treatment after flooding them with water and left overnight to equilibrate. The day after collection, the experiments were run at a nearby town (Ingham, Queensland) with similar ambient light and air temperature as in the field. The sediment cores were set in large plastic rectangular containers (1030 × 510 × 495 mm) filled with water maintained at a relatively constant temperature, which was recorded throughout the experiment (ProPlus, YSI meter, OH, USA; 29-31 • C). At the beginning of the experiment, NO − 3 -15 N was added to each core, and water samples were taken to measure NO − 3 -N concentrations before and after 15 N-NO − 3 additions. The experiments were run with final concentrations between 0.02 and 0.56 mg L −1 , all within the range of natural background conditions (Water Monitoring Information Portal, Queensland Government, https://watermonitoring.information.qld.gov.au/). Cores were topped up with water and capped to minimize headspace. Natural water movement was simulated by a stirrer bar suspended ∼3 cm above the sediment of each core driven by a magnet (rotating at ∼60-70 rpm Adame et al., 2019a. After ∼20 min, one core from each batch was sacrificed as a time zero measurement by adding 1 ml of 50% w/v of zinc chloride (ZnCl 2 ) that was mixed throughout the sediment and overlying water to halt bacterial activity. Triplicate 10 mL-water samples from each core were collected using a syringe and placed in a 12.5-ml Exetainer vial (Labco, High Wycombe, UK) with 250 µL of 50% w/v ZnCl 2 . At each time interval, a new set of cores was sacrificed. The headspace gas was analyzed by continuous-flow mass spectrometry for 28 N 2 , 29 N 2, and 30 N 2 -gas (EA-IRMS, Serco System at Griffith University).
Denitrification rates were estimated from equations of Steingruber et al. (2001), from which we calculated the following: D 15 : denitrification from labeled 15 NO − 3 as from the production rate of 29 N 2 and 30 N 2 : where r 29 and r 30 are the production rates of 29 N 2 and 30 N 2 , respectively, D 14 : denitrification from unlabeled 14 NO − 3 : D t = total denitrification or potential denitrification: D w tot : total denitrification of NO − 3 derived from the water column: where ε is NO − 3 enrichment during incubation from 15 NO 3 additions: where a and b are NO 3 concentrations after and before 15 NO − 3 addition, D w : denitrification from NO − 3 derived from the water column and corrected for tracer addition: D n : Coupled nitrification-denitrification: Denitrification rates are reported for ambient light conditions, and detection limits were 0.01 mg N m −2 h −1 .

Inundation Mapping
Between 18 and 29 March 2018, a major rainfall event occurred in the area (863 mm in 11 days; Tully Sugar Mill Station 32042, ABM, 2019) causing daily discharges in the Tully River to peak at 86,000 ML day −1 (Figure 2B, Supplementary Figure 1). During this event, the floodplain started to inundate, which usually occurs at river flows >37,000 ML day −1 (Wallace et al., 2009). This level of flooding is representative of an event with a return period of about 4 years (Wallace et al., 2009). River discharges into the coastal flooding zone are between 700,000 and 1,500,000 ML throughout an event of this magnitude (Wallace et al., 2009). Three tiles of level 1C Sentinel-2 data were acquired for 30 March 2018 and combined to generate a mosaic for the Tully-Murray catchment. The image scenes were retrieved from the European Space Agency open access data portal (https://scihub.copernicus.eu/dhus/#/home). The data were calibrated to standard surface reflectance values using the QUick Atmospheric Correction Code (QUAC, Bernstein et al., 2005) module in ENVI 5.5 (Exelis Visual Information Solutions). The flood inundation extent was quantified from the mosaic by applying an automated water extraction metric, the Modified Normalized Difference Water Index (MNDWI, Xu, 2006), which combines the green (G) and Daily stream water level and river discharge data from the Tully River (2017-2018) observed at Euramo (ID: 113006A; −17.99 S,145.94 E) were used to assess flood characteristics. Discharge and water level were integrated based on a non-parametric approach that relies on the empirical joint probability of any two hydrological units to estimate multivariate standardized indicators (MSIs) (Farahmand and AghaKouchak, 2015;Ndehedehe et al., 2016). Daily steps (493) from four different time series (i.e., mean and maximum values each of discharge and water level) were used to estimate two sets of MSIs (i.e., mean discharge/water level and maximum discharge/water level). The MSIs are interpreted similarly to the range of variability approach in which the value of one standard deviation is used to set the thresholds in the characterization of hydrological conditions (e.g., values > 1.5 are very wet and < −1.5 are very dry).

Wetland Area and Flooded Area
We used the sub-catchment areas of the Tully-Murray Rivers obtained by the stream network and a digital elevation model (DEM) as defined by the Paddock to Reef Integrated Monitoring, Modeling, and Reporting Program (P2R, Paddock to Reef program, Queensland Government, https://www.reefplan.qld. gov.au/tracking-progress/paddock-to-reef). We selected 28 subcatchments within 25 km of the coastal zone (Figure 4) and obtained their wetland area, including palustrine, estuarine, and water bodies (WetlandInfo, 2019). The area of wetlands within each sub-catchment that was flooded during the rainfall event was determined by subtracting total wetland area minus flooding area.
We estimated the volume of water overlying wetlands for one sub-catchment (SC187) located within the coastal floodplain of the lower Murray River ( Figure 4C, the mouth of the Murray River). Wetlands represented repeated information 46% of the selected sub-catchment area (Supplementary Figure 2). A Lidar image of 1-m resolution (Queensland LiDar Data, Cassowary Coast Project 2016, www.qldspatial.information.qld. gov.au/) with a vertical accuracy of 30 cm and horizontal accuracy of 80 cm was used in conjunction with the Sentinel imagery to calculate the height threshold above the main channel of the lower Murray River. In the wetland areas that were flooded, water depth was calculated by subtracting the DEM land values from the water surface elevation. Data analyses were conducted with ArcMap (v10.3 and 10.6, Esri ArcGIS, CA, USA).

Nitrogen Fate and Transport
Daily water flows (m 3 s −1 ), and dissolved inorganic N (DIN), dissolved organic N (DON) and particulate N (PN) loads (kg d −1 ) were obtained from the P2R Program for each of the sub-catchment units. The model uses the eWater CRC Source Catchments modeling framework (eWater Limited, ACT, Australia, www.ewater.org.au) to simulate sediment, nutrients and pesticide runoff within the Great Barrier Reef catchments. Simulated N outputs are obtained at the sub-catchment scale by dividing each sub-catchment into land-use types and estimating runoff from rainfall, constituent generation and a filter model (For further information, see www.reefplan.qld.gov.au/trackingprogress/paddock-to-reef). For this exercise, we obtained the daily flow of water, loads (kg d −1 ) and concentrations (mg L −1 ) of DIN, DON, and PN from 22 March to 1 April 2018.
We identified "denitrification hotspots" within subcatchments during 1 day of flooding (March 30, 2018, see inundation map Figure 2B). The hotspots were identified by multiplying the denitrification potential for wetlands within each sub-catchment (D t ; mg m −2 h −1 ), by the area of wetlands that were flooded during the day. Total denitrification was predicted from the NO − 3 concentration of the floodwater, which is a major driver of denitrification for similar wetlands in the region (Adame et al., 2019b,

Nitrogen Transformation Modeling
The fate of different forms of N (NO − 3 -N, NH + 4 -N, DON, and PON) was modeled for water flooding the sub-catchment SC187, our identified "denitrification hotspot." The sub-catchment Supplementary Table 1 has a total wetland area of 2,574 ha, of which 62% is estuarine wetland, 36% palustrine wetland, and the remainder (2%) riverine and lacustrine wetlands. The model describes the variation in the quantity of each form of N together with fine suspended sediment in the wetland (Equations 8-13). The wetland within this sub-catchment was simulated as a single box model considering the most important N processes in wetlands (Figure 1). During a flood, N enters the wetlands as rainfall and runoff. Internal releases to water within the wetland are from leaf litter and bottom-sediment release of DON and NH + 4 . Water column ammonium is converted to NO − 3 by nitrification, NO − 3 can be denitrified to N gases (N 2 , N 2 O) according to our measured rates (see above) or through anammox (NO − 2 and NH + 4 converted to N 2 ), removing N permanently from the system. Both NH + 4 and NO − 3 can be taken up by plants and algae, and water column PON (estimated as 98% of PN, Garzon-Garcia et al., 2018) can be deposited to the wetland sediment. The equations representing these processes for each nitrogen constituent in the model are given below, first for NO − 3 -N mass in the wetland as a function of time: where NO 3 is the total mass of NO − 3 -N in the wetland, C flow in NO 3 is the concentration of NO − 3 -N in the inflow, Q in and Q out are the inflow to and outflow from the wetland, respectively, A and V are the wetland area and volume, respectively, R is rainfall, C rain NO 3 is the concentration of NO − 3 -N in rainfall, k 1 is the nitrification rate, K 1 is the half-saturation constant for effect the ammonium concentration on nitrification, a 1 , a 2 , and a 3 are the areal rates of plant uptake of nitrogen, denitrification, and anammox respectively, and K a2 and K a3 are the half-saturation constants for effect of nitrate concentration on denitrification and anammox. The nitrogen conversion rates are given for a reference temperature of 20 • C with the influence of temperature (T) represented by a non-dimensional multiplier term θ T−20 . The value of the temperature multiplier was set at 1.073 (equivalent to doubling of the biogeochemical rate for each 10 • C increase in temperature) for all processes as there was not enough information to vary it among all biogeochemical processes. f(N) is a preference factor for ammonium vs. nitrate uptake: where K N is the half saturation for plant uptake Ammonium mass in the wetland is given as: where C flow in NH 4 is the concentration of NH + 4 -N in the inflow, C rain NH 4 is the concentration of NH + 4 -N in rainfall, k 2 is the ammonification rate, K 2 is the half-saturation constant for effects the DON on ammonification, a 4 is the areal rate of release of NH + 4 -N from bottom sediments, k 3 is the rate of ammonium desorption from fine suspended material <16 µm in diameter (SS 16 ), and K 3 is the half saturation constant for the ammonium desorption process.  The DON mass in the wetland is given as: where C flow in DON is the concentration of DON in the inflow, a 5 is the areal rate of release of DON from bottom sediments, k 4 is conversion of PON to DON and K 4 is the half-saturation constant for the conversion of PON to DON. For PON mass, the equation is: where C flow in PON is the concentration of PON in the inflow, a 6 is the areal rate of release of PON from leaf litter, w sPON is the settling rate of PON, and z av is the average depth of the wetland. Finally, for SS 16 , the equation is: (13) where w s is the settling rate of sediment. Table 1 provides a summary of the variables and parameters in the model. The mathematical model was solved with MATLAB routine ode45, with the initial conditions given by the mass of the four N constituents and SS 16 in the wetland. The volume of water in the wetland was determined through a water balance equating the change in volume to the net flux in and out of the wetland each day. Hourly rainfall data and mean water temperature for the flooding event were obtained from the Tully River at Euramo (Station 113006A, Queensland Water Monitoring Information Portal), with hourly evaporation rates obtained for Innisfail Aerodrome (Bureau of Meteorology Product Code IDCKWCDEA0). Time-dependent parameters (indicated by an asterisk in Table 1) were linearly interpolated with routine interp1 within the ode45 routine. The sensitivity of the model to the selected parameters was explored with a Monte Carlo simulation with 500 replicates, varying the parameters by ± 20% of the values set in the model ( Table 1). Random sampling was obtained from uniform distributions for the N conversion rates, half-saturation constants and settling rates. The proportion of TN removed by wetlands was compared among replicate results (Supplementary Figure 3).

Wetland Characteristics
Water flooding the wetlands had nutrient concentrations of (mean) 0.06 ± (standard error) 0.03 (range, 0.004-0.20) mg L −1 of NO − 3 -N and 0.13 ± 0.061 (0.02-0.53) mg L −1 of NH + 4 -N. Concentrations of PO − 4 -P were in general too low to be detected (<0.001 mg L −1 ). The water pH ranged from neutral to acidic (3.2-7.4) across the wetlands, with values close to neutral pH in estuarine wetlands (mangroves and saltmarsh). EC varied from freshwater to marine, with ranges between 5 and >25,000 µS cm −1 . The OC in the soil across the wetlands was on average 9.7 ± 4.0 (1.1-35)%, and N was 0.48 ± 0.13 (0.1-1.5)%, with highest values in the water bodies with macrophytes and lowest in estuarine wetlands ( Table 2).

Flood Characteristics
The flooding lasted ∼6 days from March 27 to April 1, 2018. In March 30, 2018, the total flooded area of the subcatchment was 15,000 ha, of which 9,421 ha were coastal wetlands (Figures 2A,B, Supplementary Figure 1). At this time, all subcatchments had some level of flooding from 0.2 to 45% of the total area inundated. This corresponds to the flooding of 43% of the total wetland area of the Tully-Murray catchment.

Denitrification Hotspots
The NO − 3 -N concentrations predicted from the Source Catchment model were highest at the sub-catchments at the mouth of the Murray River ( Figure 4B). The denitrification potential calculated from NO − 3 -N concentrations and flooded wetland area for March 30 was highest (10 t day −1 ) in subcatchments close to the mouth of the Tully and Murray-River ( Figure 4C). However, some sub-catchments with low NO − 3 -N concentrations were also denitrification hotspots due to extensive wetland areas (southernmost coastal area, Figure 4C). There were also sub-catchments with a denitrification potential close to zero, highlighting the patchiness of denitrification at the landscape scale. We estimate that for a given day during a flood, wetlands in the Tully-Murray catchment can potentially remove 48.5 t of NO − 3 -N through denitrification. If considering lower rates of denitrification (5.3 mg m −2 h −1 ; denitrification of Melaleuca spp. and mangroves, which are the dominant vegetation types), the denitrification potential is 12.0 t d −1 .

N Flows and Transformation Model
The model was run for one sub-catchment (SC187) with an area of flooded wetlands of 2,213 ha and a peak volume of water of 51,000 ML during the flood that lasted 6 days. Total N inputs to this sub-catchment during the flood were 62.2 t DIN (NO − 3 -N + NH + 4 -N), 42.4 t DON, 24.5 t PN, and 5,183 t of suspended sediment. The model simulations showed a loss of 70.4% of the NO − 3 -N that entered the wetland during the flood on the day of flooding (Figure 5). By the end of the 6 days of flooding, 85.9% of the NO − 3 -N had been removed from the wetland through denitrification and anammox processes. For NH + 4 -N, despite the internal release as water flooded the wetland, the model simulations produced a reduction of 17.6% of the load 1 day after flooding; and 19.2% after 6 days of flooding. The PON load was also reduced by 34.9% by the end of the first day of flooding and 43.6% by the end of the 6 days of flooding. After 3 days of flooding DON load increased by 7%, and after 6 days, the DON load increased by 38.8%. Predictions from the model simulation indicated that the wetlands within this sub-catchment retained 29.3% of the TN load during this flooding event.
To assess the sensitivity of the model parameters, we ran the model varying the value for each parameter by ±20%. The TABLE 2 | Nutrient concentrations in water, pH, electrical conductivity (EC), soil organic carbon (OC), nitrogen (N), and bulk density (BD), from eight coastal wetlands in tropical Australia.

Wetland type N-NH
Mangroves 0.02 ± 0.01 0.01 ± 0.003 6.7 38,000 ± 9,000 1.7 ± 0.  change in parameter values did not have a great effect on the model output, with a range in TN retention of 26.0-32.0% (Supplementary Figure 3). Increasing internal release of DON by 20.0% resulted in a reduction of TN retention from 29.3 to 21.0%. For the range of parameter values tested through the sensitivity analysis, the model consistently showed TN retention when wetlands were flooded.

DISCUSSION
During a flood with a recurrence time of ∼4-years, coastal wetlands in the tropical catchment of the Tully-Murray River remove a large portion of the incoming NO − 3 -N through denitrification. The nutrient removal was variable across the landscape, but our model simulations indicated the existence of "denitrification hotspots" corresponding to sub-catchments with high NO − 3 -N concentrations (>0.20 mg L −1 ), and large areas of wetlands (>24%) that became inundated during the flood. These "hotspots" had the potential to remove up to 10 t of NO − 3 -N d −1 through denitrification. Hotspots were located at the mouth of rivers close to National Parks, which had large areas of protected and well-managed wetlands that effectively contributed an ecosystem service of protecting the coastal ecosystem from eutrophication through N removal. In other sub-catchments, the denitrification potential was close to zero due to very small areas of flooded wetlands. Our approach provided a comprehensive spatial and temporal distribution of "denitrification hotspots, " validating that the patchiness of denitrification extends to tropical catchments and is indeed a global scale process (Groffman et al., 2009).
The importance of coastal wetlands for removing NO − 3 -N can be interpreted at different spatial and temporal scales. From a spatial perspective, at the sub-catchment scale, large areas of frequently inundated wetlands can remove most of the NO − 3 -N during a flooding event (Figure 5). At the catchment scale, they could also contribute substantially to the removal of NO − 3 -N before it reaches the coast. For instance, a flooding event of similar magnitude as the one in March 2018, i.e., a 4-year return period, was associated with a load of 423 t of DIN (Wallace et al., 2009). Based on the range of denitrification rates measured in our study, we estimate that denitrification would remove between 17 and 68% of the NO − 3 -N that would otherwise be flushed into the coastal zone (lower and higher end of the denitrification rates measured).
From a temporal perspective, this study shows that periods of high flows and flooding will increase denitrification in regions with large areas of wetlands, creating "denitrification hot times" (McClain et al., 2003). Similar results have been shown in subtropical Australia, where mangroves reduced 71% of the NO − 3 -N after a rainfall event (Wadnerkar et al., 2019). Most of the removal was suggested to be due to high denitrification rates (12 mg m 2 d −1 ); rates well within those derived from our denitrification measurements at similar NO − 3 -N concentrations (Figure 3). In riverine systems, NO − 3 -N uptake has been found to be highest after rainfall events (Wells et al., 2016) but removal rates from denitrification will also depend on the residence time and wetland area. Additionally, N loads after flooding events vary throughout the year, with highest concentrations during the first floods (Wallace et al., 2009). Thus, even though wetlands could remove large amounts of NO − 3 -N during major floods, proportionally, they will remove more NO − 3 -N per unit area during the first floods of the year, especially if they are of moderate intensity (Hansen et al., 2018).
During the dry season, much of the river flow is associated with baseflow from groundwater, and only sporadic rainfall events that generate some additional flow (Rasiah et al., 2010). Concentrations of NO − 3 -N in rivers are generally relatively low at the beginning of the dry season but increase as the season progresses with nutrient inputs accumulation in catchment soils (Davis et al., 2016). Thus, wetlands with groundwater flows or those receiving surface discharge via rivers might receive continuous inputs of water that is rich in NO − 3 -N (Karim et al., 2013). For these cases, there would be expected to be high denitrification rates throughout the year. Another major difference between the flooding season and the rest of the year is the tidal influence. During dry periods, mangroves, especially those in tidal channels, are flooded mainly by tidal water (Sadat-Noori et al., 2016), which is generally more diluted in nutrients than freshwater. Thus, removal rates could be lower during the dry compared to the wet season (Adame et al., 2010). The processes influencing nitrogen transformations are also likely to vary with rainfall; in wet periods, mangroves tend to be sinks of DIN, mostly due to microbial denitrification, but during dry periods, they can be sources of DIN associated with NH + 4 -N outputs (Adame et al., 2010), possibly due to dissimilatory NO − 3 -N reduction to NH + 4 -N (DNRA) which is common in coastal wetlands (Giblin et al., 2013). Overall, the value of these coastal wetlands for improvement of water quality at the catchment scale during dry periods is yet to be determined.
The model used to quantify N dynamics within a wetland provided insights into the main processes dominating N transformations during a flood. Denitrification was a key process which resulted in loss of a large portion of incoming NO − 3 -N loads within the first 24 h after flooding; sensitivity analysis based on the range of measured denitrification rates, revealed the same outcome. Anammox was also shown in model simulations to contribute to NO − 3 -N loss, but at a much lower rate (<10% of denitrification, Oliveira-Fernandes et al., 2012) than denitrification, resulting in TN removal of around 1% of the incoming load. Changes in NH + 4 -N and DON fluxes were even more uncertain, due to the scarcity of studies in similar wetlands and large variability in the reported rates of mineralisation and sediment releases (Fagherazzi et al., 2013). Nevertheless, it appeared that the model was not particularly sensitive to the range selected (±20% of calibrated values) for NH + 4 -N and DON fluxes, with TN removal by the wetlands varying only between 26 and 32% and concentrations in the outflow being continuously low. Physical and hydrological factors such as the surface area of the wetland and the volume of the flood are likely to be more critical compared to the variability of these parameters when assessing the role of wetlands for N removal.
There were some factors that were not included in the model that could be important and should be considered for future studies. For instance, in estuarine wetlands, dissimilatory NO − 3 -N reduction to NH + 4 -N (DNRA), is an important process in N removal (Giblin et al., 2013). The availability of NO − 3 -N and organic carbon could shift the dominance of DNRA over denitrification or vs. (Christensen et al., 2000;Hardison et al., 2015). Rates of denitrification and mineralisation in tropical wetlands could also vary greatly with oxygen availability and organic matter inputs (Chapman et al., 2019). Thus, direct field measurements of DNRA and organic N mineralisation in wetlands with different conditions will greatly improve the predictions of the fate of all forms of N during flooding. It would also be useful to differentiate N 2 and N 2 O emissions in measurements of denitrification as the latter is a potent greenhouse gas that could affect evaluations of the role of coastal wetlands in greenhouse gas emissions (Murray et al., 2018).
Extreme flooding events are increasing in tropical and temperate regions (Eccles et al., 2019), resulting in large flushes of water and nutrients into the coastal zone, which may cause algal blooms and the development of anoxic zones (e.g., Voynova et al., 2017). At a process level, increasing flood magnitude could alleviate nutrient limitation in mangroves (Lovelock et al., 2011) and increase fish production (Burford et al., 2008). Including flooding episodes will significantly improve the accuracy of estimates of fluxes of nutrients into the coastal zone, especially as flooding risk may increase in the future (Wallace et al., 2009;Abril and Borges, 2019). For instance, global models of nutrient exports to the coastal zone are still limited to the quantification of the nutrient outputs from river networks, and have not yet considered the role of coastal wetlands in ameliorating nutrient exports (Kroeze et al., 2012;Vilmin et al., 2018).
The improved understanding of the role of coastal wetlands to ameliorate N exports into the coastal zone has major management implications. The Reef 2050 Water Quality Improvement Plan (Reef 2050 WQIP, www.reefplan.qld.gov.au/) has set targets for improving water quality in the catchments that flow into the Great Barrier Reef. The desired target for reduction for DIN is 60% by 2025. Significant investment has been made to achieve this target by encouraging the adoption of improved land management practices, such as optimizing fertilizer use and application. However, in 2014, the potential for improved management practice to meet the desired endof-catchment reductions was modeled, and the conclusion was that even the adoption of best land management practices by all landholders would not be enough to meet the desired reductions. Thus, alternative nutrient management strategies need to be considered if current, and future targets are to be achieved (Waters et al., 2014). Restoration of existing wetlands or creation of new wetlands for the purpose of N mitigation have so far received relatively little attention as part of the Reef 2050 WQIP, partly because of uncertainty in the capacity of wetlands to achieve adequate levels of N removal. Our study provides an improved quantitative basis for wetlands to be assessed alongside other possible management actions to improve water quality of runoff entering the reef coastal zone.
Many waterways and floodplains in the Great Barrier Reef catchment have been drained and developed for agriculture and urban use. The regulation of river flows with dams and weirs has occurred to provide water storage for irrigation and levees to prevent floodwaters from entering agricultural lands (Dawson et al., 2017;Waltham et al., 2019). Restoration and rehabilitation of floodplain wetlands will require increased inundation frequency, depth and duration, and restoring peak flood events will be key to achieving this goal (Dawson et al., 2017). Management actions can include reconnecting waterways and their floodplains (Dawson et al., 2017;Waltham et al., 2019). Coastal wetlands within the catchments of the Great Barrier Reef are still abundant and are relatively well protected (Adame et al., 2019a), with opportunities to enhance their functionality with management actions. These wetlands provide a range of ecosystem services that are just now being incorporated into local management strategies (Adame et al., 2019a). Improved management and restoration of wetlands within tropical catchments can provide a complementary and cost-effective solution for improving water quality and the overall health of coastal and marine ecosystems.

DATA AVAILABILITY STATEMENT
The datasets generated and analyzed for this study can be found as Supplementary Data.

AUTHOR CONTRIBUTIONS
MA, MR, and MG designed the study. DH, MER, and JL designed, created, and solved the model. CN analyzed the flood hydrology. VR and GC did the geospatial analyses. MA did the field and laboratory experiments and wrote the manuscript with the help and input of all coauthors.

FUNDING
The study was funded by Griffith University and the Office of the Great Barrier Reef and Wetlands Team from the Department of Environment and Science. This work was also partially funded by the Australian Government's National Environmental Science Program, Tropical Water Quality Hub (Project 3.3.2).