Lateral connectivity between the floodplain and river channel is hypothesised to expand fish community isotopic trophic niches and increase overlap of river-floodplain food webs.  To evaluate how flood magnitude influences trophic dynamics in a large tropical free-flowing river (Roper River, Australia), we measured community isotopic niches in low and high magnitude flood years in wetland and river habitats.  Contrary to our hypothesis, isotopic niche area of the river fish community contracted following the high magnitude flood year, when compared to a low magnitude flood year. Wetland fishes maintained more similar niche space regardless of flood conditions, but showed the same pattern of contracting niche area following a large magnitude flood.  Niche overlap was lowest for river low magnitude flood and wetland high magnitude flood and highest for river and wetland high magnitude flood.  Our results suggest that river-floodplain fish communities exploit a wider isotopic range of food sources during low flood years, including enriched sources of potentially marine origin, while sharing abundant less diverse resources in high magnitude flood years. As water resources in tropical rivers are developed, our findings highlight the importance of conserving multiple dimensions of connectivity in riverine landscapes to maintain intact river-floodplain food webs.

We selected sites in the lower freshwater and upper tidal reaches of the Roper River, Northern Territory, Australia.  The Roper River is a perennial, groundwater-fed river flowing west to east into the Gulf of Carpentaria (Figure 2) through the tropical wet-dry savanna with a catchment area of 77,400 km².  Land use in the catchment is made up of Aboriginal freehold (45%) and pastoral leasehold (46%), with the remaining area occupied by national park (6%) and agriculture and mining leases (0.03%) (Watson et al., 2023).  The population of the catchment is ~2500 of which 73% are Indigenous (Watson et al., 2023).  The Roper is a unique river being unregulated, with high dry season flows sourced from the Cambrian Limestone Aquifer (CLA) and the Dook Creek Aquifer. The Roper River has the third largest median annual discharge in northern Australia, 4341 GL, and a mean annual discharge of 5557 GL (Watson et al., 2023) making it one of the largest free-flowing rivers in Australia and recognised globally (Grill et al., 2019). Based on long-term average peak wet season discharge was 1457 ML/d (n = 52), 2022 was 33% below average at 978 ML/d and 2023 was 82% higher than average at 2651 ML/d,  where 2023 provided much larger flood pulse and potential lateral connectivity than 2022 based on long-term averages (Figure 3).  Year 1 discharge did not create large overbank flows connecting floodplain wetlands to the main river channel. Of note is that over half the total flows enter the Roper River below Roper Bar, the most upstream point of detectable tidal influence, where this study took place. Two sites were sampled in the main channel of the river and four sites were sampled in adjacent wetlands (oxbow lakes) on the floodplain (Figure 2).

Fish and tissue sampling

A multi-gear approach was employed to sample the fish communities, including boat electrofishing, gill netting and hook-and-line sampling. Sampling occurred in the dry season (June 23-27 (wetland sites), August 2-4 (river sites) and 8-13 October 2022.  In 2023 surveys were 29 March – 3 April and 20-25 June (Figure 3). It should be noted that isotopic tissue turnover rates in fish muscle tissue reflects what the fish was eating at a specific time prior to sampling, which consequently does not reflect the current food sources.  For δ¹⁵N is it likely we were seeing what the fish were consuming months before sample (McIntyre & Flecker, 2006).  The retention rate for δ¹³C is slower and size dependant with higher turnover in smaller fish therefore we were likely seeing diets from 1-3 months (Vander Zanden et al., 2015; Skinner et al., 2016).  Using an ETS 5KW MBS-1DPDC-COL boat mounted electrofisher, pulsed DC output varied according to habitat, electrical conductivity, and water clarity.  Output was pulsed on and off, for a minimum of 1800 seconds on-time, at each site and time.  Fishing was carried out over as much of the edge area of each site as possible as well as two long open water shots to collect as many species and individuals from the fish community as possible.  Data on abundance and richness were not recorded due to time restraints created by sample collection and processing.  Where gill nets were used, two 20 m multi-panel gillnets with 5 m panels from 25 mm to 100 mm mesh and 2 m drop, were set one hour before sunset and observed for large animal capture.  Nets were pulled 1 hour after sunset.  Hook and line bait fishing was used opportunistically to target catfish and sharks, using cut bait taken from electrofished samples of bony bream [Nematalosa erebi (Günther, 1868)]. 

A total of 680 fin clips were analysed from 26 species across 16 families.  Two surveys were conducted in 2022 and 2023 from four wetland sites and two river sites (Supplemental Table S1).  During the low flood year of 2022 one wetland site dried completely and in the high flood year one wetland site was inaccessible due to high water during the first survey trip.  Two additional sites were sampled in early 2023 (R1 drain and WA drain) to collect fish moving across the floodplain either to wetlands or back to the main channel.  In an attempt to reduce total number of fish euthanized, tissue was collected by clipping a portion of the right pectoral fin, when large enough, otherwise a soft portion of the anal or caudal fin, whichever was in best condition, from fish >120 mm.  These fish were held in live well and measured, weighed, and sampled without anaesthesia and immediately released.  For those fish <120 mm, the whole fish was euthanised (using rapid chilling in an ice slurry) and fins were collected from the whole fish in the laboratory.

Basal food collection storage and preparation

To determine the range of isotopic variability at each site, basal food sources were collected from terrestrial vegetation, aquatic plants, algae, phytoplankton and seston.  For each basal group three replicates were collected within each wetland during each sampling occasion, one at the upstream, one at the mid and one at the downstream reach of each river or wetland site.  Summaries for all basal food sources including mean ± SE and range of δ¹³C and δ¹⁵N were calculated (Supplementary tables S2).

Vegetation

Terrestrial vegetation included collection of all leaves and grasses located within the riparian strip up to 15 m from the high-water mark. Grasses and leaves were stored separately. Leaves from each species of plant and grass at each site were combined, ground in a coffee grinder and up to 1 gram of material taken from the mixed sample and used for stable isotope analysis.  Aquatic macrophytes were divided into emergent and submerged.  The three locations within a site were then combined and homogenised.  A sample of up to 1 g was then taken and used for stable isotope analysis. 

Seston

This was intended to be a zooplankton sample but is presented as unfractioned seston or mixed seston, due to many samples being dominated by filamentous algae where it was not feasible to remove algae therefore zooplankton and algae are defined as seston for this work.   Samples were collected by towing a 500 mm mesh net for 500 m through the centre of each site at sunset. The concentrated sample was poured into 50 ml falcon tubes.  The three location samples from each site were emptied into a petri dish and most plant material was removed, then sieved using a 150 mm sieve, poured into a filter tower and filtered onto Whatman GF/F 0.45 mm pore size pre-combusted filter paper.  Filter papers were wrapped in aluminium foil and re-frozen for later processing.

Algal food sources

Algal food sources sampled included phytoplankton, periphyton and filamentous algae.  Phytoplankton was collected in 1 L containers at upstream, mid-stream and downstream locations within each site at each survey.  Each evening the water was filtered onto Whatman GF/F 0.45 mm pore size filters.  The filters were stored in aluminium foil.  Periphyton was collected using three different methods. First, toothbrush scrapes were used for the first survey in June 2022, where three 2cm² scrapes were taken and combined in one vial at each sampling location within the site.  Second, for the remainder of the surveys a scratch pad of 2 cm diameter was applied to hard surface and/or soft sediment to collect algae.  Five pads were used at each of the sample locations per site and combined by washing out the pads into a plastic bag.  Third, a core method for sediment algae was used in 2022.  We sampled sediment surfaces by dropping a cut 50 ml syringe into the sediment and using the plunger to push the core to the syringe surface and scraping the top 1 cm of algae into a 5 mL vial with water. The vials from the three locations were then combined for each site.  Algae for all the methods was extracted and filtered onto filter paper using the colloidal silica method described in Hamilton et al.,** (2005).  Briefly, a 50 mL falcon tube was filled with 21 mL of Ludox colloidal silica then diluted by adding deionized water to the 30 mL mark using a syringe. A pipette was then used to add the slurry containing the algal sample up to the 50 mL mark.  The sample was then centrifuged at 1000 rpm for 10 minutes, and a pipette was used to extract the “light fraction” upper layer into a filter tower and filtered onto pre-combusted 0.45 mm filter paper.  The filter paper was stored in aluminium foil and re-frozen until analysis.

Primary consumers

As for primary producers, primary consumers were collected at three locations within each site.  We used 3 x 5-minute sweep net samples with a net mesh of 500 mm, focusing on the most complex habitat at each sample location.  These habitats included submerged and emergent macrophyte beds, root masses, and undercut banks.  Each net sample was sorted in the field for all macroinvertebrates present and frozen in separate vials.  The three samples were combined in the lab and sorted into five orders: Diptera, Coleoptera, Ephemeroptera, Hemiptera and Odonata. Non-arthropod invertebrates collected included snails and prawns caught in the sweep nets, and freshwater mussels.  Mussels were sampled where possible and safe, using a rake modified with long tines out into the water and slowly dragged back to shore.  This worked in soft sediment with low macrophyte cover only.  All mollusc samples had soft tissue removed from the shell for analysis.  Terrestrial invertebrates, specifically Orthoptera (grasshoppers), were sampled when present.  A sweep net was used to catch grasshoppers in the vegetation adjacent to the survey sites. 

Stable Isotope Analysis

We collected fin clips (n= 680 after removing 13 outliers) sampled from 26 species across 16 families and four trophic guilds, which were analysed for δ¹³C and δ¹⁵N (Supplemental Table S1).  Fish fin clips were freeze dried and lipid extracted using a modified Bligh & Dyer (1959) method. Lipid extraction was performed to standardise the data and avoid variability between and within species caused by lipid content, which can affect δ¹³C values due to the tendency for lipids to be depleted in δ¹³C (Boecklen et al., 2011).  Lipid extraction was done by soaking samples in a 2:1 dichloromethane to methanol solution for 24 hours, then draining onto filter paper and rinsing three times with the dichloromethane to methanol solution to ensure all solvent and lipids were removed. The remaining tissue on the filter paper was collected, placed in an open test tube, and dried at 60℃ for 48 hours to evaporate any excess solvent.  The potential effect of lipid extraction on δ¹⁵N values was not verified within this study, owing to the minimal sample amounts from the sub-lethal fin clips preventing subsampling prior to lipid extraction. Given the aim of comparing between fish, such validation was deemed unnecessary, but the authors acknowledge the absolute δ¹⁵N values may be impacted.

Approximately 1500 μg of the lipid extracted tissue was weighed using a microbalance and placed into tin capsules for stable isotope analysis at Flinders University.  Samples were combusted at 950℃ using Isoprime GC5 continuous flow Isotope Ratio Mass Spectrometer with a vario ISOTOPE cube elemental analyser (Elementar Australia Pty Ltd). Three blanks and 14 standards (USGS40 (𝛿¹³𝐶=−26.39 [relative to Vienna Peedee Belemnite (VPDB)]  𝛿¹⁵𝑁=−4.52 [relative to N2 in air]) and USGS41 (𝛿13𝐶=+37.76 [relative to Vienna Peedee Belemnite (VPDB)] and 𝛿¹⁵𝑁=+47.57 [relative to N2 in air]  [Sercon Ltd, UK]) were run per tray of 96 to measure precision and correct for analytical drift between runs. The precision (average of standard deviations for both standards within each run of 96) was 0.10 for 𝛿¹³𝐶 and 0.08 for 𝛿¹⁵𝑁.  Results were exported using the IonVantage software from Elementar.  Not all samples analysed had sufficient material for accurate isotopic analysis, resulting in extreme outliers which were removed (n = 16).

Stable isotope values are written in conventional delta (d) notation as:

where X equals ¹⁵N or ¹³C, and R is the ratio of ¹⁵N/¹⁴N or ¹³C/¹²C in the sample, following the international standards of Pee Dee Belemnite for CO2, and atmospheric nitrogen for N2.

Statistical analysis

For basal food sources and fish, pairwise PERMANOVAs were applied in Primer (version 7.0.24) to test if there were significant differences in isotopic ratios between habitats and years to guide interpretation of isotopic niche comparisons.  Summary of basal food δ¹³C and δ¹⁵N values were collated in Supplementary table S2.  Isotopic niche analysis was conducted using the Bayesian Ellipse package in R SIBER v 2.1.9 (Jackson et al., 2019; R core team, 2024). Stable isotope values for δ¹³C (iso1) and trophic position (TP or δ¹⁵N-equivalent, iso2) were calculated and grouped by species, feeding guild (guild), or habitat-year combinations (Supplemental Tables S3-S4). Only species or guilds with more than three isotope observations per grouping were retained to ensure reliable ellipse estimation. Trophic position (TP) was calculated for fishes using the equation TP = 1+ (δ¹⁵N_consumer - δ¹⁵N_{basal source}) / Trophic discrimination factor (TDF) (Post, 2002; Zanden and Rasmussen, 1999), where δ¹⁵N_consumer is the δ¹⁵N value of an individual fish, and δ¹⁵N_{basal source} is the mean δ¹⁵N value of the primary producers within each site where fish were sampled (Supplemental Tables S3-S4).  To account for spatial and temporal variation in basal baseline isotopic N values, we estimated TP using a local basal mean baseline for each site.  The trophic discrimination factor (TDF), which is the stepwise increase in δ¹⁵N from diet to consumer, was estimated by three methods to test for sensitivities of the TP results to uncertainties with TDF.  A TD_fixed or constant TDF of 3.4‰ (Post, 2002) was first used.  Then a scaled TDF, based on a global metanalysis, was applied using the following equation, TP_scaled** = 1 + (δ¹⁵N_consumer** - δ¹⁵N_{basal source}) / 2.94 + 0.22 × (δ¹⁵N_{basal source}** - 3.4) (see Hussey et al., 2014).  The last method (TP_ausriv) is based on estimates by trophic guild as follows, where TDFs from basal sources to consumers are: herbivore/detritivore (3.9‰), omnivore (4.3‰), invertivore (3.4‰) and predator (5.7‰) as determined in Australian rivers by Bunn et al. (2013).  Sensitivity analysis showed there was limited variation across all three TDF methods (Supplementary Table S3-S4), so TP_fixed was used for calculation of isotope metrics and community niche polygons.

Using the δ¹³C and the TP values as described, isotopic niche areas were calculated using Bayesian Standard Ellipse Areas (SEA_B) for the fish community and presented as community polygons.  Polygon areas calculated using SEA_B correct for small sample sizes (Jackson et al., 2011) and were calculated for two separate groupings: whole community (all species, excluding any with less than three individuals), and guild.  Temporal comparisons tested for differences in niche metrics between high and low flood magnitude years, and spatially between river and wetland habitats. 

For the guild analysis, fish were divided into 4-trophic guilds as determined by Sternberg & Kennard (2014) to examine if the changes in isotopic niche area would be better represented at the guild or species level. Guilds included Herbivore/detritivore (HD) (ca > 25% plant matter), Omnivore (OM) (ca 5-25% plant matter), Invertivore (IN) and Predator (PR) (> 10% fish), food proportions taken from Sternberg & Kennard (2014).  The same script in R was run for these two additional analyses.

               To describe the community isotopic structure, Layman metrics (Layman et al. 2007) were calculated from posterior group means including δ¹³C range (CR), trophic position range (TP, based on d¹⁵N range or NR), total niche area (TA), and centroid distance (CD), as well as mean nearest neighbour distance (NND) and standard deviation of nearest neighbour distance (SDNND) as measures of packing and redundancy in the food web.  We calculated these metrics to assess differences in isotopic niche structure across river and wetland habitats (spatial) and flood magnitude years (temporal).  These results were visualized using posterior distributions of each metric as density plots with 95% credible intervals to assess uncertainty and using the lack of overlap in 95% confidence intervals (CI) as evidence of differences.  The use of Layman metrics and SEA_B polygons provided a characterization of isotopic food web structure in multidimensional space.  Layman metrics offer additional insight into the diversity, evenness, and redundancy of isotopic niches within communities, allowing for a mechanistic understanding of how connectivity and habitat variation shape food web dynamics, while the use of SEA_B provides a comparison of isotopic niche breadth across habitats and years while accounting for uncertainty and small sample sizes.

               To assess isotopic niche overlap among consumer communities across habitat and hydrological conditions, we used a Bayesian framework in the R package nicheROVER (Swanson et al., 2015).  We used δ¹³C and Trophic Position (TP_fixed) from individuals and grouped by habitat type and flood year, resulting in four combined community groups: river low flood magnitude (2022), river high flood magnitude (2023), wetland low flood (2022), wetland high flood (2023).  Posterior distributions were fit for each group's multivariate niche using the normal-inverse-Wishart prior (niw.post), generating 1000 posterior samples for each group. Groups with fewer than three individuals were excluded to ensure reliable estimation.

               We assessed niche overlap using both directional and symmetrical overlap matrices to determine the degree of isotopic niche sharing among habitat-year groups. Directional niche overlap was estimated by calculating pairwise overlap probabilities using the overlap() function with nprob = 1000 and nreps = 1000.   Directional overlap represents the probability that individuals from one group fall within the niche region of another, capturing asymmetric overlaps.  For example, group A (river / high flood) may largely fall within the group B’s (river / low flood) niche but the reverse may not apply.  This gives more ecological insight into how one group may be nested within another especially when niche breadth differs.  We applied this method as it accounts for uncertainty in niche estimates by integrating over the posterior distributions.  Directional overlaps were summarized with the mean probability of overlap between each group pair.  Results of overlap were visualized using an overlap matrix.  Symmetrical niche overlap, unlike directional overlap, is an average taken from the directional overlap matrix which compares probability of a species from group A occurring in niche space of group B and vice versa, giving more ecological interpretation.  This symmetric matrix gives an overall index of mutual niche similarity, allowing for more straightforward comparison across all group combinations.    All analyses were conducted in R version 4.4.0 (R Core Team, 2024), and follow the framework described in Swanson et al. (2015), which allows comparison of multivariate isotopic niches and associated uncertainty in a probabilistic manner.