Waterbird assemblages of Australia’s largest east-coast wetland complex: environmental determinants of spatial and temporal distribution

Sampling area and sites

The climate of the study area is seasonal tropical, with mean minimum and maximum temperatures at Ayr (Fig. 1) of 21.9°C (June) and 35.3°C (February) respectively (Australian Bureau of Meteorology data, 1952–2022). Rainfall is also seasonal, with a summer wet season (here defined as January to April), spring dry season (September to December) and a transitional period overlapping the wet and dry seasons during the period of decline in freshwater flows (May to August) (Fig. S1). Annual rainfall in Ayr (mean 1887–2022, 1056 mm) is highly variable (e.g. during the study period, 366 mm in 1991 and 1363 mm in 2021).

The study was located in the ‘Burdekin–Townsville Coastal Aggregation’ (Queensland Government 2024), and on the floodplain of the Burdekin–Haughton Water Supply System (BHWSS) and irrigation scheme. This area covers ~1140 km², including 93 km² of freshwater wetlands. The aggregation extends from Ayr north-eastwards to the coast and north-westwards to Townsville and includes the Burdekin–Haughton irrigation area (800 km²) which, despite wetland losses, still includes many wetlands. This area also includes the mostly intertidal Bowling Green Bay Ramsar wetlands (GBRMPA 2013), which were excluded from our study (Fig. 1). This complex represents one of the largest concentrations of high-value freshwater, estuarine and marine wetlands in Australia (Davis et al. 2014; NQ Dry Tropics 2016).

The floodplain is fed naturally by shallow aquifers, local rainfall and runoff, and overbank flows of the Burdekin River (rarely; Alexander et al. 1999) and Haughton River, resulting in the formation of numerous seasonal swamps and shallow lakes, deep perennial riverine lagoons and extensive coastal sedgelands (Fig. S1) (Tait and Veitch 2007; GBRMPA 2013; Queensland Government 2024). The water table is close to ground level on a large proportion of the floodplain. Patches of woodland characterised by Melaleuca, Pandanus and Livistona species occur on raised ground and, along the natural drainage channels, species-rich gallery forest or mangrove forest is prevalent, except where channels have been cleared or burned. The wetlands are of significant conservation value, providing habitat for diverse plant, invertebrate and fish assemblages (Connolly et al. 2012; Pearson et al. 2019), and many bird species, some of which occur in nationally significant numbers (Driscoll et al. 2012) or are variously threatened (e.g. 17 species of shore or wetland species (Table S2; Garnett and Baker 2021)).

Agricultural development on the floodplain historically depended on rainfall and water derived from wetlands and groundwater (from the 1870s), and subsequently directly from the Burdekin River. The Burdekin Falls dam was completed in 1987 and currently supplies both the subsequently developed irrigation area and groundwater recharge. Downstream of the dam, water is delivered via the river, from which it is pumped into irrigation channels and the Haughton River, and distributed as required to crops and to recharge pits on the lower floodplain. Irrigation runoff enters drains and eventually natural waterways, thereby boosting groundwater levels and providing water to many wetlands during the dry season (Queensland Government 2018). The dam facilitated growth of the irrigation area by ~40,000 ha to a total of 80,000 ha, providing ~25% of Australia’s sugar crop (Fig. S2) (GBRMPA 2013). The dam overflows during most years but this does not affect the frequency of rare overbank flooding to the floodplain of the BHWSS (Alexander et al. 1999; Cooper et al. 2016; Queensland Government 2018).

Development of the irrigation area for sugarcane production involved clearing of vegetation, filling of shallow wetlands, levelling of land and establishment of irrigation infrastructure. According to GBRMPA (2013), this has led to as much as 57% loss of original lacustrine wetlands and 40% loss of palustrine wetlands. Conversely, extensive development of irrigation channels and drains (387 km of channels and 366 km of drains; Sunwater 2024), weirs, balancing storages and ponded pastures has maintained the total area of wetlands at ~106% of the pre-clearance area. There were 3.8× and 1.8× increases in the areas of highly modified lacustrine and palustrine wetlands since pre-clearing of farmland respectively (Queensland Government 2024) (Table S3).

Impacts or threats to wetlands (Table S4) include changes to hydrological processes, with potentially positive or negative effects on waterbirds depending on the associated reliance on seasonal or permanent wetlands; changes to water quality that may increase productivity but deleteriously alter aquatic assemblages; loss of riparian vegetation that provides nesting or roosting sites; establishment of invasive weeds, including water hyacinth (Eichhornia crassipes), para grass (Urochloa mutica) and hymenachne (Hymenachne amplexicaulis) (the latter two species having been introduced to ponded pastures) that may choke waterways and impede reaeration (reducing dissolved oxygen saturation), thereby precluding access by many fish species; and damage by feral pigs, Sus scrofa (Connolly et al. 2012; Perna et al. 2012; GBRMPA 2013; Davis et al. 2014; NQ Dry Tropics 2016; Waltham et al. 2020).

In our surveys, freshwater wetland sites were located (i) upstream of the Ramsar site and within the BHWSS, including (ii) along Barratta Creek and (iii) Horseshoe Lagoon Conservation Park (0.8 km²); (iv) at Wongaloo Conservation Park (18 km²), adjacent to the BHWSS; and (v) at the Townsville Town Common conservation area (40 km²) (Tait and Veitch 2007) (Fig. 1). The Wongaloo and Townsville Town Common wetlands comprise mainly seasonal vegetated and open-water swamps but do not receive irrigation drainage and have no sugarcane growing in the catchments. Wongaloo is subject to partial cattle grazing. Horseshoe Lagoon is an open-water habitat situated within the agricultural area and artificially perennial because it receives irrigation tailwater.

Data collection

We use the terms ‘survey’ to describe an organised series of samples (or datasets) and ‘sample’ to describe a single census or count at a single site and time. Datasets from seven surveys and several univariate and multivariate analyses were used in this study to address the five hypotheses (Table S5) as described below.

We undertook two separate surveys: (1) comprising two dry-season and two wet-season sets of samples from each of 32 sites in 1991–1993; and (2) comprising three samples from each of 25 sites in June, July and August 2003. Sampling sites were selected from an initial listing of over 400 individual wetlands according to representativeness and accessibility, from upstream to downstream and east to west on the floodplain (Fig. 1; Table S1). These ranged from 0.1 to 400 ha in area and included only discrete, freshwater/slightly brackish waterbodies such as riverine lagoons and extensive swamps. Identification and counts of waterbirds were made using binoculars and telescope from vantage points around entire wetlands or around defined portions of more extensive waterbodies. Counts were expressed as absolute numbers per sample and as density (numbers of waterbirds per ha sampled). Density and presence/absence data were used for most analyses.

We also accessed other datasets from (3) the Birdata Atlas database (BirdLife Australia 2022) for the whole study area for 1972–2021, for 2012–2021 at (4) Wongaloo and (5) Horseshoe Lagoon (Fig. 1), during which multiple surveys were undertaken each year; (6) an aerial survey by the University of New South Wales (Kingsford and Porter 2009; Kingsford et al. 2020); and (7) sites ground-truthed during an aerial survey by Driscoll et al. (2012). Breeding activity (presence of accompanied juvenile birds) was noted during surveys of Wongaloo but was not otherwise consistently recorded. The abundance of cryptic species, especially crakes and rails (Gruiformes: Rallidae) was likely underestimated in all these surveys and occurrence of rarer species may have been missed because of limitations in numbers of sampling sites and times.

In Surveys 1 and 2, we recorded up to 32 environmental variables for each sample as potential influences on waterbird assemblages (Table S6). We recorded, for each sample (site and time), the area surveyed; estimated maximum water depth; riparian native tree diversity (number of species along ~50 m of representative wetland perimeter) as a measure of condition; whether the wetland dried during the study period; distance of the wetland from the coast and agricultural influences (using Google Earth for the year sampled) as possible proxies for agricultural or saline disturbance; species richness of aquatic plants at access points; antecedent annual rainfall at Ayr; and antecedent discharge for Barratta Creek. Additionally, for each sample in Survey 1, we further characterised each site by measuring a standard suite of water quality variables in situ (temperature, conductivity, pH and dissolved oxygen (using Yellow Springs Instruments meters) and collected water samples from 20 to 30 cm below the water surface directly into sample bottles and preserved these for subsequent analysis. Bottles were selected in accordance with requirements for preservation specified by Australian Standard 2301.1. Samples for nutrient analyses were filtered using Sartorius 0.45-μm cellulose acetate filters, stored in gamma-sterilised plastic tubes and snap-frozen upon arrival at the laboratory (a commercial water analysis laboratory in the Australian Centre for Tropical Freshwater Research (currently TropWater) at James Cook University). Preservation procedures and analysis followed standard methods (APHA 2017). Variables measured included concentrations of calcium, chlorophyll a, potassium, magnesium, sodium, ammonium, nitrate plus nitrite, phosphate, total nitrogen, total phosphorus and total suspended solids. Water quality may not directly affect waterbirds but has a strong bearing on habitat quality and productivity for food organisms (plants, invertebrates and fish), and may therefore be of indirect influence. We were particularly interested in any effects on productivity through nutrient input (positive) and suspended sediments (negative), and indicators of saline conditions (suggesting coastal encroachment). Low levels of dissolved oxygen could indicate limitations on bird food organisms (fish and invertebrates). Other measures formed part of a normal water quality scan that typically accompanies most studies of aquatic ecology.

Statistical analysis

Species counts and densities were log₁₀(x + 1) transformed to normalise data for analysis, except for some graphical display of abundance. Our analyses included only samples from wetlands that had not dried unless otherwise indicated. Surveys and the analyses applied to the data are listed in Table S5.

We compared assemblage composition among the data sources using Pearson correlation analysis to check for overall agreement. Initial observations suggested that many wetland species select a broad range wetland habitats, such that we predicted that assemblages would be nested subsets of the species pool rather than substantially different – that is, assemblage composition at species-poor sites would be a subset of those at species-rich sites (Patterson and Atmar 1986), forming part of a large metacommunity along a habitat gradient (essentially, wetland size). The degree of nestedness in Surveys 1 and 2 was compared with a null model to determine whether the samples formed a nested hierarchy or were random (Joppa et al. 2010) by calculating the ‘nestedness metric based on overlap and decreasing fill’ (NODF) metric and associated statistics (Almeida-Neto and Ulrich 2011), using the online software at http://ecosoft.alwaysdata.net/.

Overall waterbird abundance, density and species richness in Surveys 1 and 2 were compared with area, depth and riparian condition using linear regression in SigmaPlot 12.5 (Systat Software, Inc., San Jose, CA, USA). We described broad trophic relationships by grouping species into functional groups determined by the predominant diet (vegetation, including seeds, tubers, etc.; invertebrates; and fish) (Barker and Vestjens 1989; Garnett et al. 2015) to provide an outline of major trophic dependencies. Biomass of individual species and functional groups was calculated as mean density × mean individual mass (from Garnett et al. 2015) for Surveys 1 and 2.

We used the Primer package (Primer ver. 7.0.23 and Permanova+ from Quest Research Limited, Massey University, Auckland, New Zealand) for multivariate analysis. We investigated the influence of normalised environmental variables on the bird assemblages in Survey 1 with distance-based linear modelling (DistLM), using the Bray–Curtis similarity measure on proportional data, given the variation in data sources. DistLM is essentially a multiple regression procedure that uses a similarity matrix of sites based on species’ densities. This seeks the most significant relationships between assemblages and environmental variables by progressively modelling the matrix against the most influential variable, taking the residuals of that relationship, modelling the next most influential variable etc. thereby taking into account correlations between variables (Anderson et al. 2008). Thirty-two environmental variables were associated with each sample (Table S6), all of which could affect bird assemblages directly or indirectly. We used multiple regression analysis in Statistix (ver. 10, Analytical Software, Tallahassee, FL, USA) to investigate the influence of the environmental variables on individual species’ densities, total density and species richness.

We included the season, month and year of the variables in the DistLM and multiple regression analyses of Survey 1 data for analysis of short- to medium-term temporal change. We used Atlas count records to identify monthly changes over a full year graphically and via repeated measures ANOVA between seasons for the aggregated data over 7 years (2015–2021) of mean total density, mean total number of species and mean abundance of more common species, using Statistix for the Wongaloo and Horseshoe Lagoon sites. We assigned three nominal seasons, based on rainfall and hydrological characteristics: Wet, Transitional and Dry, representing January–April, May–August and September–December respectively. We similarly analysed changes in the three broad functional (diet) groups. We described short-term temporal changes in abundance during wetland drying graphically for Survey 2 (transitional period only). We also examined the effects of drying on behavioural groupings (foraging methods including aerial, swimming/diving and wading) of the waterbirds that essentially relate to the depth of the water body and associated decline.

We illustrated long-term (decadal) change in assemblages using nonmetric multi-dimension scaling (NMDS) in Primer. We ordinated the annual mean proportional composition of assemblages over three decades for the five surveys that had suitable data (Table S5). We ordinated change in presence/absence data over four decades for all datasets. The effects of time, antecedent rainfall and flow on the proportional data were tested using DistLM. Antecedent rainfall and flow were represented by total annual rainfall at Ayr and discharge in Barratta Creek for the current and previous years to provide an indication of wet and dry years that may influence assemblages. Data were sourced from the Australian Bureau of Meteorology and Queensland Government Water Monitoring Portal (https://water-monitoring.information.qld.gov.au/) station 119101A (West Barratta Creek) respectively. Differences between a priori groups of samples were investigated using Analysis of Similarities (ANOSIM) in Primer.

Hypothesis 1. Waterbird assemblages are generally similar among the wetlands surveyed

Across all datasets, 100 waterbird (including shorebird) species were recorded (Table S2), of which 44 species occurred in five or more of the datasets. Twenty-six of the remaining species were shorebirds that inhabited the intertidal zone in Australia that was not included in this study. Forty-seven species were recorded during Surveys 1 and 2 (Table 1). There were surprisingly strong correlations among the datasets that recorded abundance, given the different approaches used: for all comparisons, r ≥ 0.7, except Horseshoe Lagoon, for which r = 0.55; in all cases, P < 0.001. Nevertheless, individual datasets recorded differences in assemblages: for example, Horseshoe Lagoon and Wongaloo showed strong similarities in common species such as Pacific Black Duck and Black Swan but large differences between cormorants and Hardhead, probably due to the consistently deeper habitat in the former as a result of irrigation drainage.

In Surveys 1 and 2, 24 species occurred in more than 10% of samples, with 12 of these occurring in at least 25% of the samples (Table 1). The most frequently recorded species (among times and sites) were Masked Lapwing, Pacific Black Duck, Australasian Darter and Great Egret, while the most numerically abundant species were Magpie Goose, Pacific Black Duck and Plumed Whistling-Duck. Biomass of individual species followed a similar pattern to that of abundance, with the goose and two duck species predominant. Collective mean biomass of functional groups was estimated at 5992 kg.ha⁻¹, 1249 kg.ha⁻¹ and 370 kg.ha⁻¹ respectively for vegetation, invertebrate and fish consumers.

Nestedness analysis showed that samples from Surveys 1 and 2 were highly nested (NODF index = 67.4, Z-score = 18.85, relative nestedness = 0.43, P <0.0001), indicating that all samples were nested within the entire metacommunity, with no separate groupings.

Hypothesis 2. Wetland bird assemblages vary spatially with wetland character

Abundance of waterbirds increased with wetland area but density decreased (Fig. 2a, b). Species richness increased with abundance and consequently with area (Fig. 2c). Richness increased with maximum depth to a peak at ~3.0 m, after which it declined (Fig. 2d). Richness declined with higher riparian diversity, although the relationship was weak (Fig. 2e). Reflecting the overall data, density of functional groups was negatively correlated with wetland size (r = −0.414, −0.420, −0.366 respectively for vegetation, invertebrate and fish consumers with P < 0.01 in each case). However, there were no clear relationships with depth, riparian condition or macrophyte diversity (r < 0.20, P > 0.05).

Fig. 2.

Results of regression analyses of data from Surveys 1 and 2 to demonstrate relatonships between: (a) waterbird abundance and wetland area; (b) waterbird density and wetland area; (c) waterbird species richness and wetland area; (d) waterbird species richness and wetland depth; and (e) waterbird species richness and wetland riparian integrity. Regression lines and associated r² values are indicated, based on linear models, except for peak model in (d). Symbols (some overlapping) represent all samples (sites and times: black circles, Survey 1; grey triangles, Survey 2).

Multiple regressions of densities of common species in Survey 1 tested models with temporal, locational, site and water-quality characteristics. Models with r² > 0.25 and P < 0.05 (Table 2) included those for total density, species richness and density of 15 individual species. Total density was negatively related to sodium (Na) concentration. Species richness showed positive relationships with wetland area and phosphate concentration. All 15 individual species had significant relationships with water-quality variables, with some indicators of productivity being particularly evident (chlorophyll a and the nutrients ammonium (NH₄) and phosphate (PO₄), with four, eight and four positive instances respectively) (Table S3). Na (four instances) was negative whereas distance from the coast (two instances) was positive. Other variables, with three or fewer instances, differed among species as to whether these were positive or negative influences. Several variables had no significant relationship. These included most of the direct seasonal indicators, although temperature, drying and instantaneous discharge had minor influence (together, five instances). Correlation analysis between seasonal and water-quality variables yielded none with r² > 0.08.

DistLM was likewise applied to Survey 1 bird assemblage data and environmental variables, producing a model that included all 32 variables that together explained 46.8% of the variation (r² = 0.468), although for no individual variable was r² greater than 0.060 (Table 3).

Hypothesis 3. Abundance of waterbird populations and assemblage composition vary seasonally, with maxima in the wet season

At Wongaloo, overall abundance, abundance of functional groups, and abundance and percent frequency of occurrence of many common species varied between seasons, particularly with increases in the dry season (Figs 3, S3 and S4; Table S7). The seasonal pattern was not clear at Horseshoe Lagoon, except that the Black Swan appeared to be more abundant in the drier part of the year. At Wongaloo, individual species showed varied patterns of occurrence frequency, with many species present in all months, some more prevalent in the dry season and a few migratory waders such as the Marsh Sandpiper being recorded more frequently in the summer wet season (Fig. S3). In Survey 2, patterns of occurrence frequency across three sites that dried over the 3-month period showed the expected gradual loss of species, with deepwater species leaving first, followed by waders (Table S8). Notably, Brolga remained during drying, feeding on sedge tubers. Evidence of breeding of different species at Wongaloo suggested that breeding occurred mainly during the transitional period and into the dry season (Table S9).

Fig. 3.

Monthly relative abundance of common species (% of aggregated totals over 2015–2021), total abundance and total number of species over the same period at (a) Horseshoe Lagoon and (b) Wongaloo. Differences with P < 0.05 (Tukey test following repeated measures ANOVA) between Wet (January–April), Transition (May–August) and Dry (September–December) seasons (defined by vertical dashed lines) are indicated by letters a and b. Seasons were offset one month to the right for Hardhead analysis. ANOVA results are shown in Table S4. ‘ns’, no sample.

Hypotheses 4 and 5. Assemblages vary annually according to antecedent rainfall and assemblages have changed over several decades

In DistLM, for proportional (density) data, no relationship was evident for assemblages across years (Pseudo-F = 1.975, P = 0.098), with antecedent annual rainfall (Pseudo-F = 0.240, P = 0.875) or flow (Pseudo-F = 0.383, P = 0.840); r² for the model = 0.052. Similarly, for presence/absence data, no relationship was evident for assemblages across years (Pseudo-F = 1.687, P = 0.165), with antecedent annual rainfall (Pseudo-F = 0.137, P = 0.933) or flow (Pseudo-F = 0. 401, P = 0.788); r² for the model = 0.059.

The NMDS of proportional abundance of species in assemblages over three decades also showed little evidence of temporal change but indicated, three distinct groupings of surveys: at Wongaloo, Horseshoe Lagoon and the Burdekin-Townsville floodplain (Fig. 4a). ANOSIM demonstrated significance of the overall model (global R = 0.772, P = 0.001), and pairwise differences between all group pairs (Burdekin–Townsville floodplain (BT) and Horseshoe Lagoon (HL), R = 0. 940, P = 0.005; BT and Wongaloo (W), R = 0.68, P = 0.006; HL and W, R = 0. 884, P = 0.001).

Fig. 4.

Temporal distribution of annual data aggregated for identified samples, ordinated using multidimensional scaling plot of datasets: (a) over three decades (1991–2021) – proportional mean abundance (i.e. % composition) within a survey; and (b) over four decades (1979–2021) – presence/absence data. Samples are labelled by survey or site: A – BirdLife Atlas; HL – Horseshoe Lagoon (Atlas data); BT – Burdekin–Townsville floodplain (this study); W – Wongaloo (Atlas data); DR – Driscoll et al. (2012) ground surveys; UN – University of New South Wales aerial surveys); and by year (1979–2021).

In the NMDS of presence/absence data over four decades, again there was little evidence of temporal change, but two distinct groupings emerged: Horseshoe Lagoon and a combination of the three other groups (Atlas, Wongaloo and Other) (Fig. 4b). ANOSIM demonstrated non-significance of the overall model (global R = 0. 124, P = 0. 143), but pairwise comparisons found the Horseshoe Lagoon samples to be separate from those in Atlas and Wongaloo (R = 0.530, P = 0.001), and the ‘Other’ grouping, which included samples from the remaining surveys (global R = 0.569, P = 0.002). In these analyses, the Wongaloo site, which was not affected by intensive agriculture, notably grouped with Atlas and Other sites, some of which were subject to agricultural impacts.

Hypothesis 1. Waterbird assemblages are generally similar among the wetlands surveyed

The Burdekin–Townsville freshwater wetlands provide extensive diverse habitats for a rich, and abundant waterbird fauna, estimated at 10 times the total waterbird abundance (excluding shorebirds) from the contiguous Bowling Green Bay Ramsar wetlands (Driscoll et al. 2012). In that study, four species occurred in internationally significant numbers (Cotton Pygmy Goose, Eastern Great Egret, Pacific Black Duck and Royal Spoonbill) and six further species in nationally significant numbers (Black Swan, Hardhead, Plumed Whistling Duck, Wandering Whistling-Duck, Intermediate Egret and Little Black Cormorant). The Ramsar wetlands support globally significant populations of shorebirds (Driscoll et al. 2012), some of which occurred in lower numbers in our study area. Our data correlate strongly with these and other records, despite differences in sampling design and extent, and the three- or four-decade span of the data. The most abundant species by numbers and biomass were vegetation feeders (Magpie Goose and Plumed Whistling-Duck) or generalist vegetation/invertebrate feeders (Pacific Black Duck) (Barker and Vestjens 1989; Garnett et al. 2015), which greatly exceed the biomass of specialist invertebrate feeders and piscivores, reflecting the expected trophic pattern of productivity. The populations of the grazers, Magpie Goose and Plumed Whistling-Duck, were likely boosted by the terrestrial food supply in well-watered grassland, including sugarcane headlands. Estimates of population size (Driscoll et al. 2012) produced one of the largest waterbird counts of eastern Australian wetlands when compared with 20 other important sites (Kingsford and Porter 2009).

Nestedness analysis found that individual assemblages were subsets of the entire set of samples, with no major habitat-based divisions. Nearly all species were represented in the larger wetlands, explaining the high level of nestedness. However, the inclusion of the Ramsar site would probably have shown a separation between the waterbird- and shorebird-dominated assemblages (Driscoll et al. 2012). Therefore, the broad requirements of many waterbird species are likely accommodated by the different wetland types, especially the larger ones, although some sites show distinct signatures in the ordinations. For example, samples from Horseshoe Lagoon were separated from the remaining samples, likely due to its perennial nature resulting from irrigation drainage.

Hypothesis 2. Wetland bird assemblages vary with wetland character

Abundance and species richness of waterbirds and functional groups increased with wetland area, as expected from the species richness/area relationship generally, and reflected by observations globally, such as in Tunisia (Hamza and Selmi 2018), Amazonia (Cintra 2019) and China (Li et al. 2019). Sites with high species richness were large, with open water. Habitats included large, small, shallow, deep, permanent and temporary types, together providing most of the habitats required by the waterbirds. However, overall densities decreased with increasing area, as expected if the number of waterbirds was fixed. Additionally, the area of a particular habitat does not necessarily increase linearly with wetland size. Population size may also be limited by breeding success locally or further afield but this seems unlikely given the extent of the wetland/grassland complex for ground-nesting waterbirds and patchy woodland, gallery forest and extensive mangroves for those nesting in trees. Apart from the shorebird migrants, many of the common wetland species recorded are known or expected to breed within the wetland complex. The wetlands are also likely to provide a dry-season refuge for waterbirds whose breeding sites are elsewhere, as some of the seasonal data suggest. Waterbirds are likely to be subject to predation by native fauna and feral pigs but the impact is unknown.

The hump-shaped relationship of species richness with depth probably occurred because suitable habitat for many species was proportionately diminished in deeper waters. This was very clear for riverine lagoons that only had small areas of shallow water. Depth is a critical issue: for example, a global review indicates that a wide diversity of depths boosted diversity, with shorebirds restricted to less than 15 cm depth, dabbling ducks and large waders to about 25 cm and diving birds using much deeper waters (Ma et al. 2010). In the study area, dabbling ducks also occurred in deeper water in which submerged plants reached the surface. The Burdekin–Townsville complex provides large areas of shallow and deeper waters, thereby providing suitable feeding habitat for waders, dabblers and diving birds.

A negative relationship between bird species richness and riparian integrity was weak, possibly because (i) riparian condition declined with increasing area, as riparian vegetation was less well defined in the larger wetlands or (ii) waterbirds essentially prefer more open areas. Nevertheless, the local riparian vegetation and discrete woodland patches are important as breeding and roosting sites for several species. Total density, species richness and density of common species correlated with environmental variables (model r² > 0.250), including those relating to water quality, and especially indicators of productivity. There were no negative relationships with water quality, possibly because deep riverine lagoons, where weed infestation can cause poor water quality (Waltham et al. 2020), do not normally support many waterbirds. However, even deep lagoons can be attractive to Magpie Geese that feed on the invasive water hyacinth, and to predators such as Comb-crested Jacana and egrets that feed on animals found among hyacinth plants.

DistLM analysis of the assemblages explained ~47% of the variation in the data in a model that included all the environmental variables, each of which was very weakly associated with the bird resemblance matrix, again indicating little evidence of strong reactions to human disturbance. Therefore, although agriculture, pollution and invasive species are implicated in the degradation of wetlands in southern Queensland (Bino et al. 2016), and in our study area the same is true for aquatic invertebrates and fish (Connolly et al. 2012; Pearson et al. 2019; Tibby et al. 2019; Waltham et al. 2020), the picture was less clear for the bird assemblages. Whether or not waterbirds are indirectly affected by such impacts is unknown. Water resource development has had a significant physical impact on the Burdekin–Townsville wetlands, as elsewhere (Kingsford and Porter 2009), yet substantial assemblages continued to use the wetlands.

The weak influence of anthropogenic factors on waterbird assemblages has been reported elsewhere (Daniel et al. 2021; Khaliq et al. 2019). In the Burdekin–Townsville area, the sheer extent and diversity of the mosaic of wetlands may absorb or buffer impacts and while this may not be true of highly modified wetlands such as Horseshoe Lagoon, even this wetland provided suitable habitat for many species. Additionally, waterbird assemblages may benefit from increased wetland habitat introduced by irrigation delivery and drainage channels, ponded pastures and artificial storages, possibly offsetting wetland loss due to land clearing and levelling, as may be the case for farm dams and rice fields (Hamilton et al. 2017; Herring et al. 2019). Nevertheless, active management is required for the whole ecosystem, including control of groundwater levels and salinity, with sea-level rise a developing threat in the region (GBRMPA 2019).

Hypothesis 2 was therefore largely supported with regard to morphology but, despite signals of the impacts of agriculture, the impacts were less substantial than expected for the bird assemblages. Cumulative impacts on waterbird assemblages may not yet be evident, although this is unlikely, given that three decades have passed since the major expansion of sugarcane cultivation (see below).

Hypothesis 3. Abundance and assemblage composition vary seasonally

A seasonal pattern was not clear at Horseshoe Lagoon, in which water levels were maintained by irrigation runoff. At Wongaloo, many common species varied seasonally in abundance and occurrence frequency, typically with a dry-season peak or no peak, except that migratory shorebirds mostly occurred during the wet season. Differences between years were inconsistent, indicating that no major trend or driver of change was operating. Our finding of reduction in abundance of some species during the wet season is likely due to local dispersal as habitats expanded or movement to different areas for breeding, as occurs in the Australian Northern Territory (Morton et al. 1990a, 1990b, 1993a, 1993b). We found that waterbirds subsequently returned to the wetlands for a great deal of the dry season, trailing off as wetlands dried and diminished in extent, in which case there was a succession of waterbirds moving away, as expected. Seasonal movements were not as clear as in drier country (Halse et al. 2005; Kingsford et al. 2010), contrasting with the greater seasonality that is typical of inland Australia, where replenished habitat is exploited during and following the wet season (Kingsford et al. 2017; Caley et al. 2022) or following infrequent wet years (Halse et al. 2005).

Therefore our second hypothesis was supported, except that maximum abundances occurred after and not during the wet season.

Hypothesis 4. Assemblages vary among years according to antecedent rainfall

This hypothesis was not supported because there was little evidence that antecedent rainfall or flow were implicated in interannual variation among assemblages or individual species. As noted above, the Burdekin–Townsville wetlands are generally seasonal, although many are sustained through the dry season by normal or rising groundwater levels and irrigation tailwater that reduce interannual variability, in contrast with a large proportion of the continent.

Hypothesis 5. Assemblages have changed over several decades as a result of floodplain development

We summarised a substantial quantity of data but acknowledge that the different sources from different times may preclude substantiation of any conclusions. For this reason we were conservative in our approach and cautious in our interpretation. We found no evidence for bird assemblage change over the entire study period (three decades of proportional data, four decades of presence/absence data), despite the extensive agricultural development of the study area. Cumulative impacts were possibly not yet evident, except that it was approximately three decades since the major expansion of sugarcane growing. Undoubtedly, many wetlands were lost to or degraded by agricultural development (GBRMPA 2013; Davis et al. 2014; Waltham et al. 2020) but overall this loss appeared to be mitigated by the extent and diversity of remaining or novel wetlands, including modified and artificial wetlands as noted above. Using a rich dataset, Colloff et al. (2015) likewise demonstrated mixed trajectories or only minor decline in integrity of wetlands in the Murray–Darling area. We had limited data from before the expansion of intensive agriculture involving land clearing, levelling and irrigation development. However, the inter-decadal samples were collected in a similar timeframe to the agricultural development but no association was found. Additionally, some sites (e.g. Wongaloo and adjacent sites) were unaffected by agricultural development, providing control sites that corroborate our conclusion by grouping with our survey sites in multivariate analyses.

We should note that the bird assemblage data that we record are insufficient to track gradual trajectories in the occurrence of rarer species, for which targeted surveys would be required. For example, the Australian Painted Snipe declined nationally during the study period with a recent Australian population estimate of only 340 (Rogers et al. 2021). Lavery and Seton (1974) recorded this species as ‘uncommon’ for our study area, while the Birdata database has only two records. Our broad surveys were inadequate to assess population size or change for this or other rare species.