The Mitchell River catchment is situated in the wet-dry tropics of northern Queensland, Australia, and covers an area of approximately 72 000 km2 (Petheram et al. 2018). The westward flowing main channel of the Mitchell River stretches from the headwaters in the Daintree rainforest in the east to the river mouth in the Gulf of Carpentaria in the west. Major tributaries that flow into the Mitchell River include the Walsh, Lynd, Alice and Palmer rivers (Fig. 1). Geology and river form vary throughout the catchment, with the eastern third comprised of bedrock varying between sedimentary, granitic and volcanic lithology (Batlle-Aguilar et al. 2014). An alluvial delta megafan at lower elevation spreads west from the confluence of the Mitchell and Palmer Rivers, producing a network of braided channels and creeks on the floodplain (Rustomji et al. 2010). Rainfall in the Mitchell catchment is highly seasonal, with only 4% of annual rainfall (on average) falling across the catchment during the dry season, from May to October (Petheram et al. 2018). Historical flow data shows that the main channel of the Mitchell River experiences perennial flow most years, while the Palmer and Walsh rivers experience cease to flow conditions in most years (an average of 45 and 33 zero flow days per year, respectively).
With mostly unmodified flow, the Mitchell River catchment has a high diversity of fish, with 46 species recorded (Pusey et al. 2004). Indigenous Australians from several communities make up ̴ 23% of the catchment population and harvest fish, turtles and other resources from wetlands, creeks and river channels as an important part of their diet (Jackson et al. 2014). While cattle grazing is currently the most extensive land-use throughout the Mitchell River catchment, there is also a small area of irrigated agriculture in the upper Walsh catchment, supplemented with water from an inter-basin transfer from the eastern-flowing Barron River (Webster et al. 2009). Few barriers to fish movement currently exist within the catchment, including a natural rock cascade on the Mitchell River (Fig. 1), and one dam in the headwaters of the Mitchell River which is considered to have a hydrological impact (Southedge Dam at Lake Mitchell; Marshall 2016). However, the Mitchell River catchment has received considerable attention for potential expansion of irrigated agriculture that would require new water infrastructure development (Commonwealth of Australia 2015).
2.1 Data collection
To explore fish species composition throughout the Mitchell River catchment, we quantitatively sampled fish communities from 22 sites using electrofishing methods in June/July 2017 and June 2018 (Fig. 1). In the wet seasons immediately prior to both sampling years, flow was average in the Mitchell, low in the Palmer, and low-average, followed by average-high in the Walsh sub-catchment (compared to the past 10 years of wet season flows in each sub-catchment). Wet season flows subsided in late March in the Mitchell and Walsh sub-catchments in both sampling years, while wet season flows in the Palmer sub-catchment subsided in early April in both sampling years. We sampled fish using either a 2.5 GPP Smith-Root boat-mounted electrofishing unit or a Smith-Root LR24 back-pack. We conducted between three to five electrofishing passes at each site, ensuring that all habitat-types were sampled (i.e., overhanging vegetation, submerged wood, macrophytes, undercut banks). Each electrofishing shot was between 5 and 10 minutes. We counted all fish species and recorded standard length (mm) measurements for all collected individuals. For each shot, we also collected several habitat metrics, including water depth, percent substrate type (mud, sand, gravel, and bedrock), and percent macro-habitat type (root mass, overhanging vegetation, undercut bank, and large [> 10 cm dia.] and small [< 10 cm dia.] wood). Depending on the size of the site sampled, which varied with geomorphology and macrohabitat type (reach lengths of sites ranged between 190 and 1023 m), we measured water depth and the following physiochemical variables at between one and four locations within the site: temperature, conductivity, turbidity, and dissolved oxygen. Water column profile physicochemical measurements were taken at sites with stratification, though most sites had well mixed physicochemical profiles. We measured the physicochemical variables using a YSI handheld sonde.
We used boat electrofishing where possible due to accessibility in remote areas and dangers of estuarine crocodiles. However, we used backpack electrofishing at sites where low water levels prevented boat use. Four out of 22 sites were backpack electrofished, three of which are located on the Walsh River in the upper catchment. We included data from the backpack sites in the analyses to ensure we had sufficient data from the Walsh River, and because the species composition (abundance CPUE) was not significantly different between boat and backpack sites for all species combined (ANOSIM) and for individual species (aside from Glossamia aprion and Glossogobius sp. which were more abundant in backpack catches (ANOVA)).
2.2 Datasets and variables
2.2.1 Hydrological connectivity
To explore the relationships between hydrological connectivity (spatially weighted) and environmental and biological datasets we used two different forms of a connectivity index (O’Mara et al. 2021b): a continuous variable and a categorical one.
We used modelled discharge data (2000–2015) from stream flow gauges throughout the Mitchell catchment to create the hydrological connectivity index. We assigned in-channel sites to the nearest stream flow gauge and used the pairwise flow between sites as the basis for the connectivity metric. We first calculated the proportion of days two sites were connected by flow over a 15-year period, assuming a water depth of 10 cm over a gauge was needed for fish to pass, and then spatially weighted this proportion by the distance between each pair of sites and the slope between sites. To convert this triangular matrix of pairwise connectivity into a hydrological connectivity variable with a single values for each site, we used the mean of the pairwise connectivity value for each site, weighted by distance to the river mouth (to establish relative positions of sites within the river network). We treated off-channel sites differently because we did not quantify flow-connectivity at these locations using in-channel flows due to their large and variable distance from the main channel. As such, we assigned these sites a value close to zero, 0.1 for wetlands, and 0.15 for ephemeral floodplain creeks, which are known to have higher connectivity than floodplain wetlands in this system (Karim et al. 2018). The value for the ephemeral floodplain creeks was approximately half of the value of the least connected in-channel site (0.29). A value of zero would not be an accurate representation of the connectivity of the off-channel sites because they are periodically connected to the river channel in large floods (Karim et al. 2018).
There was a clear separation of two groups of sites evident in the histogram of the connectivity index, and we used this separation to classify in-channel sites into high or low connectivity levels. We grouped off-channel sites into their own category because of their obvious hydrological differences to in-channel sites.
2.2.2 Space
To understand effects of spatial location independent of hydrological connectivity effects, we also included spatial variables and a categorical variable describing catchment position in the analyses. The spatial variables were ‘longitude’ and ‘latitude’, and the categories of ‘catchment position’ were upper, mid, and lower catchment.
2.2.3 Environmental variables
We used minimally correlated environmental variables in the analysis (excluding one variable from each pair of variables with Pearson’s r > 0.75) that described habitat characteristics and water physical and chemical measurements (depth, water temperature, conductivity, turbidity, dissolved oxygen, mud substrate (%), sand substrate (%), bedrock (%), small wood (%), large wood (%), macrophyte cover (%), overhanging vegetation (%), steep banks (%), undercut banks (%)). Physical and chemical variables were averaged throughout the water column for stratified sites and all environmental variables (except depth) were averaged across replicates within each site. The value used for site depth was the average of the shallowest and deepest recorded site depth. Water physical and chemical variables and depth were log(x + 1) transformed while habitat characteristics (recorded as percentages) were arcsine transformed using 2/PI*arcsin(sqrt(x)). All variables were then normalized. The normalized data was transformed into a Euclidean distance matrix.
Next, we refined the environmental variables so that there were substantially more samples than variables (a requirement of the multivariate DistLM analysis used to identify which environmental variables best predicted fish abundance, using a distance-based regression approach). We kept variables with r > 0.6 in any of the five principal components of a PCA, and the reduced dataset Euclidean distance dissimilarity matrix is referred to as ‘Env’. The variables in the Env dataset were average depth, conductivity, dissolved oxygen, mud substrate (%), sand substrate (%), bedrock (%), small wood (%), large wood (%), macrophyte cover (%), overhanging vegetation (%), steep banks (%), undercut banks (%).
2.2.4 Fish species composition
We used fish abundance CPUE data to represent fish species composition. We calculated abundance CPUE by summing the number of each species per site and dividing by the total electrofishing shot time for that site. We then square root transformed the abundance data to improve normality (assessed using draftsman plots) and created a Bray-Curtis dissimilarity matrix of this data.
2.2.5 Fish functional diversity
We created a fish metrics dataset to study how the composition of fish communities across the Mitchell catchment was related to the functional diversity of the species present. We chose metrics that were most likely to influence where a species was found (trophic guild, reproductive movement classification, parental care classification, and spawning substrate and frequency). The classifications for these traits were taken from Pusey et al., (2004). The dataset was in the form of sites × traits and the number for each trait for each site was a proportion which we determined by dividing the number of species fitting the trait classification by the total number of species caught at that site. We then created a Bray-Curtis dissimilarity matrix of the functional diversity dataset.
2.2.6 Fish species turnover
We calculated fish species turnover (Simpson’s pairwise dissimilarity matrix using the ‘betapart’ package in R (Baselga and Orme 2012; R Core Team 2021)) from the presence and absence of species across sites. Species turnover refers to the difference in species between sites, where a value of 1 would indicate complete turnover (a unique fish assemblage compared to other sites). To study the effect of hydrological connectivity on turnover between adjacent sites, we used another dataset which contained only the pairwise turnover values between adjacent sites.
2.3 Statistical analysis
We used the software Microsoft Excel, Primer 6 (Clarke and Gorley 2006) and R version 4.0.2 (R Core Team 2021) in the RStudio IDE (RStudio Team 2020) for all data transformations and analyses.
Are characteristics of the environment related to hydrological connectivity?
To address the first research question, we used an ANOSIM of all the measured environmental variables to test whether environmental characteristics varied over different connectivity levels using the three-level factor ‘hydrological connectivity’. Following this, we used a SIMPER analysis to identify which environmental variables were most different between the three levels. Only variables with individual contributions > 10% were listed in the results. We then performed a second ANOSIM with the three-level factor ‘catchment position’ to test whether the separation of sites by hydrological connectivity could be attributed purely to the spatial location of sites.
Is fish community composition related to hydrological connectivity and characteristics of the environment?
To address the second research question, we used a combination of ANOSIM, SIMPER, Mantel tests, regression slope tests, and distLM analyses on the species composition dataset. Firstly, to examine the influence of hydrological connectivity on fish species composition we used ANOSIM with the three levels of connectivity or catchment position, followed by a SIMPER analysis to identify which species were most different between the levels of hydrological connectivity or catchment position. Only species with individual contributions > 10% were listed in the results. We also examined whether the abundance of individual species was directly related to connectivity using linear regression (individual species CPUE ~ mean site connectivity), only including sites for which connectivity was quantified (in-channel sites).
Secondly, to determine whether environmental characteristics were correlated with species composition we performed a Mantel test of pairwise matrices (species composition ~ Env). We then tested the differences between the slopes of species composition ~ Env for each of the three connectivity levels to assess whether the relationship between species composition and the environment changed over different levels of hydrological connectivity. We also used a forward distLM (R2) analysis to identify which Env variables best predicted species composition. Because the variable ‘longitude’ was a significant predictor when included in the model, we performed a further distLM with forced inclusion of ‘longitude’ to determine which variables best predicted species composition when the effect of longitude was accounted for.
How does fish functional diversity change across different environments and levels of hydrological connectivity?
To address the third research question, we used a combination of ANOSIM, SIMPER, Mantel tests, regression slope tests, and distLM analyses on the functional diversity dataset. Firstly, we used ANOSIM to test for significant differences in functional diversity between the three levels of connectivity and catchment position. Following this, we used SIMPER analysis to identify which functional traits were most different between the significantly different levels of hydrological connectivity or catchment position. Only variables with individual contributions > 10% were listed in the results. Secondly, to determine whether environmental characteristics were correlated with species composition we performed the same Mantel and slope comparison tests used on the species composition dataset on the functional diversity dataset. We also carried out a forward distLM (R2) analysis with the Env dataset to identify which Env variables best predicted functional diversity. Because the variable ‘longitude’ was a significant predictor when included in the model, we performed a further distLM with forced inclusion of ‘longitude’ to determine which variables best predicted functional diversity when the effect of longitude was accounted for.
Does species turnover occur across the catchment and is turnover related to functional diversity?
To address the fourth research question, we used distLM, ANOVA, and Mantel tests. We used two forward distLM (R2) models to identify relationships between turnover and spatial and Env variables. The first model entered the Env variables and the variables longitude and latitude, while the second model assessed Env variables after the effect of the significant spatial variable ‘longitude’ was accounted for. We then used an ANOVA (Type II SS) to assess whether adjacent pairwise turnover values differed among the three levels of hydrological connectivity. We did not include the site ‘Twelve-mile hole’ in this analysis because the high turnover value and observed fish movement in the catchment (unpublished data) indicate that this wetland is not connected to the river channel during seasonal floods. Finally, we carried out a Mantel test of turnover ~ functional diversity to assess whether species turnover in the Mitchell catchment occurred according to the functional diversity of the species present.