A revision of the bioregionalisation of freshwater fish communities in the Australian Monsoonal Tropics

Abstract The Australian freshwater fish fauna is very unique, but poorly understood. In the Australian Monsoonal Tropics (AMT) biome of northern Australia, the number of described and candidate species has nearly doubled since the last attempt to analyse freshwater fish species composition patterns and determine a bioregionalisation scheme. Here, we utilise the most complete database of catchment‐scale freshwater fish distributions from the AMT to date to: (a) reanalyze spatial patterns of species richness, endemism and turnover of freshwater fishes; (b) propose a biogeographic regionalisation based on species turnover; (c) assess the relationship between species turnover and patterns of environmental change and historic drainage connectivity; and (d) identify sampling gaps. Biogeographic provinces were identified using an agglomerative cluster analysis of a Simpson's beta (β sim) dissimilarity matrix. A generalised dissimilarity model incorporating eighteen environmental variables was used to investigate the environmental correlates of species turnover. Observed and estimated species richness and endemism were calculated and inventory completeness was estimated based on the ratio of observed to estimated species richness. Three major freshwater fish biogeographic provinces and 14 subprovinces are proposed. These differ substantially from the current bioregionalisation scheme. Species turnover was most strongly influenced by environmental variables that are interpreted to reflect changes in terrain (catchment relief and confinement), geology and climate (runoff perenniality, stream density), and biotic responses to climate (net primary productivity). Past connectivity between rivers during low sea‐level events is also influential highlighting the importance of historical processes in explaining contemporary patterns of biodiversity in the AMT. The inclusion of 49 newly discovered species and candidate species only reinforced known focal points of species richness and endemism in the AMT. However, a number of key sampling gaps remain that need to be filled to fully characterise the proposed bioregionalisation.


| INTRODUC TI ON
In comparison to other continents, Australian freshwater fish communities are considered highly unique but species poor, largely due to Australia's arid climate and long isolation from other land masses (Unmack, 2001(Unmack, , 2013. However, the Australian Monsoonal Tropics (AMT) biome in the tropical north of the continent is an exception (see Figure 1). The AMT encompasses 33% of Australia's landmass, but contains 65% of all freshwater fishes, and the encompassed provinces contain a moderate number of species compared to others around the globe (Abell et al., 2008). Furthermore, the number of described and candidate species in the AMT has increased by 50% since the turn of the millennium, compared to a background rate of 35% across all of Australia (see Allen, Midgley, & Allen, 2002;Le Feuvre, Dempster, Shelley, & Swearer, 2016 for comparison). Despite its importance and the disproportionate increase in biodiversity estimates, the AMT has been the subject of considerably less research into biodiversity and the evolution of biotic communities than the highly populated regions to the south (Bowman et al., 2010). There is therefore a substantial need for increased biogeographic research in this region.
The most recent analysis of freshwater fish distribution patterns (excluding diadromous species; Unmack, 2001) mapped species richness, endemism, and general biogeographic provinces at the continental scale. The analysis was constrained to a comparison of relatedness between operational geographic units (OGUs) defined by expert opinion. This work was updated by Unmack (2013) accounting for newly discovered species, but the bioregionalisation was not reanalyzed. The substantial increase in biodiversity estimates and changes recorded in species distributions in the AMT is likely to challenge aspects of the original analysis.
Furthermore, accurate and comprehensive models of landscape scale environmental data are now readily available and can be used to quantitatively explore potential drivers of freshwater F I G U R E 1 Elevation map of the Australian Monsoonal Tropics and catchment boundaries. Arrows indicate catchments discussed in text. Letters denote prominent geographic features discussed in text: (a) Kimberley Plateau, (b) Arnhem Plateau, (c) Great Dividing Range, (d) Lake Bonaparte, (e) Lake Carpentaria. Beyond the coastline, the sea floor depth to 135 m (the maximum estimated fall in sea level during the last glacial cycle; Clark & Mix, 2002) is marked by a gradient of blue. The margin of the figure represents the 135 m below sea-level contour and effectively highlights the continental shelf edge. Low sea-level drainage patterns are shown as blue lines, derived from a bathymetric 30 arcsecond (c. l km) dataset from the Australian Geological Survey Organization fish species distributions in the AMT (e.g., Hermoso, Kennard, & Linke, 2012;Sternberg & Kennard, 2013;Sternberg & Kennard, 2014). Our understanding of the historical significance of sea-level changes, driven by Plio-Pleistocene glacial cycles, on species distributions has also become clearer since phylogenetic data collection and analysis have become more widely implemented. There is an increasing amount of phylogenetic evidence that the particularly wide continental shelf around the northern coastline of the AMT provided dispersal pathways for a number of freshwater species between regions that are now isolated, for example between the island of New Guinea and Australia (see Figure 1; e.g., Baker, De Bruyn, & Mather, 2008;Cook, Adams, Mather, & Hughes, 2012;De Bruyn, Wilson, & Mather, 2004;. A key remaining challenge is to untangle the influence of past and present processes on biogeographic patterns. All of these elements combined highlight the need and opportunity to produce a comprehensive and updated analysis of freshwater fish biogeography in the AMT. By grouping biological communities into clusters of meaningful geographical units, patterns in evolution, diversity, and the ecological processes that sustain that diversity can be revealed (Morrone & Crisci, 1995). As such, a biogeographic regionalisation is essential for developing ecologically representative systems of protected areas and is a requirement of national and international conservation agreements such as the Ramsar Convention on Wetlands, the National Water Initiative, and the Australian Guidelines for Establishing the National Reserve System (Kennard, 2010;Lourie & Vincent, 2004;Olden et al., 2010). It is also an important tool to assess our current state of knowledge, identify survey gaps to prioritise future research, and to refine biogeographic province boundaries (Higgins, Bryer, Khoury, & Fitzhugh, 2005).
Here, we generate a bioregionalisation of the freshwater fishes in the Australian tropics using the Simpson's beta dissimilarity metric, and then assess the relationships of biogeographic regions, referred to here as "provinces" and "subprovinces", to their current environment using generalised dissimilarity modeling (GDM). Specifically, our goals were to: (a) map spatial patterns of species richness, endemism and turnover between river catchments; (b) propose an updated biogeographic regionalisation of freshwater fishes based on the spatial turnover of species; (c) assess the relationship between species turnover and patterns of environmental change and historic drainage connectivity; and (d) identify survey gaps that require further sampling.

| ME THODS
We investigated biogeographic patterns of freshwater fishes in the AMT, revising the most recent analysis conducted by Unmack (2001).
We did not attempt to exactly replicate the analysis in Unmack (2001) as new techniques have become available and are better suited to this dataset (see Geographical structure of species turnover).
Key differences between these two studies are outlined in Table 1.

| Study area
The AMT biome is here defined as encompassing all coastal river catchments from the Fitzroy River in the Kimberley region eastwards to the O'Connell River on Australia's east coast ( Figure 1). As a biome, the southern limit is defined by areas that receive more than 85% of their annual rainfall between November and April, TA B L E 1 Principal differences between the present study and that of Unmack (2001) for the analysis of the distribution of species dissimilarity and diversity of freshwater fishes commonly referred to as the "wet season" (Bowman et al., 2010).
South of the southern border of the AMT lies the vast arid and semiarid expanses of inland Australia. The 200 mm/year rainfall contour marks a hard faunal break for aquatic communities. The break is less distinct across catchments that drain to the east coast where rainfall is significantly higher. A strong faunal connection exists between the AMT and southern New Guinea with the regions sharing 66 species (37% of the total). The AMT landscape is largely flat, and geologically stable, with the most recent major tectonic activity (e.g., mountain building) occurring >80 Ma (Johnson, 2004). Concentrated human development is limited across the region and the river catchments represent some of the most intact riverine ecosystems in the world  (Stein, Hutchinson, & Stein, 2012). Calculations for the "Paleo-geographical distance" were based on paleo-drainage models derived from a bathymetric 30 arc-second (c. l km) dataset produced by the Australian Geological Survey Organization (see Figure 1). Modelled paleo-drainage connections at the sea floor depth of −135 m (the maximum estimated fall in sea level during the last glacial cycle; Clark & Mix, 2002) were used.
TA B L E 2 Environmental and distance variables used in our analyses . However, Indigenous Australians have occupied Australia for approximately 50,000 years (Bowler et al., 2003;Flood, 2004) and their impact on the distribution of freshwater fishes is poorly understood (Humphries, 2007). All things considered, the AMT presents an ideal region in which to study natural freshwater biogeographic patterns.

| Published species distributions
Our definition of freshwater fish includes species that reproduce in freshwater and diadromous species that spend part of their life history in freshwater. Species distributions for each catchment were primarily derived from three comprehensive and relatively recent reviews of fish distributions across the study region. These include reviews of the Kimberley region, northwestern Australia, from the Fitzroy River to the Ord River (Morgan, Allen, Pusey, & Burrows, 2011) and from the Fitzroy River to the Fitzmaurice River (Shelley, Morgan et al., 2018), northern Australia from the Keep River (immediately east of the Ord River) to the Annan River in Queensland (Pusey et al., 2017), and between the Annan River and the O'Connell River at the southern extent of our study area (Pusey, Kennard, & Arthington, 2004). Readers can refer to these papers for complete reference lists of the surveys that contributed to the underlying data.
Additional data was added to these lists from more recent surveys, which are listed in Supporting Information Appendix S1. Anomalous records were crosschecked with authorities on freshwater fish distributions in northern Australia (see Acknowledgments).

| Inclusion of genetic candidate species
Taxonomy is fluid in that names of species, genera and families change and new "cryptic species" are discovered within previously described groups as new data become available. There are complexes of species awaiting formal description in Australia (and elsewhere) that show genetic structure that likely warrants taxonomic recognition (Unmack, 2013

| Final database
River catchments were used as the base spatial unit for our analyses and included any sized creek or river catchment defined in the river catchment framework from the Australian Hydrologic Geospatial Fabric (Stein, Hutchinson, & Stein, 2014), which was accessed using

| Paleo-drainages
Drainage patterns during lowered sea levels (paleo-drainages) were modelled using Spatial Analyst 1.1 and ArcView 3.1, based upon a bathymetric 30 arc-second grid produced by the Australian Geological Survey Organization. Paleo-drainages are displayed from the current coastline to the sea floor depth of −135 m (the maximum estimated fall in sea level during the last glacial maximum; Clark & Mix, 2002).

| Current and paleo-geographic distance between drainage basins
To account for a decrease in the species compositional similarity between two localities with increasing geographical distance (i.e., distance-decay; Nekola & White, 1999), we calculated measures of geographical distance between (a) current catchments as defined in the river catchment framework from the Australian Hydrologic Geospatial Fabric (Stein et al., 2014); and (b) paleo-catchments defined in the modeled low sea-level drainage patterns (see Figure 1). Similar to the methodology of Dias et al. (2014), we defined the current geographic distance between river catchments as the number of catchment divides that would have to be crossed along the coastline when travelling from one basin to another. While some catchments may border one another at their headwaters, freshwater fish dispersal via headwater capture events are considered to be extremely rare in the geologically stable AMT (Unmack, 2001), hence we focused on dispersal across catchment divides at the coastline. In this way, we accounted for dispersal constraints on species movement, and thus species turnover, in addition to ecological niche-based processes that are accounted for by the environmental variables in the model, described above. This approach was also preferable as it overcomes the difficulty of defining the geographical Euclidean distances between river catchments.
To calculate current geographical distance, catchments (including those without fish species composition data) between the western and eastern-most rivers used in our analysis were numbered in increasing order (i.e., from the Fitzroy River [1] to the O'Connell River [301]). In this way, a distance of 1 corresponds to catchments that are adjoining at the coast, while that number increases the further apart they are. The corresponding catchment number was ascribed to each catchment for which we had fish community data.
To investigate the effect of low sea-level catchment connectivity on species turnover, we calculated paleo-geographic distances using the same approach as described above to the paleo-catchments at the −135 m contour. For example, all catchments that drained into the paleo-lake, Lake Carpentaria, during the lowered sea levels are labeled a single number. We exclusively considered the confluence of drainages at the lowest sea level reached during the last glacial maximum (−135 m) as it presents the greatest possible opportunity for connectivity between catchments.
Spearman's correlation analysis was used to test how highly correlated current and paleo-geographical distances were. The resulting Spearman's correlation coefficient (0.81) indicated that the two variables are quite highly correlated, highlighting the difficulty in untangling the influences of past and present connectivity on species turnover. In response to this, we ran the GDMs with and without paleo-geographical distance and found that its inclusion improved the amount of variance accounted for in the whole AMT model by 8.5%, and the province scale models by as much as 11.8% (results not presented here). Given the significant improvement in explanatory power of the models, we included paleo-geographical distance as a variable in our analysis.

| Species richness and endemism
We calculated species richness and endemism to identify spatial patterns of ecological and conservation significance. Many diversity patterns are scale dependent, in that some occur over geographically local extents, while others become apparent only when much larger spatial extents are considered (Laffan & Crisp, 2003). Here, we analyze endemism at three biologically relevant spatial scales defined by our analysis of species turnover (see section below): biogeographic province, subprovince, and catchment.
Species richness (SR) and corrected weighted endemism (CWE; Crisp, Laffan, Linder, & Monro, 2001) were calculated in Biodiverse 1.99 (Laffan, Lubarsky, & Rosauer, 2010) using catchments as the finest spatial unit. We considered endemism in relation to the AMT; that is, species distributed outside the study region may still be considered endemic to a given catchment, subprovince, or province within the AMT. CWE is a relative measure of endemism that indicates the degree of range restriction of a sample to a location or set of locations. In the case of this study, it can be interpreted as the degree to which species found in a spatial unit (e.g., catchment or subprovince), on average, are restricted to that unit (Laffan, Ramp, & Roger, 2013). Once the species richness and endemism scores were calculated for all catchments, we defined focal points of species richness and endemism by selecting those catchments with the highest 10% of scores.

| Species turnover
A matrix of Simpson's beta (β sim ) species turnover was generated for all pairwise catchment combinations for use in our cluster analysis. Simpson's beta was used because it reduces the effect of any species richness imbalance between locations and thus reduces the effect of unequal sampling effort (Tuomisto, 2010), which is characteristic of this poorly studied region. Simpson's beta can be calculated as where a refers to the number of species common to spatial units i and j, b is the number found in spatial unit i but not spatial unit j, and c is the number found in spatial unit j but not spatial unit i. A low β sim value indicates that many taxa are shared between two spatial units (low dissimilarity) and a high β sim means a small number of shared taxa (high dissimilarity).

| Geographical structure of species turnover
The β sim pairwise distance matrix was used in an agglomerative cluster analysis to generate a WPGMA (weighted pair-group method using arithmetic averages) hierarchical dendrogram in Biodiverse. This allowed us to visually assess the relationships between the catchments (i.e., the similarity of their species composition) and ultimately determine a bioregionalisation scheme. WPGMA weights the contributions of clusters by the number of terminal nodes (spatial units) they contain, ensuring each unit contributes equally to each merger of which it is a part. Kreft and Jetz (2010) found WPGMA was consistently among the best performing hierarchical clustering methods.
We used a tiebreaker approach such that, when multiple pairs of clusters had the minimum turnover score and thus could be merged, the algorithm selected the pair that maximised the corrected weighted endemism score in the cluster (González-Orozco et al., 2014;Laffan & Crisp, 2003). Using this approach provides a more stable and replicable result while also optimising for the degree of endemism and thus spatial compactness of the resultant biogeographic regions. Our criteria for defining biogeographic regions (provinces and subprovinces) from the clusters were: (a) respective regions are represented by a contiguous group of catchments; and (b) each cluster that represents a region is clearly separated from its children or parent in the dendrogram. Furthermore, we set a limit where only the first 10% of nodes (from the root of the dendrogram) represented divergences that were deep enough to be classed as a subprovince.

| Generalised dissimilarity modeling
We used GDM to examine how freshwater fish species composition varies with climate, geography and habitat, using the Generalised Dissimilarity Modeler 1.2.3. GDM models dissimilarity in species composition between pairs of locations as a nonlinear function of geographical and environmental distances between these locations.
Unlike other linear approaches, GDM allows for curvilinear relationships between observed compositional dissimilarity and increasing ecological and geographical separation between sites, and variation in the rate of compositional turnover at different positions along environmental gradients (for a full description of the method see Ferrier, Manion, Elith, & Richardson, 2007). The analysis was run using a Simpson's beta dissimilarity matrix, the subset of uncorrelated environmental predictors, and geographical distance between the centroids of sampled catchments. We ran the analysis for the biome (AMT) and individual provinces, to help determine the most significant environmental drivers at each level. A Wald test was applied to each parameter to determine if it significantly impacted model fit (p-value ≤0.05).

| Inventory completeness index
To identify poorly sampled regions that are most in need of additional sampling effort, we calculated an inventory completeness index (C-index). The C-index can be defined as the ratio of observed species richness to estimated species richness in a given spatial unit (Soberón, Jiménez, Golubov, & Koleff, 2007).
We calculated estimates of potential species richness using the nonparametric Chao 2 estimator (Colwell & Coddington, 1994) implemented in Biodiverse. The Chao 2 estimator calculates the total number of species present, including those species that were not sampled, by extrapolating the asymptote of a rarefaction curve. For a given spatial unit i (e.g., province) this statistic (S chao(i) ) can be calculated as where S obs(i) is the observed species richness in spatial unit i, and f 1 and f 2 are the number of singletons (species represented in a single sampling unit) and doubletons (species represented in two sampling units), respectively, found in i. The completeness index (C index ) was then calculated as We analyzed inventory completeness at the three conservation management spatial levels (biome, province, and subprovince). Due to its formulation, the Chao 2 index cannot be accurately estimated for a single spatial unit such as a catchment, so catchment level estimates were not included.

| Biogeographic provinces
Three major biogeographic provinces and 12 subprovinces are proposed within the AMT biome ( Figure 2). The dendrogram branch lengths ( Table 3) for each of the geographical clusters ( Figure 2b) show 8%-22% divergence (species-composition dissimilarity) be- F I G U R E 2 Proposed freshwater fish biogeographic provinces (in capitals) and subprovinces (in lower case) in the Australian Monsoonal Tropics (a) and the WPGMA tree of 178 species based on Simpson's beta dissimilarity matrix upon which the provinces were determined (b). Numbered boxes in (a) refer to anomalous catchments highlighted in (b) and are discussed in the Results section. Asterisks at the end of branches in (b) denote anomalous catchments. Gray catchments in (a) are those for which no or inadequate fish community composition data was available adjacent to the rivers in that subprovince. The anomalous positioning of the King George River is likely due to its depauperate fish com-

| Species richness and endemism
The most species rich and endemic regions of freshwater fishes showed little overlap (Figure 3,  (Figure 3a). None were identified in the Kimberley. In total, Wet Tropics was the most species rich subprovince (66 species) and Eastern was the most species rich province (97 species

| Inventory completeness index
F I G U R E 4 Predicted ecological distance of the generalised dissimilarity model of freshwater fish species composition in the Australian Monsoonal Tropics and each major province plotted against observed compositional dissimilarity (top row), and functions fitted to each of the most significant environmental and distance predictors (p-value ≤0.05) in the generalised dissimilarity model of freshwater fish species composition. The figure is arranged in columns by biome/province. Abbreviations for significant predictors relate to Table 2

| Kimberley Province
The Kimberley Province has experienced, by far, the largest increase in knowledge of freshwater fish biodiversity (29 newly recognised species) and distributions since the last bioregionalisation (Unmack, 2001). As a result, the most significant changes in bioregionalisation were observed here. Most importantly, we extend the Kimberley F I G U R E 5 Visual representation of the environmental predictors selected by the generalised dissimilarity models (GDMs) for the Australian Monsoonal Tropics biome, and the Kimberley, Northern and Eastern provinces: (a) Net primary productivity (b) Catchment average slope (c) Stream density (d) Stream confinement (e) Perenniality of runoff. Environmental data is only shown for catchments used in GDMs (i.e., catchments with fish community composition data). Stream confinement and Perenniality of runoff were only significant predictors at the level of biogeographic province and as such, data is presented for the relevant province only Province from the Durack River east to the Fitzmaurice River, thus including an additional eight major river catchments and numerous coastal creeks. This expansion is further supported by phylogenetic analyzes of a number of widespread species that indicate that the lowland catchments between the Kimberley Plateau and the Fitzmaurice River are more strongly related to regions to the west than those to the east (Shelley, Swearer et al., 2018;Unmack, 2013;Unmack & Dowling, 2010).
The presence of such a strong biogeographic barrier for freshwater fish around the Joseph Bonaparte Gulf, seems counterintuitive as low sea-level drainage patterns suggest that catchments between the King Edward River (Kimberley) and Howard River (Northern) were connected by a brackish lake during lowered sea levels (Yokoyama, Purcell, Lambeck, & Johnston, 2001 We also recognise six distinct subprovinces rather than the two proposed in Unmack (2001Unmack ( , 2013. These highly distinct and spatially 27 (66%) of the endemic species are restricted to one or two river systems (see Morgan et al., 2011 andShelley, Morgan et al., 2018 for distributions). Of note, seven of the nine AMT catchments that were identified as focal points of endemism occurred around the Kimberley Plateau, with NW Plateau subprovince exhibiting the highest endemism in the AMT. It has been hypothesised that the topographical complexity of the Kimberley Plateau (see Figure 1) has acted to isolate freshwater fish populations across the landscape, contributing strongly to this phenomenon (Unmack, 2013).
On the other hand, Kimberley Province and its subprovinces are the least species rich in the AMT (78 species). This is largely due to a low proportion of widespread species (16 species, 21%) that have migrated from other regions. These findings are in line with more broad-scale analyses of diversity in the AMT, which indicates the biogeographic boundaries that define Kimberley Province are particularly strong (Kennard, 2010;Unmack, 2001Unmack, , 2013.

| Northern Province
Four broad subprovinces were identified across Northern Province, as compared to the nine identified by Unmack (2001Unmack ( , 2013. The This relationship is most easily explained by drainage connections during low sea levels (Figure 1; Unmack, 2013). In the present Gulf of Carpentaria, there is a large depression that would have formed a lake containing fresh to brackish water during lowered sea levels.
This feature, known as Lake Carpentaria, provided the potential for widespread connectivity across catchments draining into the Gulf, and those connections may have been present as recently as ~10,000 years ago (Reeves et al., 2008). However, this appears to contradict the division between the Northern and Southern Gulf subprovinces. A likely explanation for this is that the free dispersal of freshwater species via Lake Carpentaria was counteracted by increased aridity during glacial phases, particularly in the south of the AMT, which would have greatly reduced river perenniality and discharge into Lake Carpentaria from the Southern Gulf catchments (Playà, Cendón, Travé, Chivas, & García, 2007;Unmack, 2001). Geological evidence suggests that during the last glacial maximum, only the northernmost rivers of the Gulf (e.g., north of the Archer-Coen river on the eastern side) contributed water to Lake Carpentaria. Rivers in the south likely received less rainfall and would have had to flow further before entering the lake, and thus may have largely evaporated before reaching the lake (Playà et al., 2007), except perhaps during exceptional flood events.
Northern Province is the second most species rich (90 spe- across Northern Province is most readily explained by low sea-level drainage patterns, which indicate that there has historically been broad hydrological connectivity between catchments draining into Lake Bonaparte and Lake Carpentaria, respectively. In Top End subprovince, low sea-level drainage patterns indicate that catchments between the Adelaide and East Alligator rivers and between the Goomadeer and Goyder rivers were also hydrologically connected when the sea level was lowest (see Figure 1). Furthermore, drainage divides across the region are typically at low elevation near the current coastline (see Figure 1) and the catchments exhibit expansive lowland floodplains that may have provided opportunity for fish to move between basins via floodplain inundation during even slightly lowered sea levels.
Furthermore, Northern Province contains seven of the 11 most species rich catchments. While these catchments occur across each of the subprovinces, they form two broad clusters around the western Top End and Daly subprovinces and the tip of Cape York Peninsula. These findings reinforce general patterns highlighted in Kennard (2010) and Unmack (2001Unmack ( , 2013. Over half (48) of Northern Provinces' species are shared with the island of New Guinea, which helps to explain the high richness and endemism values.

| Eastern Province
Eastern Province covers the tip of Cape York Peninsula and a long, narrow area that is wedged between the Great Dividing Range and the east coast of Australia. The four Eastern subprovinces were unchanged from those determined by Unmack (2001Unmack ( , 2013. The provincial divide (between Eastern and Northern) that our analysis highlights between the eastern and western sides of Cape York Peninsula, follows the Great Dividing Range to near its northern extent on the Australian mainland. At this point, the elevational divide is particularly low (see Figure 1), the opportunity for dispersal between the eastern and western sides would be much greater, and unsurprisingly the species composition is similar.
Eastern Province (within the AMT) has the most species (97 species), while endemism is equal with Northern Province (CWE = 0.68).
The species richness in the province also benefits from the presence of 50 (52%) species that are shared with New Guinea and elsewhere in the Indo-Pacific region. Wet Tropics is the most species rich (66 species) and endemic subprovince (CWE = 0.39) within Eastern province and across the AMT, with the Daintree and Mulgrave rivers being focal points of both richness and endemism. The Olive River in Cape York Peninsula is also a focal point of richness. Wet Tropics subprovince contains 13 Australian endemic species and a further 10 recently discovered amphidromous gobioid species that are locally endemic (i.e., within Australia) but also occur across the Indo-Pacific (Thuesen et al., 2011). The defining features of Wet Tropics are the exceptionally high rainfall that leads to high reliability of flow, and steep catchment gradients that, in combination with the high and consistent flow, produce an array of perennial lotic habitats that are unique within the ATM and Australia in general. Some rivers draining Cape York Peninsula, such as the Olive River, provide similar conditions. These factors have been hypothesised to be reasons for the high species richness in the Wet Tropics (Pusey, Arthington, & Read, 1995;Pusey and Kennard, 1996;Unmack, 2001), and the presence of many amphidromous gobioids that have strong habitat associations with fast flowing, perennial habitat (Keith, 2003) provides strong support. We believe that the same reasoning can be applied to the Olive River.

| Linking freshwater fish species turnover to the environment
Here, we provide the first attempt to link turnover in catchmentscale freshwater fish species composition with measures of current and paleo-geographical distance, and broad-scale environmental variables. Our GDM analysis determined that variation in freshwater fish species composition across the AMT was best explained by increases in current and paleo-geographical distance between catchments, catchment average slope, terrestrial NPP, and stream density.
The identification of paleo-geographical distance as a significant factor influencing species turnover supports strong genetic evidence for the influence of low sea-level drainage patterns on species and population connectivity in the region (Baker et al., 2008;Cook et al., 2012;De Bruyn et al., 2004;Huey et al., 2014). The widespread and/ or disjunct species distributions of species around the paleo-lakes discussed above are the clearest examples of this, although smallerscale cases of increased connectivity between catchments during lowered sea levels are also present across the entire region.
Given the AMT is a largely flat and geologically stable landscape (Johnson, 2004) it is unsurprising that elevation changes, which influence catchment slope, would also have a strong influence on The measures of terrestrial NPP reflect biological (i.e., plant) responses to the light, thermal and moisture regimes (Kennard, 2010).
Various factors influence such productivity although rainfall and temperature are the most important (Eamus, 2003). A simple interpretation is that hot and wet areas (e.g., tropical rain forests and coastal wetlands) tend to have high NPP. Stream density indicates the amount and nature of aquatic habitat available and is also heavily influenced by rainfall in addition to the physical characteristics of the catchment such as catchment slope, soil permeability and the underlying rock type (Pidwirny, 2006). Regions with high rainfall, high relief catchments and impermeable ground or exposed bedrock tend to have higher runoff and therefore higher stream density. Stream density of a catchment influences its hydraulic response to rainfall events, with catchments with high stream density exhibiting rapid rises and falls in flows (Carlston, 1963).
As NPP and stream density reflect multiple, intertwining biological and physical processes, they may present more ecologically gov.au/climate/change/acorn-sat/). As proposed by Unmack (2001Unmack ( , 2013, these gradients present likely biological filters confining mesic adapted species to the north and east coast, while only arid adapted species may occur in the south. This was exemplified in a study of biogeographic determinants of life-history indices in Australian freshwater fishes (Sternberg & Kennard, 2014), which found that NPP and associated variables such as temperature and runoff were important for explaining variation in the frequency and distribution of freshwater fish life-history strategies in Australian river basins. In a similar study investigating environmental drivers of specific life-history traits, Sternberg and Kennard (2013) identified strong associations across AMT river basins between fish exhibiting traits associated with the "periodic" endpoint strategy (Winemiller & Rose, 1992;e.g., large bodied, late maturing, broadcast spawners that produce large numbers of small eggs) with river basins dominated by high mean annual temperature and low perenniality (e.g., the most southern subprovinces). On the other hand, fish exhibiting traits associated with the "equilibrium" strategy (i.e., larger eggs, lower fecundity and intermediate age and length at maturity) were more frequent in environments typified by a high NPP, high gradient, high stream density, and low variation in mean annual temperature (e.g., the most northern subprovinces and much of the Eastern Province). This may also help explain why no subprovince encompasses both northern and southern catchments. While the physiological tolerances of AMT fishes are poorly known, this is a likely explanation for the restriction of many Eleotridae, Gobiidae and Melanotaeniidae species to Top End, Northern Gulf and north of Burdekin along the east coast.
Within Kimberley Province, current and paleo-geographical distance, and catchment confinement (Figure 5e) best explained species turnover. This aligns with our expectations as the catchments draining the Kimberley Plateau are highly constrained by the region's rugged topography limiting fish migration between catchments under current sea-level heights and driving high species turnover over short geographic distances Phillips, Storey, & Johnson, 2009;Shelley, Swearer et al., 2018). However, lowered sea levels exposed a wide continental shelf and some of the regions currently isolated rivers coalesced before reaching the ocean providing freshwater fish species an opportunity to expand their distributions to varying degrees. Therefore, it is also expected that the closer proximity or connectivity of these highly isolated catchments under low sea levels would have a particularly strong influence of community composition similarity between impacted catchments.
The clearest example of broadened connectivity during low sea levels is observed in the catchments that would have drained into Lake Bonaparte (Yokoyama et al., 2001). However, smaller-scale low sea-level connections, such as between the Prince Regent and Roe rivers (NW Plateau), also provide an explanation for the similarity of species composition between some plateau catchments.
Across Northern Province, current geographical distance and runoff perenniality best explained species turnover, based on our model.
Runoff perenniality is largely driven by the prevailing rainfall, temperature and geology (Seaman et al., 2016). As such, the declining north to south gradient in perenniality broadly reflects the patterns observed in NPP and natural forest cover. This gradient provides a reasonable explanation for the divide between Northern Gulf and Southern Gulf and Top End and Daly as any exacerbation of the length and frequency of stream drying during the dry-season would act to filter out species adapted to the more benign conditions experienced in streams with greater perenniality, as discussed above. Another explanation for the divides between Daly, Top End, and Northern Gulf subprovinces are differences in the orientation of the drainages that flow west, north, and east, respectively, off the Arnhem Plateau ( Figure 1). Low sealevel drainage patterns suggest that these subprovinces would have historically flowed into Lake Bonaparte (Daly), off the north coast (Top End), and into Lake Carpentaria (Northern Gulf) and as such the degree of isolation between their respective catchments (i.e., between river mouths) appears to have remained high throughout the glacial cycles and may have reinforced faunal differences.
Turnover in species composition across Eastern Province was most heavily influenced by current geographical distance and catchment average slope (Figure 5b). This is unsurprising given the short catchments drain directly from the Great Dividing Range into the ocean.
Even small variations in the height and direction of the mountain range strongly influence catchment slope, causing corresponding changes in habitat and catchment connectivity as discussed above. Furthermore, the continental shelf off the east coast is narrow and straight, so it was expected that that lowered sea levels would have less of an influence on species distributions in Eastern (Unmack, 2013).

| Identification of sampling gaps
Having assembled a comprehensive checklist of freshwater fishes across the AMT, we investigated the completeness of the inventory to identify under-sampled regions. We also mapped which rivers lacked published species-composition data. As a whole, the AMT and its major provinces were broadly considered to be well sampled, There is a broad consensus that these upland regions provided mesic refuge during periods of extreme aridity, driven by late Pliocene and Pleistocene glacial cycles (Byrne et al., 2008;Fujioka & Chappell, 2010 where all focal points of endemism were identified. These patterns of narrow-range endemism have been highlighted by recent surveys, and phylogenetic and taxonomic studies Shelley, Morgan et al., 2018;Shelley, Swearer et al., 2018;Thuesen et al., 2011).
An excellent example of the influence these rugged upland regions may have on patterns on patterns of diversification can be found within Australia's third most widespread freshwater fish, Amniataba percoides.
Recently, A. percoides has been proposed to comprise at least four candidate cryptic species within the Kimberley, but only one across the rest of its range that spans approximately two-thirds of Australia (Shelley, Swearer et al., 2018). In contrast, endemism in freshwater fish is relatively low around Cape York Peninsula. Broad connectivity around Lake Carpentaria during lowered sea levels (including with New Guinea) and greater opportunity for connectivity with the east coast drainages likely provided greater opportunity for wider dispersal and, as a result, there is less geographically constrained endemism. Gulf of Carpentaria (Catullo et al., 2014). These barriers present significant population-and/or species-level divides for many taxonomic groups (e.g., plants: Crisp et al., 1995;González-Orozco et al., 2014; terrestrial vertebrates: Catullo et al., 2014;Cracraft, 1991;Potter et al., 2012;invertebrates: Matthews & Bouchard, 2008). Cladistics analyses of terrestrial vertebrate (Cracraft, 1991) and plant groups (Crisp et al., 1995) indicate that these barriers divide the AMT into three broad phytogeographical and zoogeographical subregions (analogous to our Provinces) that have been described around the Kimberley Plateau, Arnhem Land and Cape York Peninsula areas of endemism. The Atherton Tablelands (similar to our Wet Tropics subprovince) is also classified as a subregion on the east coast, with the boundary being defined by lower elevation landscapes to the north and by drier, open woodland to the south (known as the Burdekin Gap), which also form the southern boundary of the AMT.
These analyses (Cracraft, 1991;Crisp et al., 1995) have been adopted into the provisional Australian Bioregionalisation Atlas presented in Ebach et al. (2013) that presents a basis for comparison with the broad biogeographic provinces determined by our analysis.
In the case of freshwater fish, the Kimberley Province overlaps exactly with the phytogeographical and zoogeographical subregions (i.e., Kimberley Plateau) defined in Ebach et al. (2013), with the Wingate Ranges forming the provinces eastern boundary. However, the Northern Province is unique in that it encompasses the Arnhem Land subregion as well as part of the Cape York Peninsula subregion; the western side of the Great Dividing Range (as far north as the Wenlock River). The simplest explanation for this is that freshwater fishes could circumvent the arid Carpentarian Gap by crossing Lake Carpentaria (elaborated on in the Northern Province section of the Discussion), while terrestrial taxonomic groups would still have to disperse around the lake, across the arid region.
Conversely, the Great Dividing Range could be more easily crossed by terrestrial taxa than freshwater fish that are largely bound to river catchments. Therefore, the Northern and Eastern freshwater fish biogeographic provinces are divided on either side of the Great Dividing Range, while the phytogeographical and zoogeographical subregions span both sides. Finally, the Atherton Tablelands subregion overlaps closely with the Wet Tropics subprovince highlighted in our analysis spanning the region from the Bloomfield River to just north of the Burdekin River catchment. This is unsurprising given the exceptionally high rainfall and strong elevation gradient that shapes the distinct freshwater habitat in the region and in turn supports its unique fish fauna, also supports the region's unique enclosed rainforests and associated endemic terrestrial fauna (Williams et al., 2010).
Overall, freshwater fish biogeographic patterns support many of the major patterns present in the AMT, while exhibiting some that are likely unique to obligate freshwater organisms. The impact of geological barriers (e.g., the Great Dividing Range) for instance, appears to have a far greater impact of freshwater fish distribution than on terrestrial plant and animal groups. Conversely, the large paleo-lakes likely facilitated dispersal of freshwater fishes, but not terrestrial taxa. Freshwater fishes thus provide important, alternative insight into understanding the complex biogeographic history of the region.

| CON CLUS IONS
Our provinces present a number of significant changes to previous broad-scale studies of freshwater fish species composition in the AMT that were based on a much smaller dataset (Unmack, 2001(Unmack, , 2013. Key differences in our study include changes to the Kimberley−Northern and Northern−Eastern province boundaries and a major refinement of subprovinces of Kimberley Province in light of substantial increases in sampling and taxonomic knowledge for the area (Morgan et al., 2011;Shelley, Morgan et al., 2018). We found that the influence of high elevation geological features, such as the Kimberley and Arnhem plateaus and Eastern Highlands, on aquatic habitat as well as stream density and light, thermal and moisture regimes are the most important environmental factors influencing species turnover. Historic low sea-level drainage patterns were also influential, highlighting the importance of history in explaining contemporary patterns of biodiversity in the AMT.
Our analysis of richness and endemism using the larger data set mostly reinforced previously described patterns (Kennard, 2010;Unmack, 2001Unmack, , 2013. The Daintree and Mulgrave catchments in the Wet Tropics are unique in that they are focal points of both narrow-range endemism and species richness and are thus of great conservation importance. The catchments surrounding the Kimberley Plateau also represent focal points of outstanding narrow-range endemism.
The patterns of freshwater fish diversity identified here provide a foundation for future biogeographic studies. Continued sampling, especially in the regions that we identified as poorly sampled or data deficient, is necessary to refine our proposed provinces. A thorough understanding of the basis for these patterns as well as the environmental/habitat preferences of the freshwater fish fauna of the AMT are vital avenues of further research to help inform conservation planning decisions.