Impact of hydroclimatic variability on regional-scale landscape connectivity across a dynamic dryland region

Impact of hydroclimatic variability on regional-scale landscape connectivity across a dynamic dryland region

Ecological Indicators xxx (xxxx) xxx–xxx Contents lists available at ScienceDirect Ecological Indicators journal homepage: www.elsevier.com/locate/e...

1MB Sizes 0 Downloads 12 Views

Ecological Indicators xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Ecological Indicators journal homepage: www.elsevier.com/locate/ecolind

Original Articles

Impact of hydroclimatic variability on regional-scale landscape connectivity across a dynamic dryland region ⁎

Robbi Bishop-Taylor , Mirela G. Tulbure, Mark Broich School of Biological, Earth and Environmental Sciences, The University of New South Wales, Biological Sciences Building (D26), Randwick, NSW 2052, Australia

A R T I C L E I N F O

A B S T R A C T

Keywords: Surface-water dynamics Graph theory Network analysis Dynamic connectivity Landscape connectivity Circuit theory

In dynamic dryland regions, accounting for spatiotemporal landscape dynamics is essential to understanding how ecological habitat networks are affected by hydroclimatic variability at regional or sub-continental scales. Here we assess how changes in the distribution and availability of surface water influence potential landscape connectivity for water-dependent organisms by combining graph theory network analysis with a Landsat-derived, seasonally continuous 25-year surface-water time-series. We focused on Australia’s Murray Darling Basin (MDB), a globally significant and ecologically stressed 1 million km2 semi-arid region recently affected by two unprecedented hydroclimatic extremes: the 1997–2010 Millennium Drought and 2010–2012 La Niña floods. We constructed potential habitat networks for two dispersal abilities using circuit theory resistance distances, and used ‘habitat availability’ graph theory metrics as indicators of regional-scale connectivity. We analysed 792 unique potential habitat networks containing over 6.6 million nodes, making our study one of the largest spatially explicit ecological network analyses yet conducted. Our indicators of connectivity revealed consistently positive but spatially heterogeneous relationships between flooded habitat area and landscape connectivity. Connectivity increased by over two orders of magnitude along the spectrum from severe drought to flood, associated with a transition from connectivity driven by intra-habitat or short-distance dispersal during drought to long-distance dispersal during wet conditions. Reductions in connectivity during drought were lower than expected given equivalent decreases in surface water habitat area, suggesting habitat network structure provides a degree of resistance to dry conditions. By providing insights into the processes driving connectivity during different phases along the drought-flood spectrum, our approach may assist in guiding conservation management aimed at maintaining or improving landscape connectivity within dynamic environments faced with increasing hydroclimatic variability.

1. Introduction Many of the world’s dryland regions are predicted to experience warmer, drier and more variable climatic conditions before the end of the 21st century (Finlayson et al., 2011; Leblanc et al., 2012). Dryland river and floodplain organisms may be particularly vulnerable to increasing hydroclimatic variability given their reliance on large, rare flooding events to provide both temporary habitat and transient opportunities for dispersal between isolated populations and breeding sites (Zeigler and Fagan, 2014). Conversely, prolonged or severe drought can adversely impact dryland species by increasing habitat isolation, reducing gene flow and inflating the risk of local extinction. In south-eastern Australia’s Murray-Darling Basin (MBD)—one of the world’s largest, driest and most dynamic river basins—changes in the frequency and intensity of flooding and drought will affect an already stressed ecological system that has lost over 50% of its floodplain ⁎

wetlands to intensive agricultural development and flow modification (Kingsford, 2000; Murray Darling Basin Authority, 2010). The potential ecological consequences of increasingly intense hydroclimatic events in the region were recently highlighted by the most severe drought on instrumental record: the 1997–2010 Millennium Drought or “Big Dry” (Leblanc et al., 2012). By both degrading surface water habitats and decreasing connectivity between remaining habitats, this 13-year drought had devastating effects on the health of many of the MDB’s water-dependent ecosystems that persist today despite record rainfall (the “Big Wet”) breaking the drought in 2010–2012 (Chessman, 2011; Mac Nally et al., 2014). Understanding how hydroclimatic variability affects habitat networks is therefore increasingly important to conservation (Pilliod et al., 2015). There is a growing recognition of the value of ecological indicator variables for assessing landscape connectivity across space and time (Hernández et al., 2015). To analyse and monitor potential

Corresponding author. E-mail address: [email protected] (R. Bishop-Taylor).

http://dx.doi.org/10.1016/j.ecolind.2017.07.029 Received 2 March 2017; Received in revised form 23 May 2017; Accepted 11 July 2017 1470-160X/ © 2017 Elsevier Ltd. All rights reserved.

Please cite this article as: Bishop-Taylor, R., Ecological Indicators (2017), http://dx.doi.org/10.1016/j.ecolind.2017.07.029

Ecological Indicators xxx (xxxx) xxx–xxx

R. Bishop-Taylor et al.

Fig. 1. The Murray Darling Basin (MDB) study area in southeastern Australia, including the maximum extent of surface water throughout the entire 25-year time-series (blue; Tulbure et al., 2016) and the location of the three subregions: the Murray, Paroo and Riverina. Insets show the proportion of important land-use classes within each subregion (ABARES, 2014).

either randomly or under the assumption that small water bodies will dry and disappear first during dry conditions (Fortuna et al., 2006; O’Farrill et al., 2014; Tulbure et al., 2014). This may result in misleading estimates of connectivity if the loss of habitats during drought does not consistently vary as a function of waterbody size. Simulating surface-water dynamics is particularly challenging across large and heterogeneous regions like the MDB, where the distribution of surface water is driven by a complex range of factors including river flow, local rainfall, evapotranspiration and soil moisture (Heimhuber et al., 2016; Heimhuber et al., 2017). Despite these limitations, prior approaches have consistently reported positive and often non-linear increases in connectivity during simulated wet conditions, particularly as ‘steppingstone’ habitats form that support long-distance dispersal (Bishop-Taylor et al., 2015). The impact of drought on habitat network structure has been more variable: forested waterhole networks in Mexico (O’Farrill et al., 2014) fragmented rapidly during simulated drought conditions, while pond networks in Spain revealed a robustness to drought with locally well-connected habitats facilitating short-distance amphibian dispersal even as most habitats were removed from the landscape (Fortuna et al., 2006). Graph theory combined with remotely-sensed surface-water timeseries represents a promising new approach for more realistically assessing the impact of hydroclimate variability on connectivity in spatiotemporally dynamic surface-water environments (Ruiz et al., 2014; Tulbure et al., 2014). Time-series data has recently been used to identify important surface-water habitats for short and long-distance dispersal along a spectrum of observed drought to inundation within the

landscape connectivity across large and dynamic study areas, previous studies have used graph theory network analysis (Fortuna et al., 2006; O’Farrill et al., 2014; Tulbure et al., 2014; Bishop-Taylor et al., 2015). Graph or network connectivity models represent landscapes as sets of discrete “nodes” (e.g. potential habitats or populations) that are connected by “edges” (or links) if ecologically connected (Urban and Keitt, 2001; Galpern et al., 2011). These connections are typically based on either Euclidean distance between habitats (i.e. two habitats are considered connected if located within an organism’s specified dispersal distance), or effective distance measures such as least-cost paths or circuit theory which use cost-distance surfaces to account for landscape resistance to movement (i.e. a water-dependent organism may move more easily through moist terrain than dry land). By analysing connectivity consistently across space and time, graph theory can provide valuable insights into network properties such as the ability of a network to facilitate long-distance dispersal and gene flow, or absorb or bounce back from the loss of habitats or ecological connections (Bunn et al., 2000; Urban and Keitt, 2001). Landscape connectivity studies in dynamic surface water environments have to date largely analysed connectivity between static sets of waterbodies (e.g. Fortuna et al., 2006; O’Farrill et al., 2014; Drake et al., 2017), between waterbodies at pre-defined dry or wet time-steps (e.g. Wright, 2010; Uden et al., 2014), or by using modelled surface water availability or flooding scenarios (e.g. Bishop-Taylor et al., 2015; Pilliod et al., 2015). In the absence of dynamic data, many of these studies have simulated the effects of drought on surface-water networks using node removal experiments that progressively remove habitats 2

Ecological Indicators xxx (xxxx) xxx–xxx

R. Bishop-Taylor et al.

pixels), enabling computationally intensive connectivity analyses for the entire extensive spatial (∼1 million km2) and temporal (25 year) data extent using reproducible Python 2.6 code (Python Software Foundation 2015) and the GDAL (GDAL Development Team, 2015), NumPy and SciPy packages (van der Walt et al., 2011). To identify potential surface-water habitats, we selected raster cells that contained water for the majority of each of 99 southern hemisphere seasonal timesteps (e.g. a cell was treated as potential habitat if it was detected as flooded in more than 50% of Landsat passes during Summer 2010). Areas inundated for less than the majority of a season were also identified to serve as optimal movement terrain in subsequent circuit theory resistance modelling. Where data was missing for a raster cell for an entire season (e.g. through consistent cloud cover), we filled ‘nodata’ cells on a per-cell basis based on the closest three years with data for the respective season (Tulbure and Broich, 2013; Bishop-Taylor et al., 2017). To consistently compare connectivity across time and space, we assigned each potential surface-water habitat cell with a unique, temporally consistent patch identifier (unique ID) using a modularity-based approach implemented by Bishop-Taylor et al. (2017) that identified 277,874 discrete ‘communities’ of raster cells that consistently occurred together across the entire time-series. The 99 seasonal layers were then labelled using these unique IDs, dividing potential surface-water habitats into discrete patches that reflected both the spatial structure and temporal synchrony of surface water throughout the time-series (Cavanaugh et al., 2014; for an extended description of this modularity approach readers are referred to Bishop-Taylor et al., 2017).

Swan Coastal Plain bioregion in south-west Western Australia (Tulbure et al., 2014). In Australia’s MDB, the approximately 25-year Landsat archive encompasses both the 1997–2009 Millennium Drought and 2010–2011 La Niña flooding, providing a unique opportunity to holistically quantify how hydroclimatic variability influences surface-water landscape connectivity across time and large spatial extents. In this study, we combined ecological indicators based on graph theory network analysis with a newly available, 25-year seasonally continuous Landsat surface-water time-series (Tulbure et al., 2016). We generated networks using circuit theory, an approach for modelling landscape resistance to movement that correlates strongly with gene flow for freshwater organisms including amphibians (Moore et al., 2011; Peterman et al., 2014). By evaluating ecological indicators consistently across time and at unprecedented spatial scale, our approach tested the following hypotheses regarding the impact of drought and flooding on regional- and sub-continental-scale landscape connectivity: 1. Long-distance movement opportunities created by increasingly large flooding will cause connectivity to increase rapidly relative to inundated potential habitat area 2. The loss of long-distance connections during drought will cause disproportionate decreases in connectivity as habitat networks fragment 2. Materials and methods 2.1. Study area

2.3. Circuit and graph theory modelling

South-eastern Australia’s MDB (Fig. 1) is the country’s second largest river basin (over 1 million km2) and home to its three longest river systems: the Darling (approximately 2740 km long), Murray (2530 km), and Murrumbidgee (1690 km). The MDB contains some of Australia’s most biologically diverse and ecologically significant wetland and floodplain ecosystems, including highly dynamic permanent and ephemeral habitats that support 124 aquatic macroinvertebrate families (including the iconic and endangered Murray Crayfish; Euastacus armatus), 55 of Australia’s 217 native amphibians (including six International Union for Conservation of Nature threatened and four critically endangered species), and five freshwater turtles (Ballinger and Mac Nally, 2006; Rogers and Ralph, 2010; Pittock and Finlayson, 2011). Surface-water habitats in the MDB have suffered severe declines in ecological health over the past century, driven primarily by intensifying agricultural water resource development (Kingsford, 2000). To evaluate connectivity across this highly heterogeneous study area, we focused on the entire MDB as well as three sub-regions (Fig. 1) with varying degrees of development selected based on previously published eco-hydrological zonations (Overton et al., 2009; Huang et al., 2013): the Paroo (a minimally disturbed and highly dynamic arid-zone catchment in the north-west MDB), the lower Murray (a southern MDB region containing both natural floodplain wetlands and dryland and irrigated agriculture), and the Riverina (a highly regulated south-east MDB region bordered containing extensive irrigated agriculture). These approximately equal sized regions (69,000–73,000 km2) were buffered by 10 km to allow for connectivity between waterbodies occurring along region boundaries (Bishop-Taylor et al., 2015).

We quantified surface-water connectivity for each region (entire MDB, Paroo, Murray and Riverina) and season in the 1987–2011 timeseries by constructing networks based on circuit theory resistance distances between all neighbouring habitat patch pairs. Circuit theory calculates effective distances between habitats that account for all possible movement pathways through a heterogeneous landscape matrix using random walks (McRae et al., 2008). To incorporate the influence of the landscape matrix on dispersal, we used a resistance layer with increasingly high resistance values in dry terrain (see Table A1) that was previously used to model landscape connectivity for waterdependent MDB species (i.e. aquatic invertebrates, amphibians and freshwater turtles; Bishop-Taylor et al., 2017). This resistance layer combined land cover and selected features extracted from topographic data (e.g. potential barriers to dispersal including major roads, urban infrastructure and saline waterbodies; see Table A.1) with our maximum flooding extent layers for each seasonal time-step (GA 2006; Lymburner et al., 2010). Circuit theory resistance distances were processed using the Circuitscape v4.04 software package (Shah and McRae, 2008) by dividing the study area into 50%-overlapping 12 × 12 km ‘moving windows’, and using seasonal habitat and resistance layers as inputs (Bishop-Taylor et al., 2017). Short and long-distance dispersal networks were then constructed by connecting unique habitats (nodes) with edges if separated by less than two specific resistance distances. These values (0.53 and 1.59) were identified by estimating the 95th percentile of circuit theory movement costs associated with maximum Euclidean dispersal distances of 1000 m and 5000 m (Bishop-Taylor et al., 2015) selected to encompass typical maximum dispersal distances for freshwater invertebrates (O’Connor, 1986), amphibians (Smith and Green, 2005) and turtles (Roe et al., 2009). We used the Equivalent Connected Area (ECA) metric as an indicator of the effects of flooding and drought on landscape connectivity (Saura et al., 2011). Based on the ‘habitat availability’ concept, ECA measures connectivity by accounting for both the habitat area contained within patches themselves, and the additional habitat made available to dispersing organisms by connections between habitat patches (Baranyi et al., 2011; Saura et al., 2011). ECA and related habitat

2.2. Seasonal surface water We used a newly available 1987–2011 validated surface water and flooding dynamics product based on Landsat TM and ETM+ remote sensing imagery for data on the spatiotemporal distribution of surface water within the MDB (Tulbure et al., 2016). To provide continuity with previous local-scale connectivity modelling in the MDB, we constructed habitat networks using a procedure to identify potential habitat areas developed by Bishop-Taylor et al. (2017). All data processing was conducted on spatially aggregated rasters (from 30 m to 120 m 3

Ecological Indicators xxx (xxxx) xxx–xxx

R. Bishop-Taylor et al.

subsequent 2010–2011 La Niña floods (e.g. from 292 to 2482 nodes in the Murray). The Equivalent Connected Area (ECA) indicator of connectivity increased with flooding and decreased during dry conditions for all regions, although by different magnitudes and with different relationships to inundated potential habitat area (Fig. 4). Across the entire MDB, the Murray subregion and the Riverina, ECA decreased only marginally during the driest seasons (i.e. from 873 km2 within the entire MDB during a 25th percentile flood to 709 km2 during the driest season for long-distance dispersers; Fig. 4a). This gradual decrease was associated with a steadily increasing ECA:area ratio, indicating that the loss of inundated habitat during dry seasons did not proportionally reduce connectivity (Fig. 4b). Large floods in the MDB and Murray saw a rapid increase in ECA for both dispersal distances, with increases of ∼170% between a 75th percentile flood and the wettest season in both regions. Although the relatively unmodified and highly ephemeral Paroo had the lowest ECA values of any region during the driest seasons, connectivity increased rapidly by over two orders of magnitude or 43400% between the driest to wettest seasons (e.g. from 3 to 1310 km2). The largest Paroo floods were also associated with the most rapid increase in ECA:area (e.g. from ∼0.24 for a 75th percentile flood to ∼0.40 during the wettest season), indicative of connectivity increasing faster than potential habitat area alone as newly flooded habitats connected large areas of usually dry arid-zone floodplains (e.g. Fig. 2). In the Murray and across the entire MDB, increases in ECA for both dispersal distances were matched with a more modest ∼5% increase in ECA:area between a 75th percentile flood and the wettest season (Fig. 4b). In the highly modified Riverina, large inundation events produced different relationships between ECA and habitat area depending on dispersal ability. While ECA for short-distance dispersers increased by 230% from 60 km2 to 198 km2 across the range of floods in the timeseries (Fig. 4a), this increase was outpaced by a ∼700% increase in potential habitat area during the same period. Accordingly, > 75th percentile flooding in the Riverina caused ECA:area for short-distance dispersers to move in the opposite direction to other subregions, decreasing to an overall minimum of 0.12 during the wettest season (Fig. 4b). The highest ECA:area ratio for both dispersal distances occurred during the driest Riverina seasons, when the only water remaining in the region occurred along the perennial Murray and Riverina river channels, adjacent floodplains, and major on-stream reservoirs (e.g. Lake Hume and Lake Mulwala in the southern MDB) that retained a degree of longitudinal connectivity during even the driest of periods (Figs. 2 and 4b). Partitioning ECA into its constituent components (ECAintra, ECAdirect and ECAstep) provided additional insights into the influence of different aspects of connectivity across the dry-wet spectrum and between regions and dispersal distances. Intra-patch connectivity (ECAintra) or the contribution of habitat area alone was consistently more important for short-distance dispersers, and was the most important component across the majority of dry and average seasonal time-steps (Fig. 4c). This was particularly true for the Paroo, where ECAintra made up between 74 and 99% of ECA for short-distance dispersal networks until at least a 75th percentile flood. During these wetter seasons, ECAintra remained relatively high at ∼33% for short-distance dispersal organisms in the Riverina (Fig. 4c), but rapidly decreased for long-distance dispersers to approximately 11–16% of ECA across all other regions. Direct connections between neighbouring habitats (ECAdirect) contributed 0–30% of overall connectivity for all regions across the timeseries (Fig. 4d). While ECAdirect increased with flooding for both dispersal abilities in the Murray and for short-distance dispersers in the Paroo, other regions either displayed little clear relationship with flooding (e.g. Riverina for short-distance, Paroo for long-distance dispersal), or a gradual decrease in importance (e.g. the entire MDB and Riverina for long-distance dispersal) as stepping-stones (ECAstep) played an increasingly important role in facilitating connectivity (Fig. 4d, e).

availability metrics can outperform more commonly used graph metrics in complex landscapes, including providing better predictions of surface-water habitat occupancy for an endangered turtle species threatened by agricultural intensification (Pereira et al., 2011). ECA is defined as: n

ECA =

n

∑ ∑ i=1

ai aj pij*

j=1

where ai and aj are the areas of habitat patches i and j, and pij* is the maximum product probability of all possible dispersal paths between patches i and j. ECA has area units (e.g. km2), and is equal to the total area of all habitat patches in the network in a completely connected network (i.e. where pij* = 1 for all patch pairs). ECA will be never be below the size of the largest single habitat patch, even in a network with zero probability of movement between all patches (i.e. an entirely fragmented network with all patches fully isolated from each other). ECA’s area units allow it to be directly compared to changes in total habitat area to study how network spatial configuration and the addition or loss of habitat area affects connectivity. In addition to ECA, we therefore calculated a ‘ECA:area’ metric by dividing ECA by total inundated habitat area for each season, giving a ratio that approached 1 for an optimally connected network, and gave low values where habitat area corresponded poorly with network-wide connectivity (e.g. a fragmented network with large but isolated areas of habitat). To provide insights into shifts in the factors driving connectivity in the MDB (i.e. short vs. long distance dispersal), we partitioned ECA into three components that quantified the contribution to connectivity provided (i) by the area within habitats alone (i.e., intrapatch connectivity; ECAintra), (ii) via direct connections between neighbouring patches (ECAdirect), and (iii) via longer distance movements using intermediate ‘stepping-stone’ habitats (ECAstep; Saura and Rubio, 2010; Saura et al., 2014). All indicators were calculated using the Conefor v2.6 command line software (Saura and Torne, 2009) using the binary IIC formulation of ECA with nodes weighted by each habitat’s seasonal inundated area. These indicators were plotted against total percentileranked inundated seasonal habitat area to assess connectivity across a spectrum from dry to wet. To aid interpretations, plot x-axes were labelled to highlight dry seasons (e.g. < 25th percentile habitat area; white bars), average seasons (e.g. 25–50th percentile; grey bars) and wet seasons (e.g. > 75th percentile; dark coloured bars). 3. Results We analysed connectivity across 99 seasonal time-steps (1987–2011), two ecologically relevant dispersal abilities and four heterogeneous regions, including an entire 1 million km2 semi-arid basin. Our resulting 792 unique potential surface-water habitat networks (e.g. Fig. 2) contained over 6.6 million nodes, making our study one of the largest regional-scale ecological network analyses yet conducted. Over the 25-year time-series, networks displayed significant variability between regions. Within the entire MDB, surface-water networks ranged between 9461 and 57354 nodes (i.e. potential habitats), with the lowest numbers corresponding largely with the 1997–2009 Millennium Drought, and highest values with either the 2010–11 La Niña flooding or other major flooding events in 1990, 1998 and 2000 (Fig. 3). Edge counts were closely correlated with nodes, but were typically 6–11 times higher for long-distance compared to shortdistance dispersal organisms. Out of the three subregions, the Paroo was the most variable overall, with node and edges varying by over two orders of magnitude across the time-series (e.g. 64 edges for long-distance dispersers in the locally driest Autumn 2005 season compared to 36503 edges in the wettest Autumn 1990 season). Nodes and edge counts were more stable in the Murray and Riverina, however increased by almost 10 times between the driest period (2006–2009) and the 4

Ecological Indicators xxx (xxxx) xxx–xxx

R. Bishop-Taylor et al.

Fig. 2. Potential habitat networks compared between the driest and wettest season by potential habitat area across the entire Murray-Darling Basin (MDB). Nodes (grey) are symbolised by potential inundated habitat area and connected with graph edges for long-distance dispersal organisms (∼5000 m, blue lines). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

directly compared with changes in potential habitat area, providing a more realistic assessment of how hydroclimatic variability and surfacewater dynamics affect the function availability (or ‘reachability’) of surface-water habitat to dispersing organisms at regional to subcontinental scales. These indicators highlighted important differences in the relationship between flooded area and connectivity both across space and by dispersal distance, and revealed key transitions in processes driving connectivity along the wet-dry spectrum.

Throughout dry and average periods, stepping-stone importance (ECAstep) was consistently highest (35–40%) within the Murray region (containing the perennially flooded Murray River) for both dispersal distances, and lowest (0–30%) in the Paroo, particularly for short-distance dispersers (0–15%; Fig. 4e). The contribution of stepping-stones rose rapidly for both dispersal distances during major (> 75th percentile) inundation events in the Paroo (e.g. from 32 to 77% for longdistance and 16–75% for short-distance dispersers) and across the entire MDB (e.g. from 33 to 79% for long-distance and 30–74% for shortdistance dispersers). Although the importance of stepping-stones also increased sharply for long-distance dispersers during large floods in the Riverina (41–83% of ECA), values for short-distance dispersers never exceeded 50% contribution to ECA (Fig. 4e).

4.1. Impact of flooding on connectivity For all regions, increasingly wet conditions were consistently associated with higher overall connectivity (i.e. ECA) as flooding increased both potential habitat area and inter-patch dispersal opportunities (Fig. 4a). Across the entire MDB and within the highly regulated Murray region, major flooding led to rapid increases in connectivity that were typically proportional to increases in flooded area itself (i.e. leading to only a marginally higher ECA:area ratio). In contrast, major flooding in the unregulated and highly dynamic Paroo region produced extreme increases in connectivity of over two orders of magnitude between the driest and wettest season in the time-series (Fig. 4a), greatly outpacing increases in inundated habitat area and resulting in a rapidly increased ECA:area ratio. Compared to floods in more regulated regions of the MDB, these floods not only increased habitat area but importantly, also increased the functional availability (‘reachability’) of surface-water habitats to dispersing organisms (Bodin and Saura, 2010). Differences in the effectiveness of additional surface water for increasing landscape connectivity across different regions of the MDB may be attributable to variation in the spatial structure of flooding. Across all regions of the MDB, moving from dry or average conditions to wet conditions resulted in a shift from connectivity driven by intrahabitat connectivity and direct dispersal opportunities between neighbouring habitats (ECAintra, ECAdirect), to connectivity influenced

4. Discussion Remote sensing time-series data represents an ideal resource for modelling landscape connectivity across space and time (Ruiz et al., 2014; Tulbure et al., 2014; Bishop-Taylor et al., 2017). In this study, we used high-resolution spatiotemporal data on the distribution of potential ecological habitats and flooding inundation to study how hydroclimatic variability affects connectivity through dynamic dryland floodplain habitat networks at regional or sub-continental scales. By combining a newly available, seasonally continuous Landsat time-series (Tulbure et al., 2016) with a circuit theory resistance model, we modelled landscape connectivity over a 25-year period which encompassed some of the most extreme hydroclimatic events on record for southeastern Australia: the 1999–2009 Millennium Drought and the 2010–2011 La Niña flood years. Using this spatiotemporally consistent graph theory framework, we calculated indicators of connectivity based on the concept of ‘habitat availability’ that accounted for both variability in habitat area and changes in the potential for organisms to disperse successfully between habitats across time (Saura et al., 2011). Importantly, these indicators allowed changes in connectivity to be 5

Ecological Indicators xxx (xxxx) xxx–xxx

R. Bishop-Taylor et al.

Fig. 3. 1987–2011 time-series showing number of graph nodes (potential unique habitats) and edges (links between habitats) for four regions: the entire Murray-Darling Basin (MDB), the Murray, Riverina and Paroo. Metrics are shown in black where values were identical for both dispersal abilities (e.g. total nodes which was not affected by dispersal distance), and in blue and orange for long-distance (∼5000 m) and short-distance (∼1000 m) dispersal organisms respectively. To aid interpretations, coloured bars on the x-axis highlight the driest 25% of seasons by inundated habitat area (i.e. white bars), the intermediate 25–75th percentile (grey bars) and the wettest 25% (dark coloured bars). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

for short-distance dispersers within highly modified surface-water environments have been previously observed in local connectivity modelling within the MDB (Bishop-Taylor et al., 2015) and agricultural wetland habitat networks in the USA (Uden et al., 2014), and may be explained by interactions between surrounding land use and the spatial structure of inundation in these regions. In the Riverina, the majority of additional water during the wettest seasons was concentrated in agricultural areas between the Murray and Murrumbidgee rivers (e.g. Fig. 5). Embedded in a mixed dryland and irrigated agricultural matrix, the large, regularly spaced inundated fields in this region remained relatively isolated from the two major river channels and gained few additional connections during even the wettest seasons. While many Australian floodplain species including the endangered Southern Bell Frog (Litoria raniformis) are known to inhabit and successfully breed in flooded agricultural fields (Pyke and Muir, 2008), our results suggest that short-distance dispersal organisms will be restricted in their ability to disperse through habitat networks despite the widespread availability of water, potentially negatively influencing metapopulation dynamics by requiring more time and effort for organisms to connect with other individuals of their species (Hanski, 1999).

primarily by long-distance movement opportunities (ECAstep). In the Paroo, major flooding inundated extensive areas between existing, highly clustered groups of surface-water habitats (Figs. 1 and 2). By providing long-distance dispersal opportunities across hundreds of square kilometres of usually impermeable movement terrain, these new potential habitats served as ‘stepping-stones’ that resulted in faster than expected increases in connectivity relative to the addition of new potential habitat (Saura et al., 2011). The prevalence of ‘stepping-stone’ habitats has been previously identified as critical for maintaining connectivity through surface-water habitats (Tulbure et al., 2014; BishopTaylor et al., 2015; Watts et al., 2015), and may improve long-term population persistence by supporting gene flow, aiding the recolonization of isolated habitats after local extinction events, and by facilitating range expansion during large-scale environmental change (Saura et al., 2014). Our findings suggest that while these habitats may be transient features of their dynamic environments, they are likely to play an outsized role in maintaining connectivity through regionalscale surface-water networks. By comparing indicators of connectivity across different regions of the MDB, we observed differences in how habitat network structure and flooding affect connectivity for organisms with different dispersal abilities. While flooding improved overall connectivity in the highly modified Riverina region (i.e. an absolute increase in ECA), improvements in ECA for short-distance dispersal organisms were small relative to the addition of newly inundated habitat area. Decreased connectivity

4.2. Impact of drought on connectivity While drought led to an absolute and significant decrease in total available (connected) habitat across all our study regions, our results 6

Ecological Indicators xxx (xxxx) xxx–xxx

R. Bishop-Taylor et al.

indicate that connectivity in the MDB and all three subregions does not decrease as severely during drought as would be expected given the loss of total surface-water habitat (Fig. 4b). During even the driest of seasons, core sets of potential habitats persisted in areas of the landscape with optimal movement conditions for water-dependent organisms (e.g. terrain with high relative moisture such as along perennial rivers or within floodplain swamps and forest). These habitats remained connected even as more isolated dryland habitats (e.g. waterholes, ephemeral lakes) disappeared during drought. Our results closely resemble those of Fortuna et al. (2006) who found that the spatial structure of a Mediterranean amphibian habitat network minimized the negative effects of drought by continuing to provide opportunities for dispersal even as the total number of flooded ponds reduced. While more regularly spaced forested waterhole networks studied by O’Farrill et al. (2014) showed a rapid decrease in connectivity during drought, our results suggest that the complex spatial structure of floodplain habitat networks and the continued presence of vital floodplain refuges may provide a degree of resistance to even severely dry conditions. Given our findings, different conservation management approaches may be required to maintain or improve landscape connectivity for water-dependent organisms at different phases in the wet-dry spectrum. Despite the potential robustness of our surface-water networks to dry conditions, river regulation and a shift towards a drier, more variable regional climate are likely to affect landscape connectivity by reducing critically important opportunities for dispersal between already fragmented and isolated surface-water habitats (CSIRO, 2008; Pilliod et al., 2015). Given the increased importance of intra-habitat connectivity (ECAintra) and direct connectivity (ECAdirect) during dry periods, protecting and managing persistent, well connected riverine and floodplain habitats and improving lateral connections between off-channel waterbodies and floodplain refuges through targeted habitat restoration may provide the greatest benefits to landscape connectivity. During wetter conditions where long-distance connections (ECAstep) drive the largest increases in connectivity, protecting or restoring ephemeral ‘stepping-stone’ habitats may provide the greatest conservation benefits, particularly in systems still subject to dynamic, relatively natural flooding regimes. By providing insights into shifting drivers of connectivity in dynamic, highly modified regions like the MDB, indicators of connectivity based on remotely sensed time-series data may help to ensure that the allocation of limited conservation funding or contested environmental water provides maximum connectivity benefits along the entire spectrum from extreme drought to major flood. 4.3. Limitations and future work To make our regional-scale indicators of connectivity broadly applicable to a diverse range of water-dependent organisms (e.g. amphibians, aquatic macroinvertebrates, turtles), our potential habitat networks relied on circuit theory resistance distances generated from a simple resistance layer based on assumed landscape moisture. Our analyses could be adapted for specific focal species or global regions as improved resistance layers based on empirical data become available (e.g. from telemetry or genetic studies). Our use of potential habitat area to weight habitat nodes in our connectivity analyses could also be improved by incorporating field-derived or modelled estimates of habitat quality as part of a quality-weighted habitat area (Minor and Urban, 2007). This could enhance our findings by accounting for variation in habitat suitability at either local (e.g. the presence of aquatic vegetation or predator species within individual habitats) or regional scales (e.g. along climatic gradients), or by incorporating possible changes in habitat quality that occur during periods of drought or flood. Importantly however, the graph theory-based ecological indicators used in this current study were applied consistently across space and time, allowing us to reveal in relative terms how flooding, drought and landscape structure combine to affect potential connectivity seasonally, annually and inter-annually, and at regional scales.

Fig. 4. Graph metrics (ECA, ECA:area ratio, ECAintra, ECAdirect, ECAstep) relative to inundated habitat area for four regions: the entire Murray-Darling Basin (MDB), the Murray, Riverina and Paroo. Low x-axis percentiles represent dry seasons (e.g. < 25th percentile; white bars), while high percentiles indicate wet seasons or major flooding (e.g. > 75th percentile; dark coloured bars). Results are shown for long (∼5000 m, blue) and short-distance (∼1000m, orange) dispersal abilities. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

7

Ecological Indicators xxx (xxxx) xxx–xxx

R. Bishop-Taylor et al.

Fig. 5. Potential surface water habitat during a typical wet Riverina seasonal time-step (Spring 1993), highlighting the distribution of inundated fields located between the Murray and Murrumbidgee rivers (Tulbure et al., 2016).

Acknowledgements

While observing the natural progression of remotely sensed surfacewater bodies along a spectrum of observed drought to flood advances static connectivity approaches, to date very few studies (Rubio et al., 2014; Lloyd et al., 2016) have compared how connectivity indicators based on dynamic remote-sensing data compare to modelling based on static datasets. Future work comparing these approaches may provide valuable insights into the advantages and disadvantages of static and dynamic graph theory approaches for providing insights into landscape connectivity, particularly in highly dynamic landscapes subject to extreme changes in both habitat availability and the potential for dispersal between habitats.

This work was funded through an Australian Research Council Discovery Early Career Researcher Award (DE140101608) to Tulbure. The eco-hydrological zonation of the MDB used in this study was originally developed by CSIRO (Overton et al., 2009) and later adapted (Huang et al., 2013). We would like to thank B McRae and S Saura for their timely assistance with the Circuitscape and Conefor software packages.

Appendix A

Table A1 Resistance to movement values used to develop resistance surfaces (based on GA topographic features and Australian Dynamic Land Cover Dataset v1.0 classes; Geoscience Australia, 2006; Lymburner et al., 2010). Value

Resistance

Example DLCD land cover or GA topographic features

1

Very low

2

Low

4 8

Moderate Moderate–high

16 32

High Very high

Surface water (including seasonal maximum flooding extent), aquatic vegetation Closed natural vegetation, irrigated cropping and pasture Open natural vegetation, dryland pasture Sparse natural vegetation, dryland cropping, minor roads Non-vegetated areas, major roads Urban infrastructure and highways, saline waterbodies

Appendix B. Supplementary data Supplementary data associated with this article can be found online at http://dx.doi.org/10.1016/j.ecolind.2017.07.029.

Baranyi, G., et al., 2011. Contribution of habitat patches to network connectivity: redundancy and uniqueness of topological indices. Ecol. Indic. 11.5, 1301–1310. Bishop-Taylor, R., et al., 2015. Surface water network structure: landscape resistance to movement and flooding vital for maintaining ecological connectivity across Australia’s largest river basin. Landsc. Ecol. 30, 2045–2065. Bishop-Taylor, R., et al., 2017. Surface water dynamics and land use influence landscape connectivity across a major dryland region. Ecol. Appl. http://dx.doi.org/10.1002/

References ABARES, 2014. Catchment Scale Land Use of Australia. Australian Bureau of Agricultural and Resource Economics and Sciences Update March 2014. Ballinger, A., Mac Nally, R.C., 2006. The landscape context of flooding in the MurrayDarling Basin. Floods Arid Cont. Adv. Ecol. Res. 39, 85–105.

8

Ecological Indicators xxx (xxxx) xxx–xxx

R. Bishop-Taylor et al.

Overton, I.C., et al., 2009. Ecological Outcomes of Flow Regimes in the Murray-Darling Basin. Report Prepared for the National Water Commission by CSIRO Water for a Healthy Country Flagship. CSIRO. Pereira, M., et al., 2011. Using spatial network structure in landscape management and planning: a case study with pond turtles. Landsc. Urban Plan. 100, 67–76. Peterman, W.E., et al., 2014. Ecological resistance surfaces predict fine-scale genetic differentiation in a terrestrial woodland salamander. Mol. Ecol. 23, 2402–2413. Pilliod, D.S., et al., 2015. Effects of changing climate on aquatic habitat and connectivity for remnant populations of a wide-ranging frog species in an arid landscape. Ecol. Evol. 5, 3979–3994. Pittock, J., Finlayson, C.M., 2011. Australia’s Murray − Darling Basin: freshwater ecosystem conservation options in an era of climate change. Mar. Freshw. Res. 62, 232. Pyke, G., Muir, G., 2008. Rice-growing and conservation of the southern bell frog Litoria raniformis in New South Wales, Australia. Aust. Zool 34, 453–458(Python Software Foundation 2015. Python Language Reference, version 2.6. Available: https://docs. python.org/2/index.html). Roe, J.H., et al., 2009. Temporal and spatial variation in landscape connectivity for a freshwater turtle in a temporally dynamic wetland system. Ecol. Appl. 19, 1288–1299. Rogers, K., Ralph, T.J., 2010. Floodplain wetlands of the Murray-Darling Basin and their freshwater biota. In: Rogers, K., Ralph, T.J. (Eds.), Floodplain Wetland Biota in the Murray-Darling Basin: Water and Habitat Requirements. CSIRO Publishing, pp. 1–16. Rubio, L., et al., 2014. Connectivity conservation priorities for individual patches evaluated in the present landscape: how durable and effective are they in the long term? Ecography (Cop.) 38, 782–791. Ruiz, L., et al., 2014. Dynamic connectivity of temporary wetlands in the southern Great Plains. Landsc. Ecol. 29, 507–516. Saura, S., Rubio, L., 2010. A common currency for the different ways in which patches and links can contribute to habitat availability and connectivity in the landscape. Ecography (Cop.) 33, 523–537. Saura, S., Torne, J., 2009. Conefor Sensinode 2.2: a software package for quantifying the importance of habitat patches for landscape connectivity. Environ. Model. Softw. 24, 135–139. Saura, S., et al., 2011. Network analysis to assess landscape connectivity trends: application to European forests (1990–2000). Ecol. Indic. 11, 407–416. Saura, S., et al., 2014. Stepping stones are crucial for species’ long-distance dispersal and range expansion through habitat networks. Frair, J. (Ed.), J. Appl. Ecol. 51, 171–182. Shah, V.B., McRae, B.H., 2008. Circuitscape: a tool for landscape ecology. Varoquaux, G., Vaught, T., Millman, J. (Eds.), Proc. 7th Python Sci. Conf. (SciPy 2008) 62–66. Smith, M.A., Green, D.M., 2005. Dispersal and the metapopulation paradigm in amphibian ecology and conservation: are all amphibian populations metapopulations? Ecography (Cop.) 28, 110–128. Tulbure, M.G., Broich, M., 2013. Spatiotemporal dynamic of surface water bodies using Landsat time-series data from 1999 to 2011. ISPRS J. Photogramm. Remote Sens. 79, 44–52. Tulbure, M.G., et al., 2014. Spatiotemporal dynamics of surface water networks across a global biodiversity hotspot—implications for conservation. Environ. Res. Lett. 9, 114012. Tulbure, M.G., et al., 2016. Surface water extent dynamics from three decades of seasonally continuous Landsat time series at subcontinental scale in a semi-arid region. Remote Sens. Environ. 178, 142–157. Uden, D., et al., 2014. The role of reserves and anthropogenic habitats for functional connectivity and resilience of ephemeral wetlands. Ecol. Appl. 24, 1569–1582. Urban, D.L., Keitt, T.H., 2001. Landscape connectivity: a graph-theoretic perspective. Ecology 82, 1205–1218. van der Walt, S., et al., 2011. The NumPy array: a structure for efficient numerical computation. Comput. Sci. Eng. 13, 22–30. Watts, A.G., et al., 2015. How spatio-temporal habitat connectivity affects amphibian genetic structure. Front. Genet. 6, 1–13. Wright, C.K., 2010. Spatiotemporal dynamics of prairie wetland networks: power-law scaling and implications for conservation planning. Ecology 91, 1924–1930. Zeigler, S.L., Fagan, W.F., 2014. Transient windows for connectivity in a changing world. Mov. Ecol. 2, 1.

eap.1507. Bodin, Ö., Saura, S., 2010. Ranking individual habitat patches as connectivity providers: Integrating network analysis and patch removal experiments. Ecol. Modell. 221, 2393–2405. Bunn, A., et al., 2000. Landscape connectivity: a conservation application of graph theory. J. Environ. Manag. 59, 265–278. CSIRO, 2008. Water Availability in the Murray. A Report to the Australian Government from the CSIRO Murray-Darling Basin Sustainable Yields Project. pp. 217. Cavanaugh, K.C., et al., 2014. Patch definition in metapopulation analysis: a graph theory approach to solve the mega-patch problem. Ecology 95, 316–328. Chessman, B.C., 2011. Declines of freshwater turtles associated with climatic drying in Australia’s Murray-Darling Basin. Wildl. Res. 38, 664–671. Drake, J.C., et al., 2017. Using nested connectivity models to resolve management conflicts of isolated water networks in the Sonoran Desert. Ecosphere 8, 1–22. Finlayson, C.M., et al., 2011. The status of wetlands and the predicted effects of global climate change: the situation in Australia. Aquat. Sci. 75, 73–93. Fortuna, M.A., et al., 2006. Spatial network structure and amphibian persistence in stochastic environments. Proc. R. Soc. B Biol. Sci. 273, 1429–1434. GDAL Development Team, 2015. GDAL − Geospatial Data Abstraction Library, Version 1.11.1. Available: http://gdal.osgeo.org. Galpern, P., et al., 2011. Patch-based graphs of landscape connectivity A guide to construction, analysis and application for conservation. Biol. Conserv. 144, 44–55. Geoscience Australia, 2006. GEODATA TOPO 250 K Series 3 (Packaged – Shape File Format). Available: http://www.ga.gov.au/metadata-gateway/metadata/record/ 64058/. Hanski, I., 1999. Habitat connectivity, habitat continuity, and metapopulations in dynamic landscapes. Oikos 87, 209–219. Heimhuber, V., et al., 2016. Modeling 25 years of spatio-temporal surface water and inundation dynamics on large river basin scale using time series of Earth observation data. Hydrol. Earth Syst. Sci. 20, 2227–2250. Heimhuber, V., et al., 2017. Modeling multidecadal surface water inundation dynamics and key drivers on large river basin scale using multiple time series of Earth-observation and river flow data. Water Resour. Res. 20, 2227–2250. Hernández, A., et al., 2015. Landscape dynamics and their effect on the functional connectivity of a Mediterranean landscape in Chile. Ecol. Indic. 48, 198–206. Huang, C., et al., 2013. GIS-based spatial zoning for flood inundation modelling in the Murray–Darling Basin. In: 20th Int. Congr. Model. Simulation. 1-6 December 2013. pp. 1700–1706. Kingsford, R.T., 2000. Ecological impacts of dams: water diversions and river management on floodplain wetlands in Australia. Austral Ecol. 25, 109–127. Leblanc, M., et al., 2012. A review of historic and future hydrological changes in the Murray-Darling Basin. Glob. Planet. Change 80–81, 226–246. Lloyd, M.W., et al., 2016. Temporal variability in potential connectivity of Vallisneria americana in the Chesapeake Bay. Landsc. Ecol. 31, 2307–2321. Lymburner, L., et al., 2010. The National Dynamic Land Cover Dataset. National Earth Observation Group, Geoscience Australia. Mac Nally, R., et al., 2014. Do frogs bounce, and if so, by how much?: Responses to the Big Wet following the Big Dry in south-eastern Australia. Glob. Ecol. Biogeogr. 23, 223–234. McRae, B.H., et al., 2008. Using circuit theory to model connectivity in ecology, evolution, and conservation. Ecology 89, 2712–2724. Minor, E.S., Urban, D.L., 2007. Graph theory as a proxy for spatially explicit population models in conservation planning. Ecol. Appl. 17, 1771–1782. Moore, J.A., et al., 2011. Effects of the landscape on boreal toad gene flow: does the pattern-process relationship hold true across distinct landscapes at the northern range margin? Mol. Ecol. 20, 4858–4869. Murray Darling Basin Authority, 2010. Guide to the Proposed Basin Plan: Overview. Murray–Darling Basin Authority, Canberra. O’Connor, P., 1986. The Biology of the Murray Crayfish, Euastacus Armatus (Decapoda: Parastacidae) and Recommendations for the Future Management of the Fishery. unpublished NSW Department of Agriculture data summary. O’Farrill, G., et al., 2014. The potential connectivity of waterhole networks and the effectiveness of a protected area under various drought scenarios. PLoS One 9, e95049.

9