Introduction

Habitat loss, which is driven by anthropogenic pressures such as agriculture, urbanisation, and resource extraction (Foley et al. 2005), is one of the greatest threats to biodiversity globally (Brooks et al. 2002; Hanski 2011; Newbold et al. 2015). Habitat loss is a substantial threat to bird biodiversity in Australia, with 30% of Australian bird species having lost at least 30% of their habitat (Simmonds et al. 2019). Habitat that remains is becoming increasingly fragmented, with particular vegetation communities subject to large declines in area and increased isolation (Tulloch et al. 2016). Such habitat loss has likely led to the disappearance of species over large areas of Australia, with some threatened species experiencing broad-scale extirpation (Ward et al. 2022). Even for species not apparently in decline, the modification and fragmentation of habitat may result in extinction debts (Tilman et al. 1994; Hanski and Ovaskainen 2002). A compounding issue for bird conservation is the geographical biases in habitat loss, which may disproportionately affect species with small ranges (which can already be predisposed to extinction (Harris and Pimm 2008), and particular habitat associations. Some habitat types are subject to greater losses and fragmentation, with approximately only 4.6% of closed forests remaining in Australia and several forest groups being composed primarily (95%) of patches (Bradshaw 2012; Tulloch et al. 2016).

Australia has retained high native bird richness in some human-modified landscapes, particularly in sprawling suburban areas with comparatively high levels of habitat preservation (Jones and Wieneke 2000; Sewell and Catterall 1998), and agricultural areas with retained native vegetation (Haslem and Bennett 2008; Martin and McIntyre 2007). However, while human-modified environments in Australia can be associated with high bird abundance (particularly of introduced species, though this varies with features of the environment), bird richness is typically low (Sewell and Catterall 1998; Chace and Walsh 2006; Evans et al. 2009). This is the case in many agricultural settings such as livestock grazing areas without tree cover (Martin and McIntyre 2007), and in highly modified urban environments, especially those with simple/non-native floristic structures (White et al. 2005). As the human population grows, Australia’s human-modified landscapes will likely not only expand but also intensify in land use, potentially moving more areas into conditions that support low bird species richness at the expense of native species. Given that the vast majority (87%) of the Australian human population lives within 50 km of the coast, and this proportion is increasing (Australian Bureau of Statistics 2020), some ecosystems may be more threatened than others by the growing human population.

Measuring the magnitude and impact of habitat modification and loss is important for estimating the extinction risk of individual species and ecosystems (Lee and Jetz 2010). Given the rapidity and unpredictability of change landscapes are undergoing worldwide, it is crucial that methods to monitor species’ habitat are dynamic enough to capture these changes so that species extinction risk can be reduced. The importance of monitoring common species’ extinction risk (Bennett et al. 2024; Lindenmayer et al. 2011) puts additional stress on the limited resources available for conservation, making broadly applicable methods that can be readily updated and applied to large suites of species are particularly valuable. One potential method for detecting habitat loss, and for prioritising research effort, is the species’ area of habitat concept (AOH, Brooks et al. 2019). AOH is the area within a species’ range that is likely habitat used, based on environmental variables such as elevation and land cover. While AOH is a useful quantification of species’ habitat, the requirement of multiple land-cover variables (and corresponding knowledge of species’ requirements) could result in reduced general applicability due to the data demands. A potential alternative, which we favour, is area of species feeding habitat (AFH), a metric developed by Simmonds et al. (2019). AFH is calculated through the restriction of more generalized spatial approaches, such as geographic range polygons created from presence-only sightings bounded by a minimum convex polygon (a method used by the IUCN (IUCN Standards and Petitions Committee 2022). Such ranges can be masked to include only areas with vegetation/land cover linked to species feeding activities. As such, species feeding habitat represents habitat within the species range more likely to be occupied by the species and reduces the risk of overestimation through the inclusion of unsuitable habitat while also being broadly applicable and easily updated. For many taxa, the available area of nesting/breeding habitat (ABH) may significantly affect useable species habitat area through a second set of vegetative requirements potentially different to AFH, which may also be seasonal. For such species, we suggest potential synergies in using AFH and ABH in tandem, though for this study we have focussed on the utility of AFH.

In the present study, we calculated AFH for Australian terrestrial bird species. As part of this, we measured habitat modification by vegetation class for Australian bird species between pre-European colonization estimates of AFH, current estimates of AFH, and a future estimate of AFH based on a projected anthropogenic land-use scenario. Using these measures, we aimed to (a) identify species that have experienced substantial habitat modification or are predicted to experience substantial habitat modification in the future, (b) identify the differential threats of urbanization and increasing agricultural land use on species (i.e., whether these land-uses drive habitat modification for a different set of species or habitats), and (c) explore habitat vegetation classes as an indicator of species habitat modification. We hypothesized that: (1) species associated with more “coastal” vegetation groups have experienced the greatest current and future habitat modification attributable to anthropogenic land use compared to other species due to the longer history of European colonisation, subsequent land cover conversion, and the large (growing) human populations in these areas (Australian Bureau of Statistics 2020, 2022), (2) different suites of species (and habitat groups) would be associated with habitat modification driven by urbanization and agriculture, and (3) species with small ranges and restricted feeding habitats would be especially vulnerable to proportionally high habitat modification (Harris and Pimm 2008).

Methods

Study region

Australia is a large continent with diverse ecoregions strongly associated with geography. The northern section of the continent is typified by tropical and subtropical conditions, while the southern section is largely temperate and Mediterranean. The centre of the continent is mostly desert and shrublands (Australian Government 2021). Forests cover 17% of Australia’s land area and are located primarily in areas of high rainfall such as coastal regions (Australian Bureau Of Agricultural Resource Economics And Sciences (ABARES), 2019). Agriculture represents the largest land-use category, and continues to increase in area (ABARES 2024). Many of Australia’s largest cities are also growing faster than other cities internationally (Hill et al. 2021; Oke 1982), though urban and residential environments represent a small proportion of land area in Australia (< 0.5%) (Australian Collaborative Land Use and Management Program (ACLUMP), 2017).

Estimating historical and present species’ area of feeding habitat

Species’ geographic range data were sourced from the BirdLife International Birds of the World Database (BirdLife International and Handbook of the Birds of the World 2016), which are provided as range polygons. Species range data were restricted to core habitat range (areas where the species is extant or probably extant, excluding areas of potential habitat where the species is now extinct or where there are no records of the species) within Australia. The areas of these range polygons were calculated to provide a baseline for comparison with calculated AFH. All spatial data handling and analyses were conducted in R 4.1.1 (R Core Team 2021) using the terra package (Hijmans 2022). Three datasets were collated for the purposes of calculating feeding habitat area: (1) the Australian National Vegetation Information System (NVIS) major vegetation groups (MVG) maps (Version 6.0), which show the spatial distribution of vegetation groups based on extensive Governmental vegetation data (Department of Climate Change, Energy, the Environment, & Water, 2021b), and are considered to be the best available national-scale vegetation data for Australia (Simmonds et al. 2019): (2) a database of Australian bird species characteristics data (Garnett et al. 2015): and (3) Birdlife International species range data. The NVIS maps include a map of contemporary vegetation (2020 – hereafter referred to as “current”) and estimates of pre-1750 vegetation (i.e., before European colonization and subsequent habitat modification, land clearing, and urbanization (Department of Climate Change, Energy, the Environment, & Water, 2021b), hereafter referred to as historical).

The NVIS categorizes land into 32 different vegetation groups (e.g., forests, woodlands, shrublands) with 1-ha resolution, and includes categories for unvegetated or contemporarily human-modified landscapes (excluding indigenous stewardship, as much of Australia’s vegetation has been human-modified for many millennia) (Department of Climate Change, Energy, the Environment, & Water, 2021b). Species feeding habitat associations classified by MVG were available from a database of Australian bird characteristics (Garnett et al. 2015). In some cases, feeding habitat associations in this database were not explicitly linked to an MVG, or vice versa, due to changes in the MVGs since the establishment of the database. In these cases, feeding habitats in the database were assigned to the most appropriate MVG based on the descriptions of the dominant vegetation species in the feeding habitats (see Supplementary 1 for a list of MVG – Feeding Habitat Associations).

As per Simmonds et al. (2019), the BirdLife International species geographic ranges were used to clip the extent of both the contemporary and historical MVG layers for each terrestrial species with at least one feeding habitat association encompassed by a terrestrial, vegetated MVG. This was then further clipped to include only the usable terrestrial feeding habitat for each species (by removing any area associated with MVGs not used by the species) that remains unmodified anthropogenically in a contemporary context. The NVIS category “cleared, non-native vegetation, buildings” was not included in AFH estimations, as this class represents contemporarily unmodified habitat, though we recognise that this may be usable feeding habitat for some species. The area (km2) of current and historical suitable feeding habitat within each species’ range was calculated for MVG habitats individually and totalled across all MVGs. The area of contemporary feeding habitat that occurred within a protected area from the Collaborative Australian Protected Areas Database (Department of Climate Change, Energy, the Environment, & Water, 2021a) was calculated for each species, as was the area of contemporary species feeding habitat within 50 km of the coast (selected as 87% of the human population lives in this region (Australian Bureau of Statistics 2020). These calculations were also repeated for the Australia-wide MVGs to get the total area of each group in each period.

Current and future anthropogenic land-use

Current and future areas of urban and agricultural land use were estimated using land-use harmonization (LUH) scenarios (Hurtt et al. 2020) based on shared socio-economic pathways (SSP) (Popp et al. 2017). These scenarios estimate future land-use based on projections of human use of resources and the effect of this on Earth systems. The LUH includes the proportion of land in a 0.25 × 0.25 arc-degree cell within anthropogenic land-use classes, allowing for attribution of habitat loss in the current scenario (which could not be distinguished using the NVIS data), and provides indications of potential future habitat modification. We opted to use the LUH associated with SSP2, which is described as the “middle of the road” scenario and represents a future with intermediate challenges driven by human population growth and carbon emissions (Popp et al. 2017) as it provides a moderate prediction of future anthropogenic land use. We summarized the land-use classes into two broad groups for this analysis; urban and agricultural (including C3 crops, C4 crops, and managed pastureland) for the years 2015 (which was the base year for these data) and 2100. These rasters were cropped to the extent of Australia, resampled, and then clipped by the species feeding association layers, and the total area (km2) encompassed by each land-use category in each period was calculated. These calculations were also repeated for each MVG (the entire area of the MVG Australia-wide) to get a future projection of loss for each MVG.

Analysis of trends

Proportional habitat metrics were calculated (Table 1) to represent the historical, current, and future habitat conditions. The proportion of species’ historical ranges represented by each MVG were also calculated. The MVGs were grouped into broader vegetation categories for further analysis, hereafter referred to as macro MVGs (see Supplementary 1 for grou**s). Principal component analysis was used to explore the relationship between variables and the clustering of species among these variables with package FactoMineR (Husson, 2020) in R. Beta generalized linear models were then used to determine which variables significantly predicted the proportion of species habitat lost between time periods for six different scenarios (Table 1). To minimize the risk of overfitting, variables were systematically removed from each global model to minimize the Akaike Information Criterion (AIC) value. Models within 2 AIC of the minimum AIC were retained, and the model estimates were averaged using the MuMIn package (Barton 2020) in R.

Table 1 List of proportional variables calculated, the associated terms used and whether the variable was used as a response variable in a model. For response variables, the independent variables included in the corresponding global model are indicated
Full size table

Results

Bird species results

A total of 467 species were included in the final dataset. Most (95%) species (n = 442) experienced habitat modification between historical and current estimates. About a third of species (151) had less than 50% of their historical range unmodified in the current conditions, while in the future scenario this number grows to 176. Based on the current vegetation maps, 39 species have an AFH less than 5,000 km2, 18 species have less than 1,000 km2, and four have less than 100 km2. In the future scenario, using the same thresholds these values are updated to 43, 19, and six respectively. Of the species identified, many of them had very small ranges historically, and most species lost much of their unmodified habitat between the historical and current period, and are predicted to experience further habitat modification between the current and future period (Fig. 1). Several of these species had very small proportions of their AFH protected. For example, the black grasswren (Amytornis housei) had a current and future AFH of 24 km² but had only 14% of their range protected, and the Kalkadoon grasswren (Amytornis ballarae), which has a predicted future AFH of 771 km², but currently only ~ 1% of their AFH within protected areas. Overall, most species had less than 50% of their AFH protected under the current scenario, with a median percent of habitat area protected of 27.3%, and a right-skewed distribution.

Fig. 1
figure 1

Area of unmodified feeding habitat (km2) for endemic species with less than 5,000 km2 remaining in the future scenario. The full column represents the historical habitat area for the species, and modification between scenarios is indicated

Full size image

The species with the largest conversion of habitat overall between the historical and future estimates was the long-billed corella (Cacatua tenuirostris), having a total of 95% of the potential feeding habitat converted to anthropogenic uses. Most (n = 6) of the species that appeared in the top 10 for habitat modification between the historical to the current period (see Supplementary 2) also appeared in the top 10 for habitat loss from historical to future periods (Table 2). The noisy scrub-bird (Atrichornis clamosus), a species with a small AFH in both the historical and current scenarios, had the greatest proportional habitat modification between the historical and current periods, having lost a total of 93% of potential unmodified feeding habitat between the historical and current range. Of the top 10 species with the greatest proportional habitat modification, most (n = 8) were predominately eucalypt-reliant (Table 2). Both the noisy scrub-bird and long-billed corella’s feeding habitats were predominated by Eucalypt MVG, and agriculture and urbanization both contributed to the loss of unmodified habitat for these species.

Table 2 Summary of AFH (km2) for the 10 species with the largest historical-future decline in potential unmodified habitat range. Percentage change from the historical baseline to the current period is represented in parentheses in the Current and Future columns. Coastal range (and percentage of historical in parentheses), and the change due to agriculture and urban between C-F are also listed. Current IUCN Red List listings are also provided for each species. Percentage of species historical range in applicable vegetation groups is also represented (R = Rainforest, E = Eucalypt, A = Acacia, Cas = Casuarina, M = Mixed, Mal = Mallee, G = grassland). Species represented in bold did not appear in the top 10 species between historical-current
Full size table

Vegetation groups results

At the macro MVG level, Rainforests and Mallee groups experienced the largest decline between historical and current scenarios (64% and 46% lost respectively), and the largest overall changes between historical and future scenarios (66% and 56% respectively) (Supplementary 3). Callitris and Mallee macro MVGs were predicted to have the most substantial decline in proportional habitat under the future scenario, losing an additional 19% and 17% area respectively. At the MVG level, “Mallee Open Woodlands and Sparse Mallee Shrublands” and “Rainforest and Vine Thickets” had the largest declines between the historical and the current periods, with declines of 71% and 64% respectively, and these continued to be the groups with the largest loss under the future scenario (see Supplementary 4). Eucalypt vegetation groups experienced the largest absolute loss of area, primarily from the historical to the current period (Fig. 2). All macro MVGs loss was primarily caused by agriculture in both comparative periods. Urban land use represented between 0% and 1.5% of all areas modified for macro MVGs between the historical and current periods. Conversely, modification attributed to urbanization was proportionally higher between the current and the future scenarios, with Rainforest and Eucalypt macro MVGs having the highest proportions at 16% and 10% respectively. Of the declining MVGs, those with larger percentages within 50 km of the coast experienced larger declines (Fig. 2).

Fig. 2
figure 2

Representation of vegetation coverage area (km2) for the macro MVGs between periods, with the percentage of future modification attributed to urbanization or agriculture for C-F (H-C plot not included as all groups had ~%1 lost area attributable to urbanization)

Full size image

Predictors of habitat loss

Species with large historical AFH maintained a large AFH through current and future conditions, as indicated by the clustering of those variables in the PCA (Fig. 3). There was also a close positive association between range size and proportion of species range in Acacia MVGs. The proportion of species habitat modified between the historical and current ranges is clustered with the proportion of historical area modified by urbanization, and the proportion of current range lost to urbanization in the future scenario. This is also clustered with the proportion of range in Eucalypt MVGs. The proportion of species habitat lost between current and future scenarios were more closely associated with the proportion of species habitat converted to agriculture. The proportion of habitat in the rainforest MVG was closely associated with species with the proportion of habitat within coastal areas.

Fig. 3
figure 3

Principal component analysis exploring the relationship between habitat values across species (n = 442). Variables that are positively correlated point the same direction on the plot, while those negatively correlated point opposite. The colour (contrib) indicates the contributions of the variables (%) to the principal components. Habitat values are proportional relative to historical habitat range. H-C = Historical to Current proportional change. C-F = Current to Future proportional change

Full size image

For the generalised linear models, species with increases between historical and current scenarios were excluded. For the agricultural and urban models, species were also excluded if they experienced no change in habitat area attributable to that source in the reference period (species counts post exclusions: Agri H-C n = 442, Urban H-C n = 439, Agri C-F n = 442, Urban C-F n = 441). The GLMs showed a positive association between species proportional association with Eucalypt and Casuarina macro MVGs between the historical and current periods, spanning across urban and agricultural modification (Table 3). Mallee habitat proportion was also associated with high habitat modification both current and future stemming from agricultural land use (not urban). Species with a high proportion of Melaleuca habitat were less likely to experience habitat modification across all models. The proportion of coastal area was positively associated with the proportion of habitat modified in all the historical to current loss models. This did not have a significant effect on habitat modified between current and future scenarios overall, or the agriculture-driven modification of habitat (Table 4). The proportion of habitat modification to urbanization (historical to current) was positively associated with future habitat modification overall and for both future urban and agricultural modification. Likewise, the proportion of habitat conversion to agriculture (historical to current) was also positively associated with future habitat modification overall, and for future urban and agricultural habitat conversion.

Table 3 Generalized linear model results for the macro MVG variables as predictors of species level proportional unmodified habitat loss, with rows represent the response variable in the GLM (which are all values representing proportional decline) and columns as the macro MVG variables (i.e., the proportion of a species’ habitat consists of each macro MVG) input into the model*
Full size table
Table 4 Generalized linear model results for the non-MVG variables as predictors of species level proportional unmodified habitat loss, with rows as the response variable in the GLM (which are all values representing proportional decline) and columns the macro MVG variables input into the model*
Full size table

Discussion

This study examined changes in Australian bird species area of feeding habitat (AFH) across three periods, historical (pre-1750), current (2015), and future (2100) scenarios. The source of habitat changes were quantified, whether from agricultural or urban development, and species’ AFH was analysed by major vegetation groups (MVG). We found that the species experiencing the largest losses of unmodified habitat between the historical and current AFH are likely to continue experiencing the largest losses of unmodified feeding habitat into the future. We also found that urban-related habitat modification, agriculture-related habitat modification, and overall habitat modification was positively associated with each other. Many species, including common species, have had substantial declines in available unmodified habitat that is likely to continue into the future, which may result in an extinction debt to be paid for these species. Our findings highlight the urgent need for proactive conservation efforts aimed at safeguarding the remaining unmodified habitat of species with little left, particularly those species that are not adapted to the complexities of human-modified landscapes, and those that are impacted by the intersection of multiple complex threats.

A key finding of this work for the conservation and management of Australian bird species was that species associations with Major Vegetation Groups (MVGs) were indicative of habitat modification. Specifically, species with large proportions of their habitat in Eucalypt, Rainforest, Mallee, Callitris, and Casuarina MVGs were more likely to have lost a greater proportion of their unmodified habitat. As predicted, species with large proportions of their habitat close to the coast were also more likely to have lost a higher proportion of unmodified habitat between the historical and current range. These results highlight the importance of ensuring thorough protection and increasing protected areas in these key MVGs, as they are crucial to preventing species extinctions stemming from excessive loss of habitat. While identifying key habitats for protection is an important step to securing conservation outcomes for birds, this is unlikely to be achieved entirely on public land as Kearney et al. (2022) highlight that many species remain distributed across private land. Furthermore, comprehensive representation of species would require a multifaceted approach that includes acquisition, co-management, and threat mitigation within and outside protected area boundaries (Ivanova and Cook 2020; Watson et al. 2011). The list of species results accompanying this study (see Supplementary 6) provides a numerical basis for species in need of additional protection through the clear delineation of remaining habitat, potential future loss, and percentage of habitat already protected, making this an important tool for conservation planning. The scale of the study does not lend itself to analysing individual species results in a substantive capacity in this manuscript.

For both agriculture and urbanization-related habitat modification between the historical and current scenarios, species with high proportions of Eucalypt MVGs in their habitat experienced significantly higher loss of unmodified habitat. Eucalypt forests are the most common type of forest in Australia, found everywhere except the driest areas of the continent (Australian Bureau Of Agricultural Resource & Sciences 2019). In what is described as one of the most substantial changes to vegetation in Australian history, European colonizers cleared large areas of eucalypt woodland upon their arrival due to their relatively fertile soils and their suitability for agriculture and the development of towns (Yates and Hobbs 1997). Eucalypt forests continued to be subject to considerable modification and clearing, to the extent that more than 80% of Australia’s eucalypt forests have been affected by humans (Resource Assessment Commission (1992) in Bradshaw (2012). As such, it is unsurprising that there was such a strong signal between Eucalypt MVGs and habitat loss. It is very clear that to protect species from excessive habitat loss, there must be more strategic protection of eucalypt habitat, particularly given that the future scenario indicates that these areas are likely to continue to be modified without intervention. Other MVGs have also been the focus of historical and ongoing land clearing, particularly in Queensland, that impacts biodiversity more broadly (Neldner et al. 2017; Reside et al. 2017). For example, Acacia woodlands (e.g. Brigalow – Acacia harpophylla) have been extensively cleared (Thornton and Elledge 2022), while rainforests are another area in need of strategic protection, particularly from agriculture, given that the range is so limited. Species with large proportions of melaleuca habitat had significantly less habitat loss. This is likely because much of Australia’s melaleuca forest is located in the northernmost areas of the continent (Australian Bureau of Agricultural Resource & Sciences 2016), where deforestation is limited (Deo 2011).

In this study, we excluded urban and modified land areas from habitat calculations for all species. This is certainly an oversimplification, as many species are known to have densities higher in human-modified environments or have had range expansions associated with spreading anthropogenic modification. The long-billed corella, the species with the greatest proportional reduction in area, is an example of this. Long-billed corella have adapted to using agricultural areas where their diet predominantly includes grains, sunflower seeds and corms of introduced weeds (Temby and Emison 1986), so it is likely that these values should be interpreted with that in mind. However, it is worth considering though that many of these anthropogenic environments are in flux, and so species dynamics may also be changing. For example, many species commonly seen in urban environments still depend on retained urban greenspaces, and have urban bird communities that are changing (Campbell et al. 2022) through pressures such as the declining extent of urban backyards and gardens (Hall 2010, 2015; Osborne et al. 2021). An example of this is the crested pigeon (Ocyphaps lophotes), which experienced a huge range increase coinciding with their ability to inhabit agricultural areas as European agriculture spread across Australia (Higgins and Davies 1996). Crested pigeon numbers have seen a subsequent decline in prevalence in recent years in some urban areas, likely due to the increasing intensity of anthropogenic land-use in these areas (Campbell et al. 2022). As such, while many urban, suburban, and agricultural areas may be somewhat suitable habitat for species now, that is not guaranteed to continue as Australian urban and agricultural areas grow and change to cope with the growing human population. Likewise, even though some species may persist in these areas, there may be less obvious physiological consequences for the species. For instance, recent research has shown that birds foraging in agricultural lands are exposed to, and experience varied effects from, neonicotinoid insecticides (Lennon et al. 2019, 2020), while those in urban environments may not be able to satisfy their nutritional requirements (Meillère et al. 2015). Coupled with the ongoing threat of habitat loss this may result in extinction debt in the long-term (Tilman et al. 1994; Hanski and Ovaskainen 2002).

The findings of this study are not intended as a prescriptive model of risk, but more as a representation of the potential magnitude of risks of current and future land-use change for species, especially those that have not had any habitat research done with a conservation focus. For species that are well known to be at risk, such as those in the Action Plan for Australian Birds 2020 (Garnett and Baker 2022), more detailed assessments of habitat availability exist. We also note that some species may have an apparently very small remaining habitat area; however, this is only representative of habitat not largely modified. This is not to say that these species will not persist under certain conditions in these modified habitats.

It is important to recognise that there will be some over and underestimation of feeding habitat stemming from the use the available data sources. Using major vegetation groups would have associated error, as the borders between vegetation groups are not absolute, rather there would be a continuum of intermediate vegetation mosaics. This means that species feeding habitat range will likely extend beyond or retract within the mapped areas in complex ways. It will likely be very important to update species distribution maps, such as those used in this study, to reflect species distributions more accurately, particularly as they change due to habitat loss and a changing climate. This is an interesting point for further consideration and further research.

We also note that we calculated pre-1750 habitat based on the current range maps, not from an estimation of pre-1750 species habitat. This was done as there is no comprehensive resource denoting the historical range of these species and means that the difference between historical and current estimates is likely to be underestimated. There may also be some circularity in the models presented, as species with small ranges are likely to have lost more habitat and are also more likely to have their range predominated by one macro MVG (and thus have a larger proportion in an MVG). We account for this by including habitat area baselines in all models.

There was some mismatch between data due to the use of two different sources of modified land use data (from NVIS and LUH). The current habitat area was based on NVIS classifications of vegetation, not solely on the LUH urban and agricultural data. Given that the NVIS map does not quantify the cause of MVG change (i.e., agriculture or urbanization), we opted to use the LUH 2015 baseline for attribution of change. As such, there is some unexplained habitat loss between the historical and current species’ habitat that may stem from a transition between vegetated MVGs, anthropogenic modification, or the differing resolution and scope (global versus national) of the datasets. We believe that these estimates are still fit for purpose, as they are likely just a conservative estimate of these land uses, especially given that there is a 5-year difference between the data. Likewise, the use of the SSP2 “middle of the road” scenario is also potentially a conservative estimate of future habitat modification, and there may be value in exploring the other potential pathways.

This work emphasizes the importance of a multifaceted approach to species conservation, ensuring that we have robust and widely applicable metrics to assess species extinction risk. The use of species area of feeding habitat (AFH) allowed for in-depth analysis of changes in species’ geographic ranges, providing insights into at risk Australian bird species. Our results indicate that the habitat of species that are running out of unmodified habitat must be actively conserved; there is no substitute due to the complex interplay of environmental changes that are typical in human-modified landscapes. However, we have also highlighted that many species have almost no unmodified habitat left, so we need to work toward improving the capacity of anthropogenically modified landscapes to maintain (or regain) local biodiversity and habitat connectivity. Options for such interventions are being explored, including large-scale urban greening projects, with a focus on diverse floristic structures representing local species to provide interstitial greenspaces (Shanahan et al. 2011; Snep et al. 2016; Sushinsky et al. 2013; Threlfall et al. 2016), and smart urban planning to decrease urban sprawl, which is a major challenge in Australia (Jim 2004; Sushinsky et al. 2013). We identified that species associated with particular vegetation types, such as Eucalypt and Mallee habitats, are more vulnerable to habitat loss now and into the future. This study provides a comprehensive list of terrestrial bird habitat loss at the species level as of now and into the future and indicates the amount of species habitat area protected, making it a key tool for protected area planning in the future.