Impacts of climate change on high priority fruit fly species in Australia

Tephritid fruit flies are among the most destructive horticultural pests posing risks to Australia’s multi-billion-dollar horticulture industry. Currently, there are 11 pest fruit fly species of economic concern in Australia. Of these, nine are native to this continent (Bactrocera aquilonis, B. bryoniae, B. halfordiae, B. jarvisi, B. kraussi, B. musae, B. neohumeralis, B. tryoni and Zeugodacus cucumis), while B. frauenfeldi and Ceratitis capitata are introduced. To varying degrees these species are costly to Australia’s horticulture through in-farm management, monitoring to demonstrate pest freedom, quarantine and trade restrictions, and crop losses. Here, we used a common species distribution model, Maxent, to assess climate suitability for these 11 species under baseline (1960–1990) and future climate scenarios for Australia. Projections indicate that the Wet Tropics is likely to be vulnerable to all 11 species until at least 2070, with the east coast of Australia also likely to remain vulnerable to multiple species. While the Cape York Peninsula and Northern Territory are projected to have suitable climate for numerous species, extrapolation to novel climates in these areas decreases confidence in model projections. The climate suitability of major horticulture areas currently in eastern Queensland, southern-central New South Wales and southern Victoria to these pests may increase as climate changes. By highlighting areas at risk of pest range expansion in the future our study may guide Australia’s horticulture industry in developing effective monitoring and management strategies.


Introduction
Tephritid fruit flies are one of the most destructive and economically significant pest insect families, attacking a wide range of fruit and vegetables. While the family contains more than 4000 species, around 350 are recognized as economically important horticultural pests [1] that have significant impacts on global horticultural production and market access. In Australia, the average annual value of crops susceptible to fruit flies is~$4.8 billion [1], and the National Fruit Fly Strategy has identified 46 species as 'high priority pests' [2] of concern. The majority of these species are exotic to Australia, primarily found in South-East Asia and the South PLOS ONE | https://doi.org/10.1371/journal.pone.0213820 February 13, 2020 1 / 19 a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 fruit fly species, based on their performance along climatic gradients. While highly useful in furthering our understanding of climate impacts on fruit flies, these studies have either focused on other countries or have explored global patterns of the distribution of suitable climate [22][23][24][25][26]. Here we assess how climate change may result in shifts to the distribution of climatically suitable habitat for the 11 high priority fruit fly species present in Australia, using the correlative SDM, Maxent [27]. This SDM has been used extensively to assess the distribution of suitable habitat for a broad range of pests and invasive species [26,[28][29][30][31]. We also highlight areas at risk of pest range shifts, to guide Australia's horticulture industries in developing effective monitoring and management strategies.

Species occurrence data
We collected occurrence data for the 11 species from five sources: the Australian Plant Pest Database (APPD; http://www.planthealthaustralia.com.au/resources/australian-plant-pest-database, accessed 15th March 2017), the Atlas of Living Australia (ALA; http://www.ala.org.au, 22nd December, 2016), the Global Biodiversity Information Facility (GBIF, https://www.gbif.org, 28th June, 2017, see S1 File for DOIs), trap data, and existing literature. APPD is a national digital Chilli, tomato [3] Bactrocera halfordiae (Tryon) Halfordia fruit fly North Queensland to the Sydney region in NSW Citrus [3] Bactrocera jarvisi (Tryon) Jarvis' fruit fly Northwest Western Australia, NT, northwest Queensland, eastern Australia from Cape York to Sydney, NSW Mango, peach, banana, pear, apple, pawpaw, persimmon [3] Bactrocera kraussi (Hardy) Krauss' fruit fly Torres Strait Islands, northeast Queensland as far south as Townsville Grapefruit, mandarin, orange, mango, peach and banana [3,7] Bactrocera musae (Tryon) Banana fruit fly Torres Strait Islands, northeast Queensland as far south as Townsville Banana [3,7] Bactrocera neohumeralis (Tryon) database of plant pest and pathogen specimens held within herbaria and insect collections across Australia. It is a powerful tool for market access and emergency responses to pest incursion, and supports associated research activities. ALA is Australia's largest digital database of species occurrence records, containing information from a wide array of data providers including Australia's major museums and government departments. GBIF provides similar data at a global scale. Before downloading data from APPD, ALA and GBIF, we applied filters to restrict records to those that were resolved to species-level, were dated no earlier than 1 January 1950, contained valid geographic coordinates, and were not flagged as 'environmental outliers'. We also collected trap data from various state government departments (Biosecurity and Food Safety, Department of Primary Industries, NSW; Biosecurity Queensland and the Queensland Department of Agriculture and Fisheries; Department of Economic Development, Jobs, Transport and Resources, Victoria; and Department of Primary Industries and Regions South Australia (PIRSA)). Trap data from these sources were collected at different periods from 1996 to 2017. Finally, we also obtained occurrence data from the literature [1-4, 6, 7, 11, 33-37].

Major commercial fruit and vegetable hosts
For each of the 11 fruit fly species, we compiled information on the major commercial hosts on which infestation has been recorded. For this purpose, we defined major fruit and vegetable host species according to the Australian Horticulture Statistics Handbook (HSHB; www. horticulture.com.au) for the year 2016/2017 [32]. This document consolidates horticulture statistics of interest to industry members and other stakeholders. The data contained in HSHB were derived from the Australian Bureau of Statistics, projects funded by Hort Innovation, international trade sources and horticulture industry representative bodies where available.

Climate data
For baseline and future climate conditions we used the bioclimatic variables available within the WorldClim database (v1.4), at a spatial resolution of 30 arc-seconds [38] (approximately 1 km; http://www.worldclim.org). These data, based on meteorological records for the period 1960-1990, comprise 19 climatic variables, 11 of which are temperature-based while eight relate to precipitation. Combined, the data represent annual trends, seasonality, and limiting or extreme environmental conditions. Assuming that host plants are available, temperature and moisture are the key factors influencing fruit fly reproduction and survival [18,39]. Thus, these variables were chosen as predictor candidates based on the fruit flies' biology and ecological requirements, and similar habitat suitability studies undertaken on other insects [40]. For each species, we identified a set of ecologically-relevant variables, with minimal collinearity, that resulted in high predictive power for the model [41] (described below).
When projecting future suitability, we considered a range of climate scenarios to acknowledge this important aspect of uncertainty. CSIRO recommends eight global climate models (GCMs) as being useful for Australian climate impact assessments [42]. Data from six of these models were available from the CCAFS GCM Data Portal (http://www.ccafs-climate.org/data_ spatial_downscaling/), at a spatial resolution of 30 arc seconds. These data were developed from anomalies of the original GCM data that were statistically downscaled using a thin plate spline spatial interpolation, and then applied to the WorldClim v1. and GFDL-ESM2M (Global Coupled Climate Carbon Earth System Model Part-1). CanESM2 projects a hot future with drying across central regions of Australia and higher precipitation in the north-east. ACCESS1.0 projects a hot and dry future across most areas of Australia, while MIROC5 projects moderate warming, with drying in the north-east and south-west but higher precipitation in central Australia. NorESM1-M projects moderate warming. HadGEM2-CC and GFDL-ESM2M project a hot future with greater warming typically in central regions.
We downloaded the 19 bioclimatic variables from these six models from CCAF, for 20-year periods centred on 2030, 2050 and 2070, for the Representative Concentration Pathway 8.5 (RCP8.5) [43]. With a radiative forcing exceeding 8.5 Wm -2 by 2100, this is the highest of the four RCPs presented in the Intergovernmental Panel on Climate Change's Fifth Assessment Report [43]. It is also the RCP that emissions are currently tracking most closely [44]. After downloading, we reprojected data to a spatial resolution of 1 × 1 km (Australian Albers Equal Area, EPSG: 3577) via bilinear interpolation, using the gdalwarp function provided by the R package gdalUtils [45] in R version 3.1.2 [46].

Species distribution models
We used the machine learning approach, Maxent (v3.3.3k [27]), to assess climate suitability for species under baseline and future climate scenarios. Maxent accommodates presence-only data and has performed well in multimodel assessments [47]. It produces a continuous probability surface, which can be interpreted as an index of climatic suitability given the predictor variables included in model calibration. Detailed descriptions of Maxent are given elsewhere [48,49]. We optimized models by assessing the effects of different combinations of feature types, of competing predictor sets deemed ecologically sensible a priori, and of the extent of regularization on model performance. We found that Maxent performed best when product (first-order interactions), linear and quadratic features were used, with a regularization multiplier of 1 (the default), and used this configuration to calibrate our final models.
Maxent requires background data, to which it compares the environmental characteristics of presence locations. There is flexibility for users to specify which points to use as background, as well as the number of records and the spatial extent from which they are chosen [48]. Following Ihlow et al [50], we generated background points by randomly selecting up to 100,000 cells from terrestrial areas within 200 km of occurrence records of the target species. Our choice of background achieves a balance between fine-scale discrimination of suitable and unsuitable sites along environmental gradients, and generalization of model predictions.
To assess model performance, we used five-fold cross-validation to reduce model errors that may occur from the random splitting of data into test and training subsets. The performance of each model was evaluated using the area under the receiver operating characteristic curve (AUC), which describes the consistency with which a model ranks randomly chosen presence sites as more suitable than randomly chosen background sites. AUC ranges from 0 to 1, with a value of 0.50 indicating discrimination ability no better than random, while values greater than 0.75 indicates that the model has a discriminative ability that is better than "fair" [51]. Cross-validated AUC scores were presumed to reflect the performance of a single final model for each species, which used all available data.
Following previous studies of pest species [26], continuous suitability scores projected by Maxent models were converted to binary layers (0 = unsuitable, 1 = suitable) using the 10th percentile training presence threshold (i.e. the value that corresponds to 10% training omission). We note that the selection of a threshold value may vary depending upon the goals of the study [52], thus we also provide continuous output for baseline climate as supplemental data (S1 File). For each species, the six binary suitability grids (i.e., one for each GCM, with cells assigned 0 when unsuitable and 1 when suitable) for each time period were summed to produce a consensus map, identifying agreement about the suitability of grid cells across the six climate scenarios. Each species' consensus map was then converting to a binary map indicating whether cells were projected to be suitable under the majority of GCMs (i.e., suitable in < 4 GCMs = 0, suitable in 4 or more = 1). The resulting binary maps were summed across species to identify hotspots-grid cells suitable for multiple pest species. Finally, we compared the distribution of hotspots to that of major horticultural crops.
When projecting models, extrapolation to conditions beyond the range of the training data may be unreliable. Following Elith et al. [53] we developed MESS (multivariate environmental similarity surface) maps to identify regions of extrapolation [53]. By revealing areas with novel environmental conditions, MESS maps can be used as a projection mask, highlighting regions for which less confidence can be placed in projections, or as a quantitative measure of prediction uncertainty [53]. We then recalculated the size of projected suitable climate with novel environments excluded.
All modelling and post-modelling analyses and calculation of statistics were performed in R version 3.1.2 [46]. We used the sp [54] and raster [55] packages for preparation and manipulation of spatial data, the dismo [56] package to fit Maxent models, and custom R code for rapid projection of fitted models.
Bactrocera aquilonis. Our model suggested that, during the baseline period, climatically suitable habitat for B. aquilonis existed in the northern regions of the Northern Territory and Western Australia, as well as northern Queensland where this fly has not been reported (S1 File). The variables with the highest permutation importance were precipitation of the wettest quarter (68.9%) and annual mean temperature (28.9%) (S1 File).
As the century progresses, the geographic extent of climatically suitable habitat for this species is projected to increase and expand southwards under all six scenarios, with many areas suitable in the baseline period projected to remain so until at least 2070 (S1 File). This includes northern Western Australia, much of the Northern Territory, and north-western Queensland (S1 File). We note, however, that climate scenarios beyond 2030 frequently contain novel conditions across the northern regions of Australian, highlighting uncertainty in Maxent projections within these areas (S1 File).
Key horticultural crops for B. aquilonis are Mangifera indica (mango), Citrus × paradisi (grapefruit), Malus domestica (apple), Prunus persica (peach) and Citrus sp. (citrus) (S1 File). The major regions where these crops are currently grown include the Northern Territory and north-east Western Australia. These regions may remain suitable for B. aquilonis until at least 2070. Similarly, fruit growing regions in the Wet Tropics (north-east Queensland) are likely to increase in suitability in the future. Other major host-plant growing regions in the south and east of the continent will likely remain unsuitable (S1 File).
Bactrocera bryoniae. Suitable habitat for B. bryoniae during the baseline period is projected to have occurred along the northern and eastern coastlines (S1 File). Temperature annual range and precipitation of the driest month contributed the most to the model for this species (42.2% and 27.4%, respectively) (S1 File).
By 2070, suitable habitat is projected to increase under all scenarios except GFDL-ESM2M (which projects a hot, very dry future) (S1 File), expanding to the southern coastlines of Victoria and Western Australia. Under 1-3 scenarios, suitable habitat is projected to shift inland in Queensland and NSW. However, the amount of habitat projected to be suitable under all six scenarios remains relatively stable from 2030-2070 (S1 File). Beyond 2030, novel conditions are primarily restricted to the north-western regions (S1 File).
The major horticultural host for B. bryoniae is Capsicum annuum (chilli) (S1 File). Our model indicates that key growing regions for this crop in Queensland contained suitable habitat for B. bryoniae during the baseline period, and this will continue to be the case until at least 2070 (S1 File).
Bactrocera frauenfeldi. Climatically suitable habitat for the baseline period is projected to have been mostly confined to Cape York Peninsula and the Wet Tropics, although small areas in northern Western Australia and the Northern Territory from which the species has not been recorded were also classified as suitable (S1 File). The most important variable in the model for B. frauenfeldi was precipitation of the wettest quarter (75.4%) (S1 File).
As the century progresses, suitable habitat is projected to expand under all scenarios except CanESM2 (S1 File). This scenario projects a hot, very dry future, leading to loss of suitable habitat in northern Queensland by 2050. However, the extent of suitable habitat for this species is likely to remain small, relative to other species. In addition, the far north-east of Queensland contains novel conditions, decreasing confidence that this area will be suitable as the century progresses. As with other species, the Wet Tropics is projected to remain suitable and is not a region in which the model is extrapolating.
The major crops for B. frauenfeldi are Mangifera indica (mango) and Carcica papaya (pawpaw) (S1 File). Major production regions in north-western Northern Territory may remain suitable for this species until at least 2070, although there is substantial uncertainty across the climate scenarios. In contrast, it is very likely that the Wet Tropics will remain suitable until at least 2070, irrespective of the climate scenario (S1 File).
Bactrocera halfordiae. Baseline climatically suitable habitat for B. halfordiae occurred in the Wet Tropics and subtropics, from north Queensland to eastern New South Wales (S1 File). Precipitation of the driest month (66.8%) and annual mean temperature (32.3%) contributed most to this model (S1 File).
The geographic extent of suitable habitat is projected to vary considerably across the six climate scenarios. As the century progresses, gains in new habitat may exceed losses under some scenarios (e.g. see ACCESS and MIROC5 in S1 File) while losses are projected under the CanESM2 scenario (which projects a hot future, drying across central regions and higher precipitation in the north-east), mostly due to contractions in the south and east. Across the scenarios there is consensus that lower elevation regions in the south-east will be suitable. Furthermore, MESS maps indicate little model extrapolation for this species (S1 File).
Crops in the Wet Tropics may continue to be at risk from this species, until at least 2070. However, only 1-2 scenarios project horticultural regions in southern Queensland to retain suitable climate (S1 File). Although horticultural regions along the NSW-Victorian border are projected to have been unsuitable for B. halfordiae in the baseline period, these areas may become suitable between 2050-2070 under some scenarios (S1 File).
Bactrocera jarvisi. Suitable habitat for this species is projected to have been mostly confined to northern Western Australia, the Top End of the Northern Territory, and eastern Australia from Cape York to NSW during the baseline period (S1 File). Annual mean temperature (38.0%) and precipitation of driest month (37.2%) had the highest contributions to the model for this species (S1 File).
There is substantial consensus across the six scenarios that regions currently suitable for B. jarvisi will remain so until at least 2070 (S1 File). In addition, across some models gains are projected to occur in central Queensland, Western Australia, and the Northern Territory, although model extrapolation occurs under several climate scenarios (S1 File).
Comparing the distribution of suitable habitat for this fly with that of its major host crops indicates that crops grown in the Top End of the Northern Territory, and in eastern Australia from Cape York to New South Wales, may continue to be at risk until at least 2070. Other major host-plant growing regions in the south and west of the continent will also remain suitable for this species until 2070 (S1 File).
Bactrocera kraussi. For the baseline period, suitable habitat for B. kraussi is projected to have occurred across the northern tip of Australia and northeast Queensland, as far south as Townsville (S1 File). Precipitation of the wettest quarter (75.19%) had the highest contribution to the model of B. kraussi (S1 File).
There is consensus across the six scenarios that the geographic extent of climatically suitable habitat may increase slightly (S1 File), although this is still confined to the Wet Tropics and far north of the continent. In addition, little extrapolation to novel conditions occurs (S1 File). Horticultural production regions in northeast Queensland may remain suitable for this species by 2070, although production regions in the south are likely to remain unsuitable (S1 File).
Bactrocera musae. For the baseline period, suitable habitat for B. musae extended from the Torres Strait Islands through to the Wet Tropics (S1 File). The most important variable in the model for B. musae was precipitation of the wettest quarter (78.7%) (S1 File).
Suitable habitat for this species is projected to remain restricted to the Wet Tropics and northern-most regions of the country under the climate scenarios. While there is consensus across the six climate scenarios, less confidence can be placed in projections to the north-west (S1 File).
B. musae mainly attacks Musa × paradisiaca (banana), the production areas for which are located primarily in tropical and subtropical regions of the continent (S1 File). The major commercial growing region in the Wet Tropics is projected to remain climatically suitable for this species until at least 2070 (S1 File). Bactrocera neohumeralis. Suitable habitat for this species during the baseline is projected to have been mostly confined to the Torres Strait Islands, eastern Queensland, and north eastern NSW south to Wollongong (S1 File). Precipitation of the wettest month (47.4%) contributed most to the model for B. neohumeralis (S1 File).
As the century progresses, considerable differences in suitable habitat are projected across the six scenarios. For example, under the CanESM2 scenario,~one quarter of current suitable habitat is projected to be lost by 2030, although by 2070, range expansions are projected to exceed losses (S1 File). Similarly, under the hot, very dry scenario simulated by GFDL-ESM2M, total range size may decline by 2030, mostly due to contractions in the south and east, although limited gains in habitat may occur in northern Australia (S1 File). There is consensus in projections of suitability across the north tips of the continent, however MESS maps indicate that Maxent has extrapolated suitability in these regions. In contrast, greater confidence can be placed in projections of consensus along the east coast (S1 File).
Production regions in eastern Queensland and north-eastern NSW will likely remain suitable for this species until at least 2070, although there is substantial uncertainty across the climate scenarios. In contrast, regions along the NSW-Victorian border and further south are projected to remain unsuitable for B. neohumeralis (S1 File). Bactrocera tryoni. Highly suitable habitat for B. tryoni is projected to have occurred in south-western Western Australia, south-eastern South Australia, Victoria, and eastern Australia from Cape York to NSW in the baseline period (S1 File). Coastal zones in northern Western Australia, the Northern Territory and the eastern half of Tasmania have moderate suitability (S1 File). Annual mean temperature (33.06%) and mean temperature of the coldest month (32.42%) had the highest contributions to the model for this species (S1 File).
The geographic extent of suitable habitat varies across the six climate scenarios. As the century progresses, gains in new habitat may exceed losses under some scenarios (e.g. see ACCESS1.0, MIROC5 and NorESM1-M; S1 File), while substantial declines occur under others (e.g. GFDL-ESM2M S1 File), mostly due to contractions in the south and east. Areas of consensus occur along the coastline, although less confidence can be placed in these projections for the Northern Territory and northern Western Australia due to model extrapolation (S1 File).
Key regions for host crops in the Top End of Northern Territory, eastern Australia from Cape York to NSW, Victoria, and some parts of Tasmania, may remain suitable for B. tryoni until at least 2070. Major host-plant growing regions in South Australia may also remain suitable for this species until 2070 (S1 File).
Ceratitis capitata. Our model suggests that suitable habitat for C. capitata for the baseline occurred throughout Western Australia, the Northern Territory, the east coast of Queensland to NSW and South Australia (S1 File). We note that scattered records within inland regions of Western Australia are projected as having had low suitability. Annual mean temperature (47.2%) and mean temperature of the coldest month (46.2%) contributed most to the model for this species (S1 File).
Under the future climate scenarios, the geographic extent of suitable habitat is projected to increase and expand inland (S1 File) with much of Victoria and Tasmania likely to be suitable. There is considerable consensus in the distribution of suitable habitat, although consensus declines in New South Wales as the time horizon increases (S1 File). As with other species, MESS maps indicate extrapolation across the northern regions under scenarios from 2050 onwards (S1 File). However, there is high consensus in suitability across the major host plant regions in Queensland, Victoria, and Western Australia.
Zeugodacus cucumis. For the baseline period, suitable habitat for Z. cucumis is projected along the northern region of Western Australia and the Northern Territory, north-east Queensland, and south along the east coast to NSW (S1 File). Precipitation of the driest quarter (54.3%) and mean temperature of the coldest quarter (36.2%) had the highest permutation importance in the model for this species (S1 File).
Under future climate scenarios, the geographic extent of suitable habitat may increase, expanding southward and inland, with most areas that are currently suitable projected to remain so until at least 2070 (S1 File). There is considerable variation among projections for inland regions, likely due to differences in precipitation patterns, indicating higher uncertainty about the future suitability of these regions. There is high consensus in suitability along the east coast, and while consensus is also high in the north MESS maps identify this as a region of extrapolation. There is little agreement on the suitability of inland regions of New South Wales and Queensland (S1 File).
Major commercial growing regions for host crops in Queensland and the Northern Territory are projected to remain climatically suitable for this species until at least 2070 (S1 File). Other major host-plant growing regions in the south and west of the continent will likely remain unsuitable under the time periods considered in this study (S1 File).

Future hotspots of pest fruit flies
For each time period, we stacked climate suitability maps for all species, to identify regions most likely to contain suitable climate conditions for multiple pest species (i.e. hotspots). As the century progresses, the geographic extent of climatically suitable habitat for most of the 11 species is projected to expand and shift south regardless of whether novel environments are included or excluded (Figs 1 and 2 and Table 2). When regions containing novel climate are included, 31.6% of Australia (i.e.~2,400,800 km 2 ) is projected to have been suitable in the baseline period for at least one of the 11 species, increasing to more than half of the continent by 2070 (Table 2). However, only Queensland's Wet Tropics is likely to be suitable for all 11 species into the future.
When novel environments are excluded from maps, less than 30% of Australia is projected to be suitable for at least one of the species by 2070 (Table 2). Hence, exclusion of novel environments substantially impacts the size of suitable habitat (i.e., projections of suitable habitat frequently occur in areas with novel climatic conditions). However, extrapolation primarily occurs in northern regions of Western Australia, Northern Territory and the Cape York Peninsula, decreasing confidence in projections across these regions. From the Wet Tropics and southward, little extrapolation occurs. As such, the Wet Tropics bioregion is projected to remain suitable for 10-11 species, indicating that the major commercial host plants within this bioregion may continue to be at risk of invasion by most or all of these high priority species.
Major commercial host plant regions along the coastal strip of south-east Queensland and north-east NSW are likely to have areas that are suitable under all future scenarios for B. bryoniae, B. jarvisi, C. capitata and Z. cucumis (S1 File). Under some scenarios, these regions may also be suitable for B. halfordiae, B. neohumeralis and B. tryoni (S1 File). Some major commercial host plant regions in southern NSW and Victoria are also projected to be suitable for B. jarvisi, B. tryoni and C. capitata under all scenarios (S1 File) and for B. halfordiae, B. neohumeralis and Z. cucumis under a limited number of scenarios (S1 File). Horticultural regions in Tasmania are projected as suitable for B. jarvisi, B. tryoni and C. capitata (S1 File).
In south-west Western Australia, major horticulture regions are likely to remain suitable for B. jarvisi, B. tryoni and C. capitata, although the latter species is currently not found in this region (S1 File). Commercial horticulture regions in northern region of the Northern Territory are also likely to be suitable for B. jarvisi, B. kraussi, B. musae, B. tryoni and Z. cucumis under all scenarios, and B. frauenfeldi under some climate scenarios.

Discussion
Our study suggests that the Wet Tropics bioregion has climatically suitable habitat for the largest number of high priorities tephritid pest species under both baseline conditions  and as a result of climate changes projected to occur through to 2070. Cape York Peninsula and the Northern Territory are also likely to be vulnerable, although novel climates are projected to occur in these regions, and the extrapolation of SDMs to these conditions may be unreliable. The east coast of Australia is also likely to remain suitable for multiple species until at least 2070. As such, major horticulture regions in north-western Australia, the Northern Territory, southern-central regions of NSW, southern Victoria and north Tasmania may become increasingly suitable to high priority fruit flies. Two species, B. tryoni (Qfly) and C. capitata (Medfly), are projected to have suitable conditions in all states and territories of Australia, under all considered climate change scenarios, until at least 2070.
Over the past 30 years, numerous studies have modelled suitable habitat for Qfly using CLI-MEX and Maxent, at various spatial resolutions [8,40,58,59] and extents. While generally giving similar projections, a key difference is that our model projects Tasmania to have been suitable in the baseline period , whereas fine scale modelling using CLIMEX indicates that it is unlikely to become suitable prior the mid-21 st century [59].
Our models for both Qfly and Medfly were driven primarily by temperature parameters, rather than precipitation. Previous studies have identified climatic constraints on the distribution of Qfly. For example, it has been reported that Qfly pupae do not survive in the winter months in Melbourne and near Sydney [60], and adults fail to emerge later than mid-April [61]. Further, many subtropical sites in Queensland are marginal in winter for Qfly breeding and general activity [8,61]. As such, slight temperature increases associated with climate change are projected to substantially elevate the threat that this species poses to horticultural industries [62]. For instance, using data from the late 1990s, it was estimated that annual control costs for apple growers around Adelaide may increase by between $346,000 and $1.3 million with a 0.5-2˚C increase in temperature [62].
With the exception of Western Australia, all Australian states and territories are currently free from Medfly, with market access protocols inhibiting movement into other states [21], and incursions met with immediate eradication programs [13]. Our model of baseline habitat indicates suitable conditions for Medfly around most of Australia's coastal regions. In addition to identifying suitability in the subtropical coastal fringe of Queensland, our model suggested that much of the low-altitude regions in the south-east, including parts of Tasmania, are also Hotspot maps of baseline ) and future habitat suitability for 11 fruit flies. Suitability was modelled with Maxent, and thresholded using the 10th percentile of suitability at training presence localities. These maps include projections under novel environments (see S1 File for individual species maps with novel environments included). Colours indicate the number of species for which habitat is projected to be suitable under the majority (� 4) future climate scenarios. Figure was created in R version 3.3.3 [57] (https://www.R-project.org/). https://doi.org/10.1371/journal.pone.0213820.g001 Impacts of climate change on high priority fruit fly species in Australia suitable. While our result for Medfly is similar to that obtained from CLIMEX and Principle Components Analysis [25,58], the model GARP projected a far broader potential distribution in Australia for this species [58], while Szyniszewska and Tatem's [63] Maxent model suggested more of Queensland may be suitable, compared to our analysis. Competition with Qfly may be responsible for exclusion of Medfly from much of Queensland [25], and similar biotic interactions may suppress the species elsewhere [13]. However, Medfly may be more tolerant to low temperatures and dry summers than Qfly [4], rendering Medfly the stronger competitor in areas with these conditions. Medfly was recorded in Tasmania in the 1920s but reportedly failed to survive an unseasonably hot and dry summer [4]. Due to their age, these records were not used to calibrate our model, yet our projections indicate that Tasmania continues to have conditions suitable for this species.
Allwood and Angeles [64] reported that B. jarvisi is recognized as a pest in north-western Australia, infesting mango, guava and pomegranates (as reported in [65]). Dominiak and Impacts of climate change on high priority fruit fly species in Australia Worsley [9] concluded that the current south-eastern range limit lies north of the Queensland-NSW border (~25.5˚south), while the south-western limit lies at approximately 18˚south.
However, previous analysis suggested that this species' current climatic range could extend into the cooler temperate areas of southern NSW, and eastern and northern Victoria [4]. Our models partly agree, indicating that suitable conditions along the east coast of Victoria in the baseline period. This species can also withstand very warm conditions, with eggs known to be more heat tolerant than those of the sympatric Qfly, surviving temperatures of 48.2˚C [65]. Given that these species infest many of the same hosts, competition is likely, hence eradication of Qfly may result in the competitive release of B. jarvisi, increasing the threat it poses to horticulture [4,65]. Further, as the cultivation of B. jarvisi host plants expands geographically, this species may increase in abundance and extend its range, potentially becoming a major pest in north-western Australia [6,36]. However, across north-western Australia, and to a lesser extent the far north-east, models for most species were projected onto novel conditions, decreasing confidence in suitability estimates for these regions. In contrast, MESS maps demonstrated that extrapolation rarely occurred across eastern and southern regions, although novel interactions between climate variables cannot be ruled out.
While widespread throughout Queensland, Z. cucumis currently has a restricted distribution in the Northern Territory, although there is a disputed single record from northern Western Australia [66]. Both Fitt [67] and the Horticultural Policy Council [4] reported that if the cucurbit industry expands in the Northern Territory, the pest status of Z. cucumis may increase. However, while the species has been trapped frequently in the Northern Territory, it has not been found on cucurbits growing in this region [6]. In NSW, Z. cucumis appears to be currently limited to regions close to the Queensland border, with rare detection as far south as Sydney [66] . It has not been detected in the (former) Fruit Fly Exclusion Zone in southern NSW [37]. Our model also estimates the southern limit of suitable climate for this species to be around Sydney. However, with climate change this may extend further southward, with parts  [3,7,37]. Our model suggests that as climate changes, the range of this species may extend southward and, under some scenarios, into parts of Victoria. Previous climatic analysis also suggested that this species is well adapted to conditions on the east coast of Queensland, with large populations occurring in areas north of Townsville [4]. Similar ecological characteristics are shared by B. neohumeralis and Qfly [68], yet while Qfly is prevalent in sub-tropical and temperate areas of Queensland and NSW, B. neohumeralis is more prevalent in northern wet tropical areas [4,5,69]. The reason for this difference between the geographical ranges of these species is unclear, as both are polyphagous and use similar host fruits for their larval development [68,69].
Our model for B. aquilonis indicates suitable conditions for this species in northern Queensland under baseline conditions, although it is presently only known from north-western Australia [5]. The hosts of this species now include 40 commercial crops [6]. Expansion of the range of this species, or the growth of host plant industries in north-western Australia may necessitate the development of new monitoring, control and disinfestation procedures [65]. In addition, it has been argued that if B. aquilonis hybridises with Qfly, the resulting strain may have greater potential for spread than B. aquilonis [4]. This, in turn, would require that disinfestation procedures be developed for the hybrids [65].
The distribution of B. bryoniae ranges from the Torres Strait Islands, across northern Australia, and along the east coast to north of Sydney, NSW. Our results indicate that suitable climate may exist in Victoria, south of the species' known range. However, previous studies have demonstrated that populations in northern NSW experience a marked decline in abundance through November-January [37]. This may be explained by a decline in the fruiting and flowering of native host trees, or seasonal climatic constraints that are not reflected in our model [37], which may also explain their absence in Victoria.
Northern Queensland has the highest diversity of fruit flies in Australia, and some species with significant economic impacts are found only in this region [7]. The distribution of B. kraussi, B. musae and B. frauenfeldi is limited to north Queensland [3,70], with recent trap data suggesting that these species do not occur south of Townsville [7]. Royer et al. [35] predicted that B. frauenfeldi has suitable habitat in the Northern Territory and northern Western Australia, which is also suggested by our model. This species has expanded its range in northern Queensland due to continued planting of hosts, such as mango and guava [35]. Further increases within these horticulture industries in northern Queensland may increase the pest status of this fly [70].

Model errors and uncertainties
SDMs are useful for developing a broad understanding of how the distribution of suitable habitat may be influenced by climate change. However, the output of SDMs is known to be influenced by characteristics of the occurrence sample, including its size [71], sampling bias [72], and spatial autocorrelation [73], as well as the extent of the study area, selection of predictor variables [74], and selection of background points [75]. We addressed these issues by: (1) exploring alternate settings in Maxent to optimise models and reduce overfitting that may generate unreliable estimates [48]; (2) reducing the number of predictor variables by assessing collinearity; and (3) critically examining response curves.
In addition, we acknowledge that the selection of a threshold for converting Maxent's continuous output into binary data (typically defined as distinguishing between "suitable" and "unsuitable" conditions) can be subjective. A region classified as unsuitable may not be free of the pest; rather, these areas are considered less likely to support a population compared with regions above the threshold. In reality, the choice of threshold is based upon a comparison of the importance of false positives and false negatives [76]. For invasive species, the latter may be more serious because it can result in an underestimate of the geographic extent of suitable conditions, and hence, invasion risk [77]. This, in turn, can lead to poor decision-making and failure to establish appropriate surveillance or containment measures. As such, in this context a precautionary approach to defining a threshold, as undertaken in the present study, is warranted. However, since overprediction of suitable habitat can also prove problematic (potentially leading to ineffective allocation of monitoring resources), we provide maps of continuous (unthresholded) suitability (S1 File), permitting stakeholders to modify this threshold according to their objectives.
Sampling bias is another challenge faced when fitting correlative SDMs, particularly when incorporating data from sources of incidental observations such as museums and natural history collections [78]. As such, it is difficult to determine whether a species is observed in a particular environment because of habitat preferences or because that region has received the largest search effort [75,78]. For presence-background approaches to habitat modelling, a target-group background sampling strategy goes some way to handling biased occurrence samples [79]. However, while imposing environmental bias on the background counteracts similar bias in the occurrence sample, this strategy may increase the extent of novel environments to which the model must be extrapolated.
While SDMs consider sensitivity to climate change, species responses may also include microevolution [80] or plasticity [81]. As accessibility to genomic data increases, and experiments on plasticity are conducted, SDM output can be refined [82]. In addition, as mean conditions change, so too will the distribution and magnitude of extremes. Presently, there has been little work undertaken to assess how different fruit fly pest species tolerate extreme weather events such as heatwaves and moisture stress, or their potential for adaptation to climate change. These remain key areas for future research.
We also note that our analysis does not take into consideration the potential necessity for horticultural industries to shift geographically to adapt to climate change. Analysing shifts in climatic suitability for horticultural crops is complicated by our capacity to modify the environment (e.g. through irrigation), and thus was beyond the scope of this study.
To conclude, surveillance activities, pre-and post-harvest treatment, and control activities for fruit flies present a substantial cost to Australia's horticultural industries [2,4,14]. Our analysis highlights that the major horticultural production regions are likely to remain suitable for multiple economically important fruit fly species as climate changes. Furthermore, given that knowledge of species current distributions remains the basis for market access decisions, the potential for range shifts to occur is of critical interest to horticultural industries. Outputs from this study provide guidance to pest managers, such that they can assess pest risks and design appropriate ongoing surveillance strategies. Our results emphasize the importance of vigilance and preparedness across Australia to prevent further range expansion of these 11 species, and underscore the need for ongoing research and development into monitoring, control, and eradication tools.