Detailed assessment of the reported economic costs of invasive species in Australia

The legacy of deliberate and accidental introductions of invasive alien species to Australia has had a hefty economic toll, yet quantifying the magnitude of the costs associated with direct loss and damage, as well as for management interventions, remains elusive. This is because the reliability of cost estimates and undersampling have not been determined. We provide the first detailed analysis of the reported costs associated NeoBiota 67: 511–550 (2021) doi: 10.3897/neobiota.67.58834 https://neobiota.pensoft.net Copyright Corey J. A. Bradshaw et al. This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. RESEARCH ARTICLE Advancing research on alien species and biological invasions A peer-reviewed open-access journal


Introduction
Biological invasions continue to erode economies, ecosystems and societies worldwide, with no sign of abatement (Simberloff et al. 2013;Bradshaw et al. 2016;Pyšek et al. 2020). As the rate of introductions of invasive alien species accelerates given an increasingly connected world (Seebens et al. 2017), the extent and magnitude of these impacts will ipso facto also increase. While in recent decades, much research has examined the ecological effects of invasive species across habitat types, geographic regions and taxonomic groups (Crystal-Ornelas and Lockwood 2020, and references therein), quantification of the economic impacts has remained diffuse. In particular, a lack of resolute, comprehensive and synthesised economic cost estimates precludes adequate comparisons and compilation at, for example, the national level. Such information can help to assist in setting priorities by policy-makers and organisations for managing invasive species in some of the most impacted countries.
Recently, the InvaCost database was developed to provide the most comprehensive and standardised compilation of invasion costs globally (Diagne et al. 2020b). This advance now addresses the aforementioned limitations by presenting economic costs at a global scale, yet with sufficient resolution to enable assessment in more granular national, taxonomic and socioeconomic contexts. Further, InvaCost allows for assessment of the reliability of cost estimates, as well as for whether costs are predicted to occur or have been empirically observed. While broad-scale perspectives of the economic costs of invasive species are needed because of the transboundary nature of invasions, national or regional assessments are still required in much greater detail (Diagne et al. 2020a).
Australia -the sixth largest country (7,688,287 km 2 ) and thirteenth largest economy (2017 gross domestic product = US$1.23 trillion; worldbank.org) in the world, as well as the only true 'island' continent apart from Antarctica -has a long history of deliberate and accidental introductions of invasive species (Hoffmann and Broadhurst 2016). Introductions by humans go back as far as 5,000-10,000 years with the deliberate introduction of the dingo (Canis dingo) (Smith et al. 2019) and, today, many different alien species occupy almost every terrestrial, freshwater and marine habitat in the country. Indeed, some of the most infamous international examples of deleterious invasive species are Australian -cane toads (Rhinella marina) (Lever 2001), prickly pear cactus (Opuntia spp.) (Freeman 1992), swamp buffalo (Bubalus bubalis) (Ridpath and Waithman 1988), foxes (Vulpes vulpes) (Saunders et al. 2010) and European rabbits (Oryctolagus cuniculus) -to name a few. While there have been some successes in suppressing various alien species using biological control and corresponding savings in averted damage, such as the prickly pear cactus (Raghu and Walton 2007) and European rabbits (Cooke et al. 2013), most invasive species represent major ongoing ecological, agricultural and economic problems for the country.
While there have been previous attempts to evaluate the costs of invasive species to the Australian economy, these have either focussed on one or only a few taxa, or have been restricted to particular regions. Only the impacts of invasive plants have been the subject of analyses at the kingdom level (e.g., Sinden et al. 2004). Moreover, most assessments have been reliant on flawed assumptions (Sagoff 2008;Holmes et al. 2009) and extrapolations (Pimentel et al. 2001) or have applied more top-down approaches to estimate costs by sector, rather than to divide the estimates among species, regions, sectors or decades (McLeod 2004;Sinden et al. 2004;Gong et al. 2009;Hoffmann and Broadhurst 2016;Llewellyn et al. 2016).
Here we focus on Australia and its territories to provide the first detailed assessment of the reported economic costs of invasive species since the 1960s, based on records extracted from the recently published InvaCost database (Diagne et al. 2020b), combined with both an independent database of costs restricted to invasive herbivore species (previously unpublished) and recent data describing the costs of invasive plants and other disease-causing agents. Our aims are to (i) assess the reliability (values based on actual measures as opposed to non-sourced estimates) of the Australian cost estimates, as has been done previously for invasive insects (Bradshaw et al. 2016) and invasive species globally ), (ii) provide a State/Territory summary of those costs, (iii) identify the costliest species nationally and by State/Territory, (iv) investigate the most impacted environments and sectors and (v) estimate robust temporal trends in the economic costs of invasive species over the last five decades.

Data collection
To determine the cost of invasive species to the Australian economy, we used cost data collected in the InvaCost database (Diagne et al. 2020a, b) (n = 2,419 entries) concerning the global costs of invasive species, based on published literature, enabling comprehensive quantification of costs associated with invasive species at various spatio-temporal scales. Of these, 877 (36%) entries pertained to Australia. The data in InvaCost were collected following a series of literature searches using the Web of Science platform (webofknowledge.com), Google Scholar (scholar.google.com) and the Google search engine (google.com) and all the retrieved costs were converted to a common, up-to-date currency (2017 US$; data.worldbank.org).
We complemented the InvaCost data in three ways. We first added supplementary cost data from new references containing cost information (~ 2300 entries; https://doi.org/10.6084/m9.figshare.12928145). Next, we added data from the "Costs of Invasive Herbivores in Australia" database compiled by Biosecurity South Australia in 2018. The latter is an unpublished database compiled by L.A. to collate peer-reviewed and government documents reporting estimated costs specifically for 'invasive' herbivores [this can include native species, which compete with human interests in some cases -for example, kangaroos (various species, notably Osphranter and Macropus spp.)]. That database also includes pigs (Sus scrofa) and birds, even though these species are not all strictly herbivores. Based on the top five commodities in each of the categories of livestock, crops and horticulture as a starting point, the impacts from pest animals on those commodities were compiled for each. Estimates were identified using Google Scholar and Google search engine for peer-reviewed papers, conference papers, surveys and reports (e.g. Australian Bureau of Agricultural and Resource Economics, Invasive Animals Cooperative Research Centre, government and industry reports).
As a last step, we augmented the database with additional, missing cost estimates identified during the review process, as well as additional searching. We included all new costs following the structure, decision points and rules of the original InvaCost data (Diagne et al. 2020b). Many of the additional costs were derived from a single, large report on weeds of cropping systems by Llewellyn et al. (2016). The reporting units used in that report were the Grain Research and Development Corporation agroecological zones and some of these zones crossed state-government boundaries. To assign costs to the State level where an agroecological zone crossed State boundaries, we assumed that costs were evenly distributed across each zone and divided the reported costs proportionally into their respective States. Furthermore, Llewellyn et al. (2016) reported the annual ongoing costs of weed management and these costs were updated by McLeod (2018) for the year 2018 and onwards. To avoid double counting these costs, we extended the Llewellyn et al. (2016) costs up to 2017and used McLeod (2018 from 2018 onwards. These added costs included new estimates that included the present year (2020). At the time of writing, there were no exchange rates or consumer price index data available from the chosen InvaCost sources. As such, we used an 11-month average (Jan-Nov 2020) exchange rate taken from rba.gov.au/statistics/historical-data.html. We calculated the relevant consumer price index by taking the 12-month average change to November 2020 reported at bls.gov and applied it to the 2019 consumer price index reported by the chosen InvaCost data source (data.worldbank.org).
We reviewed all sources, as well as the references they cited, to identify additional sources. Each entry recorded: (i) species identity ('general' if unspecified); (ii) reported cost (including range if available; no hypothetical costs included); (iii) jurisdiction (including area of coverage if provided); (iv) applicable year(s) (set to year of publication if not provided); (v) implementation (observed or extrapolated); (vi) method (field, desktop, both); (vii) verification (whether approach could be identified/repeated); and (viii) type (control, loss, research, damage, mixed).
After combining the separate databases and standardising/aligning columns, we removed obvious duplicate cost estimates (i.e. same cost figures from (non-)identical sources) following previous protocols (Bradshaw et al. 2016;Diagne et al. 2020b). Following our data processing (see below), we finished with a total of n = 2257 unique entries pertaining to Australia for analysis (database available for download at https:// doi.org/10.5281/zenodo.4455979).

Estimating total costs
Deriving the total cumulative costs of the impacts and management of invasive species over time requires considering the temporal period to which a particular cost estimate applies. We calculated the duration of a cost as the number of years between the probable start and end years provided in the full database. When the exact start year was unknown, we conservatively considered the year of publication of a primary data source as either the start year or the end year, to which the duration (if mentioned) in number of years was considered (by adding or subtracting the number of years) to derive either, respectively, the end or the start year (Diagne et al. 2020b). We did not use data describing costs prior to 1960 to avoid inconsistencies in currency conversion. We also removed all costs identified as 'avoided' because of a given intervention (i.e. unrealised costs).
To calculate the total cumulative cost, we first recalculated all the annual costs for the defined periods of their occurrence using the invacost package in R via the expandYearlyCosts function (Leroy et al. 2020) and then summed them to obtain total costs. We also estimated the invasion costs for a series of sub-categories by summing all entries according to six descriptive columns in the database: (1) method reliabilitythe perceived reliability of cost estimates, based on the type of publication and method of estimation (low or high); following Diagne et al. (2020b), 'high reliability' is accorded if either provided by pre-assessed materials (e.g. peer review, official reports) or using a documented, repeatable and/or traceable method when provided in other grey literature; (2) region -here, we split costs by major political unit in Australia (States and Territories), as well as costs not associated with any particular unit (i.e. national-scale or multiple states/not stated); (3) implementation form -this refers to whether the cost estimate was actually realised in the invaded habitat or merely predicted (observed or potential); (4) type of environment: aquatic, terrestrial or mixed habitats (species that spend part of their life cycle in water); (5) type of cost -(i) damage/loss (damage or losses incurred by invasion), (ii) expenditure (control-related expenditure, such as monitoring, prevention, management or eradication), (iii) general costs, including research and administrative costs and (iv) mixed types; and (6) impacted sector -the activity, societal or market sector that was affected by the cost -these were agriculture, authoritiesstakeholders, energy, environment, forestry, health, public and social welfare, protected areas and trade. We modified individual cost entries not allocated to a single sector to mixed in the impacted sector column. We also provide several taxonomic summaries of the costs to provide the reader with a full appreciation of the relative scale of costs among different contributors. These include by taxonomic Kingdom and taxonomic Class.

Temporal development of costs
For the temporal estimation of the average annual costs, we used the custom invacost package in R (Leroy et al. 2020). This package provides functions for modelling the temporal trend of costs using a selection of both linear and non-linear models to provide a summary and comparison of their respective outputs. Given the evidence that numbers of invasive species show no sign of saturation (Seebens et al. 2017), we expected their associated costs to be stable or increase. We accounted for the effects of time lags between the occurrence of the costs and their reporting by examining 'impact year' relative to 'publication year'. This is because there were often several years between the occurrence of costs and the time when they were reported in the literature ). Here, we determined from both the highly reliable, observed costs and all costs combined that the lag quantiles were: 25% = 0 year; 50% = 1 year and 75% = 3 years. We therefore estimated the 'final' costs for the year 2017 (i.e. three years prior to 2020) in the trend analysis described below -this ensures that we include only the most complete years in the trend analysis (i.e. years expected to have > 75% of cost data).
We applied five different models to quantify the temporal dynamics of reported log 10 costs (costTrendOverTime function in the invacost package; now modelCosts in the latest version of R) because we had no a priori reason to assume that the trends were monotonic (linear or otherwise). The simplest approach is an ordinary least-squares regression (two variants: linear and quadratic to test for monotonic trends or non-linear behaviour, respectively). Additionally, we applied two variants of a robust regression (linear, quadratic -R package robustbase) (Maechler et al. 2020) because the cost data are heteroscedastic (unequal variances) and temporally autocorrelated. We therefore estimated the covariance matrix with heteroscedasticity and autocorrelationconsistent estimators (Andrews 1991) to derive 95% confidence intervals for our models. Robust (MM-type) regression (Yohai et al. 1991;Koller and Stahel 2011) applies iteratively reweighted least-squares to reduce the influence of outliers on parameters estimates. Finally, we applied a generalised additive model (GAM -R package mgcv) (Wood et al. 2016). Generalised additive models use smoothing functions to account for heteroscedasticity, based on a Gaussian location-scale model family. A more detailed description of the methods we applied is provided in Diagne et al. (2021).
We applied these five different models to both the entire cost dataset for Australia, as well as the highly reliable, observed costs only, to predict model-averaged 'final' (for 2017) ) estimated costs, based on the temporal trends of the full and subset data. This incorporates both parameter uncertainty estimated in individual models, as well as model uncertainty regarding the true underlying fit. We did this in two ways: (1) we first calculated Akaike's information criterion weights (Burnham and Anderson 2002) for the three likelihood-based models (ordinary least-squares regressions and generalised additive model) and (2) using the root mean-squared errors as weights to calculate a weighted-mean cost in 2017 (all five models). In addition to the invacost package, all R code and the Australia-specific dataset needed to reproduce the analyses can be accessed on Github via https://doi.org/10.5281/zenodo.4455979.

Total cost
Since 1960, the total estimated cost of invasive species to Australia was US$298.58 billion (2017 value), based on 2078 unique entries (after removing 179 records pertain-ing to avoided costs) in the combined database (6674 expanded yearly values), which is approximately equivalent to AU$389.59 billion (2017 average exchange rate). Of the total costs, the majority (91.6%) were observed (US$273.37 billion) rather than predicted or extrapolated ('potential'; US$25.21 billion) (Fig. 1a). Of the observed costs, most (61.3%) were considered highly reliable (US$183.04 billion). Considering all costs regardless of reliability and implementation type, 27.6% of the total (US$82.29 billion) was not attributable to a single kingdom or was unspecified (Fig. 1b). This arises mainly from a multi-species assessment of costs of invasive species across all of Australia (Hoffmann and Broadhurst 2016). However, when considering only observed, highly reliable estimates, the costliest kingdom of invasive species was plants (US$151.68 billion), followed by animals (US$26.43 billion), with 'diverse/ unspecified' making up only 2.7% of these (135 estimates amounting to US$4.93 billion) (Fig. 1b). There were few entries for Kingdoms Chromista (n = 3; US$27,970), Fungi (n = 3; US$14.69 million; all low reliability; many of the fungal plant pathogens were not specified to Kingdom in the source data and so were designated 'diverse/unspecified') and Bacteria (n = 1; US$16.49 million; low reliability) (Fig. 1b).
There was a large disparity in the proportional attribution of costs by major political unit (States and Territories) whether estimating all costs or focussing on the highly reliable, observed costs only. Aside from the costs not clearly associated with a particular State or Territory (i.e. nation-wide or not specified), Western Australia had the highest total costs (52.7%) when considering all costs (US$17.88 billion) (Fig. 2a) -69.3% of this value is attributed to rats Rattus rattus (US$12.39 billion), but > 99% of this estimate is considered to be of low reliability. When considering only the highly reliable, observed costs, New South Wales had the highest costs (US$5.25 billion), followed by Western Australia (US$4.58 billion) and Victoria (US$3.09 billion) (Fig. 2b).
There was an approximate power-law relationship between the number of unique database entries and the total costs per political unit for both all costs combined (Fig. 2c) or highly reliable, observed costs only (Fig. 2d). These relationships indicate that, with an increase of one order of magnitude in the number of estimates, the estimated costs increase on average by 2.0 (all costs) or 1.9 (highly reliable, observed costs) orders of magnitude. These power-law relationships were also evident for the cumulative data over time (Suppl. material 1: Fig. S1). The magnitude-order increase in costs with the number of database entries appears to be driven mainly by the variation in land surface area among political units (Suppl. material 1: Fig. S2); however, there is no relationship between costs and the number of database entries per unit area (Suppl. material 1: Fig. S2e, f ), suggesting that the intensity of assessment of costs among political units is not systematically different. The Australia-wide or unspecified (to State/Territory) values probably represent some inevitable overlap with the cumulative estimates from the different regions; however, it is impossible to discern to what extent given unspecified attribution in many national-scale analyses (e.g., Hoffmann and Broadhurst 2016).
The costliest kingdom (plants) grouped most (96.5%) of its costs into the 'diverse/ unspecified' category (Fig. 3b). Of the remaining highly reliable, observed costs identified to species, six species accounted for most (61%) of the remaining costs: annual ryegrass (Lolium rigidum), parthenium (Parthenium hysterophorus), ragwort (Senecio jacobaea), cucumis melons (Cucumis spp.), common heliotrope (Heliotropium europaeum) and wild radish (Raphanus raphanistrum) (Fig. 3b). Other invasive plants have Figure 2. a sum of all costs according to attributable major political unit (States and Territories) b sum of highly reliable costs only by political unit c relationship between the number of database entries and all cost estimates by political unit -this also includes the estimate for 'Australia' ('AUS'; not directly attributable to a single State or Territory). The power-law relationship is also shown (evidence ratio = 18013, R 2 = 0.90) d relationship between number of database entries for highly reliable costs estimates by political unit (evidence ratio = 38550, R 2 = 0.91). Abbreviations: ACT = Australian Capital Territory; NSW = New South Wales; NT = Northern Territory; QLD = Queensland; SA = South Australia; TAS = Tasmania; VIC = Victoria; WA = Western Australia; AUS = nation-wide or not specified to which political unit the estimate belongs. 'Australian territory' refers to regions outside State/Territory jurisdication (e.g. Christmas Island, Lord Howe Island).
historically had enormous negative impacts on Australian agriculture, but successful biological control programmes have largely eliminated these costs (e.g. prickly pear cactus and Paterson's curse Echium plantagineum) (Cullen et al. 2012). Some high-cost invasive grasses, such as gamba grass (Andropogon gayanus) (Northern Territory Government 2008), were invariably grouped within this 'diverse/unspecified' category and so species-specific cost estimates were not available. In Australia, exotic grasses have major environmental (e.g. gamba and buffel Cenchrus ciliaris grasses) and agricultural impacts (e.g. Nassella tussocks).
The costliest taxonomic classes of invasive species across all of Australia are mammals, insects and eudicots (respectively), although most estimates cannot be attributed to a single class (Fig. 3a). Among the mammals, cats, rodents (mice Mus musculus and rats Rattus spp.), pigs, rabbits and foxes had the highest costs, accounting for 95% of the total highly reliable, observed costs in this class (US$20.19 billion; Fig. 3c). We generally ordered these by the highly reliable costs, but, in some cases where there were no highly reliable costs for a particular category, we placed the category in the order suggested by total costs.
In fact, the category of 'diverse/unspecified' included these five taxa in many multispecies assessments; so, the costs attributed to these are actually higher. Including low-reliability costs would suggest that rodents -namely, house mice and rats Rattus spp. -were the second-costliest mammals, but most (89%) of this total was attributed to the low-reliability category. We also reported cost estimates for five native species groups, including various kangaroo species, koalas (Phascolarctos cinereus), common wombats (Vombatus ursinus), dingoes (Canis dingo) and Queensland fruit flies (Bactrocera tryoni) given that they are often considered 'overabundant' native 'pest' species because they compete for grazing resources (kangaroos), consume trees in Eucalyptus spp. plantations (koalas), burrow in paddocks (wombats), kill livestock (dingoes) or damage crops outside their native region (Queensland fruit fly). Kangaroos, koalas and wombats together account for only 3.1% of the total including all costs and 2.4% of the total highly reliable, observed costs (99.9% of which is attributed to kangaroos alone). Dingoes are native to Australia (Smith et al. 2019), but here we included all accounts of 'wild dogs', 'dogs' and 'dingoes' as dingoes -adding dingoes to the nativespecies groups increases the percentage represented to 3.5% (all) and 3.1% (highly reliable) (although this percentage is slightly higher in reality because dingo-related costs are sometimes combined with other species). Of course, many other native species cause extensive damage to the agricultural industry, such as birds and many insect species, but reliable estimates of the costs associated with most of these species have not been made for Australia. Within the second-costliest class (insects), most (41.5%) of the highly reliable, observed total is within the 'diverse/unspecified' category ( Fig. 3d). Of the highly reliable, observed cost estimates attributed to single species, 70.7% of the total is from the red imported fire ant Solenopsis invicta (US$1.29 billion), 11.8% from the (native to tropical Australia, but considered invasive elsewhere) Queensland fruit fly Bactrocera tryoni (US$215.45 million), 8.7% from the Pacific fruit fly Bactrocera philippinensis (US$158.91 million) and 7.1% from the bollworm Helicovera spp. (US$129.2 million) (Fig. 3d).
For the third-costliest class, based on all costs combined (Eudicots), five species account for most (56.7%) of all costs attributed to this class: parthenium (18.1%; US$740.66 million), ragwort (10.4%; US$425.37 million), cucumis melons (10.1%; US$412.12 million), common heliotrope (9.4%; US$384.25 million) and wild radish (8.8%; US$361.42 million) (Fig. 3b). Many of the other classes are dominated by one or a few species (Suppl. material 1: Table S1); for example, bird costs are either unspecified or from a single species: the common starling (Sturnus vulgaris); the Arachnids include only two mites: the red-legged earth mite Halotydeus destructor and varroa mite Varroa destructor; the Ulvophytes are represented solely by Caulerpa taxifolia; the Secernentids include only two nematode species (Heterodera avenae and Pratylenchus spp.); the Amphibia include only the cane toad Rhinella marina; the Polypodiopsids are represented only by Salvinia molesta; and the Phaeophyceae (brown algae) include only one species, wakame Undaria pinnatifida (see full species list in Suppl. material 1: Table S1).
The costliest species also vary among States/Territories ( Fig. 4; also see Fig. 5). Mammals (cats Felis catus, red foxes, rabbits) are the costliest species only for Australian Capital Territory and New South Wales (Fig. 4). The Northern Territory's costliest species (36% of its costs) is the fungus Phyllosticta cavendishii that causes banana freckle disease. Queensland's costliest species is the red imported fire ant, representing 27% of the total   Table S1. highly reliable, observed cost for that State (Fig. 4), whereas the common heliotrope is the costliest species for both South Australia and Victoria (6% of the total highly reliable, observed costs for those States). Tasmania's costliest species (62% of all costs) is the ragwort and Western Australia's is annual ryegrass (9% of total costs) (Fig. 4).
The proportional attribution of the highly reliable, observed costs by species per State/Territory is presented in Fig. 5.
The most impacted habitat is the terrestrial environment (39%), although most (60%) of the total highly reliable, observed costs could not be attributed to a single habitat type (Fig. 6a). Damage by or loss of economic opportunity (cf. management) from invasive species has the highest value (US$133.35 billion) among cost types (Fig. 6b), representing 72.9% of the total highly reliable, observed costs. The most-affected sectors are the agriculture (24.1%; US$44.03 billion), health (4.6%; US$8.37 billion) and environment (4.1%; US$7.58 billion) sectors, although most (65.8% of the total highly reliable, observed costs) affected multiple sectors (mixed; Fig. 6c).
Tracking temporal trends , the costs attributed to invasive species in Australia increased from the 1970s to the present. Using all costs irrespective of reliability, the average annual cost increased from US$57.65 million in the 1970s to $20.19 billion during the last decade (Fig. 7a). Although highly variable from decade to decade, this equates to an average decadal increase of ~ 6.3-fold (or 3.2-fold, based on the slope coefficient for the linear robust regression only to compare directly to the 3-fold increase estimated from the global dataset) . Taking only the reliable, observed costs, the average annual cost increased from over US$52.35 million in the 1970s to US$15.12 billion during the last decade (Fig. 7a) or an average 6.0fold increase per decade (or 1.8-fold, based on the slope coefficient for the linear robust regression). This translates into a mean annual cost of US$5.85 billion (all costs) or US$3.58 billion (reliable, observed only) over the study interval (Fig. 7a). Examining the temporal trends in the observed, reliable costs only for three of the main taxonomic groups (plants, mammals, insects) shows the general increasing trend, although the most recent decade's increase is driven primarily by costs attributed to plants (Suppl. material 1: Fig. S3).
For both all-costs and observed, reliable datasets, the general additive model had the best fit assessed using the highest Akaike's information criterion weights (wAIC). However, the quadratic ordinary least-squares model had the best fit for the highly reliable, observed costs, based on the lowest root mean-squared error (RMSE; Table 1).
Using these weights to predict the annual costs in 2017 for both datasets, those based on wAIC are dominated by the GAM prediction, whereas those based on RMSE weights accord relatively more importance to the quadratic models (Fig. 7b, c).
For the all-costs dataset, the estimated annual costs in 2017 are US$18.77 billion (US$6.09 billion-US$57.91 billion) according to wAIC or US$17.88 billion (US$7.56 billion-US$45.44 billion) according to RMSE (Fig. 7b). For the highly reliable, observed data only, the predictions for 2017 are US$731.48 million (US$225.31 million-US$2.38 billion) according to wAIC or US$1.85 billion (US$484.85 million-US$6.84 billion) according to RMSE (Fig. 7c).  . a raw annual costs for all costs (black) and reliable, observed costs (grey). Also shown are the decadal and overall means b predicted annual costs across Australia from 1970 to 2020 for all costs and c reliable, observed costs only. Fitted models include OLS l = linear ordinary least-squares, OLS q = quadratic ordinary least-squares, RR l = linear robust regression, RR q = quadratic robust regression, GAM = general additive model. Also shown in each panel are wAIC $ 2017 = Akaike's information criterion-weighted (wAIC)-average of the predicted annual cost in 2017 (all costs; OLS l , OLS q , GAM only), wRMSE $ 2017 = root mean-squared error-weighted average of the predicted annual cost in 2017 (all costs; all models), wAIC $ 2017 , and wRMSE $ 2017 (reliable, observed costs only).

Discussion
Aggregated economic costs of the impacts and management of invasive species in Australia have amounted to at least US$298.58 billion (~ AU$389.59 billion) since the 1960s and US$183.04 billion (~ AU$238.83 billion) when conservatively considering highly reliable, observed costs only. Sampling biases notwithstanding (see below), the greatest economic burden to Australia imposed by invasive species originates from weedy plants, although most of these costs are shared across a wide range of species. This arises because of the 'top-down' approaches employed by others previously to estimate costs associated with losses and control specific to particular industries, rather than individual species (e.g., McLeod 2004; Sinden et al. 2004;Gong et al. 2009;Llewellyn et al. 2016). In many circumstances, this approach is more tractable and efficient for estimating total costs to particular sectors.
There are an estimated 2700 exotic plant species established in Australia, of which > 400 are declared weedy or noxious (Hoffmann and Broadhurst 2016). Our database contained highly reliable cost estimates for only ~ 100 species of declared weeds and many weeds did not have species-specific costs as described above. The cost of controlling and the damage done by weeds to the Australian agriculture sector alone are estimated at ~ AU$4 billion year -1 (Sinden et al. 2004;Hoffmann and Broadhurst 2016). However, from the perspective of single species, exotic mammals dominate the costs, with cats and rabbits, in particular, showing some of the highest estimates across the entire sample across Australia. For rabbits, it has been estimated that, without the highly successful biological control programme started in the 1950s, the impacts of rabbits in Australia would have been at least US$53.5 billion (~ AU$70 billion) higher over the last 50 years (Cooke Table 1. Model fits to the temporal trend of annual costs from 1970 to 2020 for all data combined and for reliable, observed data only. Fitted models include OLS l = linear ordinary least-squares, OLS q = quadratic ordinary least-squares, RR l = linear robust regression, RR q = quadratic robust regression, GAM = general additive model. Also shown are Akaike's information criterion weights (wAIC) for the three likelihood-based models (OLS l , OLS q , GAM), the root mean-squared error (RMSE) for all models, the R 2 for each model (% deviance explained in the case of GAM) and the estimates of the relevant model coefficients (β year and β year 2 ) and their standard errors (± SE). See also Fig. 7 Hoffmann and Broadhurst (2016) hail from five different sources (Canyon et al. 2002;McLeod 2004;Sinden et al. 2004;Gong et al. 2009;de Hayr 2013) of unknown reliability and/or derived from stakeholder surveys. The management ('national') expenditure component of these were AU$3.0 billion and AU$3.8 billion for 2001-20122017 value). In contrast, our study incorporated appraisals of method reliability and implementation type when considering economic costs, presenting both 'total' and more conservative figures. Considering only those more conservative numbers, we found that damage and resource losses attributable to invasive species outweigh (73% of that total) management expenditure, but to a greater extent than indicated in those previous studies. This likely mirrors the relatively small investment of government funding for the management of most invasive species (Hoffmann and Broadhurst 2016) compared to the actual economic damages they incur. However, we acknowledge that our broad categorisations of cost type and implementation likely obscure subtleties associated with production losses, control costs and environmental impacts on a case-by-case basis. Reporting cost categories at finer resolution would likely invoke unacceptable subjectivity in reporting given the diversity of species, approaches, sectors, cost types, analyses and assumptions made in individual reports. We acknowledge, however, that the environmental and social costs recorded in InvaCost should be considered with some caution regarding their interpretation, because they are not strictly similar to market costs recorded in economic sectors (Diagne et al. 2020b). Further, cost categorisations for particular species likely shift in terms of emphasis during the course of invasions, meaning that management investment for many species begins with eradication costs and ultimately changes to suppression via control management as the species becomes established. Indeed, government investments typically target new incursions first, meaning that many of these are unlikely to be captured in the relevant literature.
As most cost estimates are damage arising from invasive plants (weeds), it is understandable why management-related costs represent such a small proportion of the total. However, one invasive mammal species is problematic in this regard -cats. According to the definition of 'reliable' provided for the overall InvaCost database -"Peer-reviewed articles and official documents (e.g. institutional or governmental reports) are likely validated by experts before publication. We assumed, therefore, that all cost estimates collected from these materials may likely be of high reliability" (Diagne et al. 2020b)we were objectively obliged to include the damage estimate of US$5.95 billion for this species from Pimentel et al. (2001). However, that particular estimate was based on an unverified national population of 18 million feral cats and a subjective value of a bird eaten of US$30 (amount to US$540 million year -1 ) (Gregory et al. 2014). The subjective extrapolation of the costs of cats was also noted for the USA (Fantle-Lepczyk et al. 2021).
Compared to the global estimates , the relatively well-sampled region of Oceania represents ~ 8% (range: 3-22%) of the total average annual costs globally in 2017 according to our database. Further, we found that Australia's rate of cost increase was up to ~ 2 times the rate of cost increase estimated from the global dataset , although this observation might be explained in part by a lack of data in other regions compared to relatively well-studied Australia. However, Australia is still likely to be recording only a portion of the total costs of invasive species in the region. Although InvaCost, in general, as well as our enhanced sample from Australia more specifically, represent the most comprehensive and resolute assessments of the costs of invasive species yet available, there are several lines of evidence to suggest that the totals we report here still represent a vast underestimate of the real costs.
The first line of evidence is that the estimated total costs increased by approximately two orders of magnitude with every order-of-magnitude increase in the number of entries. This accords well with other assessments revealing that, as the number of estimates increases, so too do the total costs (Bradshaw et al. 2016;Cuthbert et al. 2021a;Diagne et al. 2021) -in other words, the more economic assessments are done, the more costs are discovered. While this could arise in part from the increasing rate of scientific and related publishing over the last 50 years (Richardson and Pyšek 2008), under-sampled or under-assessed species and regions will necessarily underestimate total costs. This particularly holds true to aquatic and semi-aquatic alien taxa in Australia, with the majority (99.9%) of costs attributed to a particular habitat (i.e. excluding mixed-habitat costs) being terrestrial (observed, highly reliable costs) and terrestrial taxa dominating in most regions. On the global scale, this aligns with the under-representation of aquatic invaders relative to terrestrial ones (Cuthbert et al. 2021b).
The second line of evidence is that many well-known invasive species established in Australia have no associated cost estimates in the database. For example, there was not a single estimate from the Reptilia in the database, yet species like red-eared sliders (Trachemys scripta elegans) and corn snakes (Pantherophis guttatus) are potentially costly species in some States of Australia (García-Díaz et al. 2017;Toomes et al. 2020). Neither were pet trade-sourced bird species like rose-ringed parakeets (Psittacula krameri) (Vall-llosera et al. 2017;Toomes et al. 2020) or fish pests, such as European carp (Cyprinus carpio) (Koehn 2004), identified in our cost database. Neither have native birds (Bomford and Sinclair 2002) or insects (Gu et al. 2007) that can heavily damage various crops been adequately assessed for costs (apart from the Queensland fruit fly). Indeed, Gong et al. (2009) reported that birds were the costliest vertebrate group to Australian agriculture. Despite reporting the costs for several fungal plant pathogens, there were notable absences; for example, we could not identify any reasonable costs estimates for Phytophthora cinnamomi, despite its being a major cause of crop losses and damage to biodiversity in Australia (Cahill et al. 2008;Hee et al. 2013).
The third line of evidence is that the InvaCost approach mandates avoiding the extrapolation of on-going costs beyond the time period specified by a particular source (Diagne et al. 2020b). We therefore included costs without a specified time window as single-year costs, meaning that the resultant annualised costs represent a lower boundary of the true costs. While this avoids propagating positive errors through time, it downwardly biases the true mean costs. The fourth line of evidence is that the number of invasions in Australia has been increasing linearly for some time (CSIRO 2020), which is a notable improvement from the current exponential trend seen globally (Seebens et al. 2017); this accords well with our temporal analysis indicating an ongoing increase in recorded costs over last few decades (Fig. 7b, c). More broadly, lags in invader impacts considering their year of introduction (Rouget et al. 2016) could mean that it takes decades for economic costs to be realised and reported, just as it takes decades of introductions to become invasions (Essl et al. 2011). Accordingly, future economic impacts will likely result from a different suite of invasive species for which the effects have not yet been fully realised.

Conclusions
While the major costs of loss and damage arising from invasive species, where tangible, are probably captured reasonably well by our database (the under-sampling bias notwithstanding), management-expenditure estimates are perhaps less reliable. The component of the total costs of invasive species attributed to management expenditure is particularly problematic for several reasons. Indeed, there is no standard procedure for reporting expenditure or costs at any level of government or for private organisations, nor is there a national database of expenditure available (Hoffmann and Broadhurst 2016). A similar argument could also be mounted for damage and loss assessments regarding a lack of a standardised reporting protocol.
As our assessment highlights, the large and growing costs of invasive species to the Australian economy are substantial, but under-estimated because of insufficient coverage and a lack of standardised reporting by management authorities and other agencies. As invasive species continue to increase their ranges and associated impacts across the planet (Bellard et al. 2013Seebens et al. 2017;Seebens et al. 2021), we can reasonably surmise that Australia will also suffer many additional, negative economic repercussions from invasive species over the coming decades. Developing better methods of estimating environmental impacts of invasive alien species will also contribute to this. We recognise that such types of economically intangible costs arising from invasive species (Bradshaw et al. 2016) are not captured by the database -for example, ecological damage, erosion of ecosystem services and loss of cultural values are inherently challenging to measure in this regard.