Economic costs of invasive rodents worldwide: the tip of the iceberg

Background Rodents are among the most notorious invasive alien species worldwide. These invaders have substantially impacted native ecosystems, food production and storage, local infrastructures, human health and well-being. However, the lack of standardized and understandable estimation of their impacts is a serious barrier to raising societal awareness, and hampers effective management interventions at relevant scales. Methods Here, we assessed the economic costs of invasive alien rodents globally in order to help overcome these obstacles. For this purpose, we combined and analysed economic cost data from the InvaCost database—the most up-to-date and comprehensive synthesis of reported invasion costs—and specific complementary searches within and beyond the published literature. Results Our conservative analysis showed that reported costs of rodent invasions reached a conservative total of US$ 3.6 billion between 1930 and 2022 (annually US$ 87.5 million between 1980 and 2022), and were significantly increasing through time. The highest cost reported was for muskrat Ondatra zibethicus (US$ 377.5 million), then unspecified Rattus spp. (US$ 327.8 million), followed by Rattus norvegicus specifically (US$ 156.6 million) and Castor canadensis (US$ 150.4 million). Of the total costs, 87% were damage-related, principally impacting agriculture and predominantly reported in Asia (60%), Europe (19%) and North America (9%). Our study evidenced obvious cost underreporting with only 99 documents gathered globally, clear taxonomic gaps, reliability issues for cost assessment, and skewed breakdowns of costs among regions, sectors and contexts. As a consequence, these reported costs represent only a very small fraction of the expected true cost of rodent invasions (e.g., using a less conservative analytic approach would have led to a global amount more than 80-times higher than estimated here). Conclusions These findings strongly suggest that available information represents a substantial underestimation of the global costs incurred. We offer recommendations for improving estimates of costs to fill these knowledge gaps including: systematic distinction between native and invasive rodents’ impacts; monetizing indirect impacts on human health; and greater integrative and concerted research effort between scientists and stakeholders. Finally, we discuss why and how this approach will stimulate and provide support for proactive and sustainable management strategies in the context of alien rodent invasions, for which biosecurity measures should be amplified globally.


Introduction
Rodents, the most abundant and diversi ed order of living mammals (~ 40% of mammalian biodiversity; Burgin et al. 2018), are undoubtedly the vertebrate group that has most often accompanied humans throughout their history of global dispersal (e.g. Cucchi et al. 2020). The ever-increasing intensi cation of human enterprise (e.g. maritime trade, road development) together with habitat modi cations (e.g. landuse changes, urbanization) has resulted in a global spread of numerous non-native rodents, with some species continuing to proliferate; Dalecky et al. 2015;Di Febbraro et al. 2019;Hassell et al. 2021). In addition, the ecological exibility of these rodents has allowed adaptation to heavily human-modi ed habitats, facilitating both their spread and acclimatization to new areas (Hima et al. 2019;Mazza et al. 2020). Once established, these commensal non-native rodents are usually highly proli c and represent a multisectoral threat to local biodiversity (e.g. (Sainsbury et al. 2020)), public health (Han et al. 2015;Meerburg et al. 2009a), human well-being (Colombe 2019) and socio-economic activities (Murray et al. 2018).
Indeed, invasive rodents have numerous detrimental impacts on invaded ecosystems, resulting from both direct (e.g. competition, predation, destruction through digging and gnawing) and indirect (e.g.  (Diagne et al. 2021). Annual production losses attributed to rodents have been assessed at US $1.9 billion in Asia (Nghiem et al. 2013), US$ 45 million in the United Republic of Tanzania (Leirs 2003), US$ 19 billion in the United States of America (Pimentel et al. 2005), and US$ 60 million in Australia (Brown & Singleton 2000). However, we lack an essential, global overview of these economic costs, which is necessary for both research needs (e.g. identifying gaps and priorities) and management strategies (e.g. providing a basis for coordinating regional biosecurity measures) (Diagne et al. 2020a;Early et al. 2016).
Here, we provide the rst global synthesis of the reported economic costs of invasive alien rodents. For this purpose, we relied on the recently developed InvaCost database, which is the most up-to-date and comprehensive living database of the economic costs of invasive alien species reported worldwide (Diagne et al. 2020b). Speci cally, our goal was to describe and model the cost of invasive rodents to human society over time, and summarize how costs are distributed across rodent taxa, geographic space, socio-economic sectors and types of costs. From this, we identify research needs for consistent identi cation and use of costs across disparate sectors as well as crucial societal interactions in the perspective of an e cient management of invasive alien rodent impacts.

Data collection and processing
We considered the most recent version of the InvaCost database (version 3.0, available at https://doi.org/10.6084/m9. gshare.12668570 ). This database includes 9,823 cost entries collated from scienti c and grey materials in multiple languages (Angulo et al. 2021;Diagne et al. 2020b). Each cost entry recorded is standardized to 2017 US dollars and categorized by a range of 64 descriptive elds (Online Resource 1, Tab Descriptors). We selected cost entries identi ed as Rodentia in the 'Order' column of the database (Fig. 1). We carefully checked the data for any duplicates or mistakes, and all modi cations made were sent to updates@invacost.fr as recommended by the database managers. The resulting dataset (n = 349 cost entries) is provided as the suitable subset in the Online Resource 1 (Tab. Original subset). This suitable subset was homogenized so that all cost entries were considered on an annual basis, meaning that costs spanning multiple years were divided according to their duration (e.g. $20 million between 1991 and 2000 becomes $2 million annually across those years). Annual costs were calculated through a subset 'expansion' process using the expandYearlyCosts function of the 'invacost' R package (Leroy et al. 2020; R Core Team 2019). The duration time of each cost occurrence was calculated as the number of years between the recorded cost entry's starting ('Probable starting year adjusted' column) and ending ('Probable ending year adjusted' column) years. Any cost entries without available information in one or both columns were conservatively removed from this expansion process, and thus our analyses. The resulting subset (n = 718 cost entries) is provided as the expanded subset in the Online Resource 1 (Tab. Expanded subset). In addition, we applied two successive lters to this expanded subset to obtain a conservative subset (Fig. 1): rst, we kept only observed costs (rather than potential costs, under the 'Implementation' column ; thereby removing, for example, all predicted costs); second, we retained only high-reliability costs (rather than low-reliability costs, under the 'Method reliability' column; thereby removing, for examples, all costs without sourced information) -see Online Resource 2 for distribution of cost data within both descriptive elds. Our conservative subset contained 426 annualized cost entries between 1930 and 2018 ( Fig. 1, Online Resource 1, Tab. Conservative subset). From there, total costs were obtained by summing all annualized cost entries ('Cost_estimate_per_year_2017_USD_exchange_rate' column) from this conservative subset.

Temporal dynamics of costs
We examined how costs developed over time (since 1980) using the modelCosts function to t multiple models to the conservative subset. Such modelling of the trend of costs over time allows for a more reliable estimation of the dynamics of total annual costs by taking into account the time lags between the real occurrence of the costs and their reporting in the literature, as well as the heteroscedastic and temporally auto-correlated nature of cost data (Leroy et al. 2020). We therefore removed post-2013 years from this analysis, due to time lags in cost reporting. We subsequently employed a range of modelling techniques on the conservative subset data: ordinary least squares regression (linear and quadratic), robust regression (linear and quadratic), multivariate adaptive regression splines (MARS), generalised additive models (GAMs) and quantile regression [0.1 (lower boundary of cost), 0.5 (median cost value), 0.9 (upper boundary of cost)]. Model evaluation was based on the assessment of their predictive performance (via root-mean-square deviation, RMSE) and the level of variance explained. Although predictions will inherently vary among models, combining these diverse modelling procedures offers strong support for the resulting temporal trends if most or all of them provide consistent outcomes.

Taxonomic bias
To identify the proportion of invasive rodent species for which cost data is available, we compared the individual rodent species reported in the original subset with comprehensive lists of invasive rodents recorded worldwide, following an approach similar to Cuthbert et al. (2021). Lists of known invasive rodents were extracted and compiled from the Global Invasive Species Database (GISD; http://www.iucngisd.org/gisd/) and the sTwist database (version 1.2; Seebens et al. 2020b). We ltered these databases to select only species belonging to the order Rodentia and used the GBIF.org Backbone Taxonomy to standardize species names and removed any duplicated species. Then, for the latter rst records (sTwist) database, we selected only taxa that were known to be presently established. We classi ed all such species as invasive, but note that the de nitions of invasiveness may differ slightly between these datasets (Cuthbert et al. in press). Within each taxonomic family, we thus obtained the proportion of invasive rodent species with costs recorded in InvaCost.

Cost distribution
We subsequently investigated how economic costs of invasive rodents were distributed across key database descriptors using the conservative subset (see Online Resource 1, Tab Descriptors for details on all descriptors and categories). We included the (i) 'Species' (undetermined species were aggregated by genus, where possible), (ii) 'Geographic region' and 'O cial country' where the cost occurred, (iii) 'Type of cost' (Damage [economic losses due to direct and indirect impacts of rodents] vs. Management [monetary investments to prevent and/or mitigate impacts, further separated according to type of actions undertaken: pre-invasion management, post-invasion management and research/funding]) and (iv) 'Impacted sector' (Agriculture, Authorities-Stakeholders, Environment, Fishery, Forestry, Public and social welfare). We also included an additional comparison (v) insular habitat status ('Island' = yes or no). For each descriptor, we grouped under mixed all cost entries that were not unambiguously assigned with one of the above-mentioned speci c categories.

Global cost and temporal dynamics
Based on costs reported in our conservative data subset, we found that invasive alien rodents have already cost the global economy at least US$ 3.28 billion between 1930 and 2018 (Online Resource 2). A less conservative approach would have produced a gure of around US$ 35.53 billion worldwide ( Fig. 1).
Models considering the temporal dynamics of costs were generally convergent in showing an increase in invasion costs over time (Fig. 2), con rming the raw temporal trends directly based on the cost estimates (Online Resource 3). All models displayed a relatively similar goodness of t (RMSE 0.75-0.77), with costs in the year 2020 projected between US$ 511 million (linear ordinary least squares regression) and US$ 10 billion (quadratic robust regression) (Online Resource 2). Quantiles were increasingly divergent through time, indicating greater amplitudes between lower and upper cost quantiles in recent years. This global cost was unevenly distributed across taxonomic groups, geographic areas, types of costs and societal sectors (see below). Note that all costs provided here are summarized in the Online Resource 4.

Taxonomic cost distribution and bias
Invasion costs were reported for 12 individual rodent species in our conservative subset, while there are 48 invasive alien rodents recorded worldwide (i.e. across InvaCost, sTwist, GISD; Fig. 3). Two further species recorded in the original InvaCost database were not included in our conservative subset ( Fig. 1;  Fig. 3). Speci cally, costs for Hystrix brachyura and Sciurus niger, either reported (for H. brachyura in the UK) or expected (for S. niger should it arrive in the Netherlands), were respectively deemed as lowreliability or potential estimates. The most underrepresented rodent families in our subset include Sciuridae (13 species without costs out of 17), Muridae (9 species out of 13) and Cricetidae (5 species out of 6). Additionally, the families Cavidae, Dasyproctidae and Heteromyidae did not have any reported costs (Fig. 3).
Costs were skewed towards the muskrat Ondatra zibethicus (US$ 378.1 million; n = 18 annualized cost entries), unde ned rats Rattus spp., (US$ 329.3 million; n = 82), the brown rat R. norvegicus (US$ 145.8 million; n = 29) and the North American beaver Castor canadensis (US$ 103.9 million; n = 15). These four taxa constituted about a third of the total costs reported. All remaining species-speci c costs totaled less than US$ 100 million, but mixed costs (diverse or nonspeci c taxa) collectively amounted to US$ 2.17 billion. Despite being the species with the higher number of annualized entries (n = 110) in our conservative subset, costs from the coypu M. coypus totaled only US$ 70 million.

Cost distribution across types, space and sectors
Most costs were due to resource damages or losses (91%; US$ 2.99 billion, n = 134). Management actions (or mixed damage-management) comprised the remainder, though they had a higher number of annualized entries (n = 269 for management actions, 21 for mixed damage-management). In turn, management spending was dominated by post-invasion management (US$ 260 million, n = 196), which was over 170-times greater than pre-invasion management (Fig. 4a). While the aforementioned costliest species remain overall the same regardless of cost type, their relative ranking changes when considering damage versus management. Regarding damage costs, O. zibethicus (US$ 328.6 million; n = 8) was the costliest species, followed by R. norvegicus (US$ 68.5 million; n = 2) and C. canadensis (US$ 65.4 million; n = 9). The only speci c species with more than 10 damage cost entries were M. coypus (n = 69) and Callosciurus erythraeus (n = 32), which totaled, respectively, US$ 64.3 million and US$ 1.98 million. Conversely, management costs were mostly associated with R. norvegicus (US$ 68.6 million; n = 26), and then O. zibethicus (US$ 49.5 million; n = 10) and C. canadensis (US$ 38.5 million; n = 6). While being the species complex incurring the highest damage costs (US$ 304.2 million; n = 4), unde ned Rattus spp. represented the fourth taxon for which money was spent for management actions (US$ 24.9 million; n = 75).
Regionally, most costs were incurred in Asia (65%; US$ 2.16 billion, n = 87), followed by Europe (20%; US$ 659.3 million, n = 206) and North America (9 %; US$ 297.9 million, n = 21), with remaining regions contributing around US$ 100 million or less each. Most species recorded impacts in only a few geographic regions (Fig. 4b). In particular, the costliest species O. zibethicus only incurred costs in Europe. Where de ned, mainland areas incurred higher rodent invasion costs than islands overall (US$ 457.3 million, n = 199 vs. US$ 314.2 million, n = 179) (Fig. 5). Rodent damage represented most costs (88%) on mainland areas, but only a third of the total costs reported in islands. Conversely, management spending was considerably greater on islands (US$ 192.7 million, n = 157) compared to mainland areas (US$ 54.5 million, n = 89). While post-invasion actions dominated management spending overall, preinvasion management actions were only reported for islands (Fig. 5).
Regarding impacted sectors, most costs were incurred by the Agricultural sector (66%; US$ 2.16 billion; n = 102) with two-thirds of this cost recorded in Asia, followed by expenditures by Authorities and stakeholders (22%; US$ 741.4 million; n = 288), of which slightly more than half occurred in Europe. Almost all (~ 95%) of the agricultural costs were attributed to diverse or unspeci ed taxa, while for the costliest species, O. zibethicus, 46% of the costs were borne by Authorities and stakeholders. All other speci ed sectors represented less than US$ 10 million and ten annualized entries.

Discussion
Tremendous, increasing and uneven economic costs Rodent invaders have conservatively cost the global economy at least US$ 3.28 billion between 1930 and 2018. Inclusion of all costs through less conservative data ltering leads to a global amount more than ten times higher (US$ 35.53 billion; Fig. 1). Whatever the actual cost gure, these costs are undeniably increasing. All models of temporal cost dynamics converged to depict an exponential increase over time in invasive rodent costs (Fig. 2). Although projections were variable due to underlying model characteristics, annual costs of rodent invasions were predicted to reach as much as US$ 10 billion in 2020. This gure is striking, and to provide perspective, is higher than the European Union's negotiated budget for addressing the COVID-19 crisis (US$ 7.3 billion, consilium.europa.eu) during the same year. The fact that these annual costs show no sign of slowing re ects the ongoing increase in rates of biological invasions globally (Seebens et al. 2020;Seebens et al. 2017). Although increasing reporting of costs cannot be clearly disentangled from empirically rising cost gures, the ongoing intensi cation of global trade, transport networks and human-induced habitat modi cation continues to provide new opportunities for further rodent invasions, and their associated costs worldwide (Hassell et al. 2021;Seebens et al. 2020).
This global cost gure is unevenly distributed across taxa, space, sectors and types of costs. From a taxonomic perspective, most costs were attributed to species belonging to the genus Rattus, with a cumulative cost of around US$ 480 million. Whether due to their actual impacts (e.g. role as disease reservoir; Morand et al. 2015); alteration of socio-economic activities, Murray et al. 2018; impacts on biodiversity and ecosystems; Doherty et al. 2016)) or intensive research effort (this is probably the most documented rodent genus worldwide: e.g. invasion history and introduction pathways; Zeng et al. 2018), Rattus spp. are recognized among the worst invaders worldwide (Lowe et al. 2000;Luque et al. 2014). Ondatra zibethicus, the individual species with the highest reported costs but exclusively in Europe) causes huge impacts through its burrowing ability (which can severely damage local habitats, roads and hydraulic systems) and its capacity to transmit zoonotic diseases (Nentwig et al. 2018).
From a geographic perspective, Asia (US$ 2.16 billion) comprises the highest proportion of the total cost, mainly due to a single estimate associated with agricultural losses from Mus and Rattus species in Malaysia, Myanmar and Thailand (Nghiem et al. 2013). Interestingly, despite having the third highest number of cost entries (n = 48), Oceania ranked 5th regarding total costs, likely because this region is mainly associated with management costs rather than more costly damages. The higher reporting rate (number of entries) of cost estimates observed in Europe, North America and Oceania could also re ect biases in research efforts and/or economic capacities rather than an accurate spatial distribution of costs, as shown for invasion science in general (Pysek et al. 2008). In turn, scarce cost reporting in low income regions likely re ects low priority given to IAS research and/or limited capabilities to act against invasions (Early et al. 2016). Indeed, costs of invasive rodents in Africa represented less than 1% of the total cost estimates among continents. Yet, globally common invasive rodents (R. rattus, R. norvegicus and M. musculus) are known to have the same described detrimental impacts there (Dossou et al. 2020; van Wilgen et al. 2020).
From a sectoral perspective, our results highlight that most costs from invasive rodents are simultaneously associated with several societal and activity sectors, illustrating the intrinsic multisectoral nature of rodent impacts (Colombe 2019). For example, a single invasive rodent -the Eastern grey squirrel S. carolinensis -may simultaneously impact local biodiversity, affect people's possessions, consume ornamental plants and bark stripping activity (Broughton 2020). Considering individual sectors, agricultural losses unsurprisingly comprised the greatest proportion. Globally, rodents are among humans' most important competitors for food resources, particularly through the damage they cause to growing crops and stored products (Belmain et al. 2015). Relative to agriculture, other sectors have reported more marginal rodent impacts, but this should clearly be viewed in light of the di culties present in monetizing substantial ecological and health impacts (Diagne et al. 2021).

An undervalued economic burden
However high these costs may be, there is no doubt that they are a massive underestimation of the total costs incurred by invasive alien rodents globally. As for all invasive alien species (IAS) costs, there are a number of logistical, methodological and cost-intrinsic factors that fail to encompass the full diversity, and thus total cost of invasive rodents (Diagne et al. 2021). Although our synthesis is based on the most complete and up-to-date compilation of reported IAS costs worldwide, it is not exhaustively comprehensive. For example, the accessibility of grey literature materials varies (Angulo et al. 2021;Diagne et al. 2020a), monetary valuation of non-market ecosystem services is not straightforward (Kallis et al. 2013;Spangenberg and Settele 2010), and there is active ethical debate surrounding the principles of monetary valuation processes (Meinard et al. 2016). In addition, our choice to only examine the most robust subset (Fig. 1), and thus exclude any unsubstantiated costs (e.g. those relying on unsourced hypothetical calculations) also contributed to reduce the total cost over 1930-2018, which also contribute to explain the striking discrepancy between the resulting global annual cost (US$ 7.71 million) and some local estimates previously provided elsewhere (e.g. US$ 19 billion per year in the United States; Pimentel et al. 2005).
Further evidence of underestimation is seen in the number of invasive rodents for which no invasion costs have been reported so far (Fig. 3). In fact, reported costs were only available for one-quarter of known invaders, though it is unlikely the other species have no signi cant economic impacts. Another challenge to estimating costs from rodent IAS is the lack of systematic distinction between invasive and native rodent species when assessing economic losses and expenditure. This may be because of di culty in attributing costs between often morphologically similar species. Moreover, particularly for invasive mice and rats, their commensal habits and long-standing invasion history may mean that most of the time they are classi ed as generic pests rather than speci cally as invasive species (Stenseth et al. 2003), and so are treated differently within the literature, especially outside ecology (e.g. agriculture and health). In this instance, the search terms used within InvaCost may be not optimally designed for capturing such costs. Additionally, rodents (including most invasive ones) are also major reservoirs of pathogenic agents responsible for both zoonotic and veterinary diseases (Colombe 2019;Han et al. 2015;Meerburg et al. 2009a). These diseases are associate with substantial costs from both direct (e.g. disease control; medical care) and indirect impacts (e.g. disabilities resulting in decreased productivity and loss of income; disturbed tourism). However, such costs can again be di cult to monetize (Diagne et al. 2021), or they may be attributed to the pathogens or arthropod vectors rather than explicitly to invasive rodents. In the same vein, economic losses or expenditures associated with (invasive) rodents are often provided in terms of incurred damage rather than in speci c monetary terms. For example, rats were estimated to consume food crops that could feed 200 million people in Asia for an entire year (Singleton 2003), and it was estimated that 280 million cases of undernourishment could be avoided worldwide through proactive rodent control (Meerburg et al. 2009b). Similarly, rodent-borne zoonoses are responsible for over 400 million human illness cases each year, leading thus to a cascade of socioeconomic consequences (Meerburg et al. 2009a); as an illustration, the invasive R. norvegicus plays a pivotal role in the epidemiological cycle of leptospirosis in many urban settings, which is associated with a global loss of 2.9 million Disability Adjusted Life Years (DALY) annually (Torgerson et al. 2015). Such gaps and underestimates of costs identi ed by our synthesis highlight the need for improved explicit monetary valuation of the economic impacts of invasive alien rodents to better disseminate the costbene t tradeoffs of addressing this global problem.

Research and management implications
In light of the evident knowledge gaps currently impairing our quantitative understanding of the economic costs of invasive rodents, we suggest efforts need to be targeted at multiple scales towards currently under-reported regions, taxa and sectors. In addition to multiscale studies, we stress the need for more accurate and standardized economic estimations in order to improve cost reporting following recommendations from Diagne et al. (2021). A key insight from these is to provide cost estimates at the nest taxonomic resolution possible. In this study, 66% of the total estimated costs were associated with mixed rodent species. This means that there is no possibility to disentangle species-speci c contributions to this total cost, thereby limiting opportunities to set priorities and evaluate cost-effectiveness of management actions at the species level (Gruber et al. 2021). For instance, currently underestimated health costs are expected to dramatically increase as ongoing trends in land-use change and urbanization lead to ampli cation of the role of (invasive) rodents as important zoonotic reservoirs in many locations (Gibb et al. 2020;Hassell et al. 2021;Mendoza et al. 2020). Obtaining accurate estimates of the true magnitude of these health costs will be imperative for incentivizing control efforts targeting multiple invasive rodent hosts. Along with increased reporting of species-speci c costs, we strongly encourage separation of invasive versus native status in rodent impact assessments, rather than considering both species as pests versus non-pests. Producing this greater granularity across scales will enhance our understanding of the rodent impacts, but will also help to improve the effectiveness of rodent management actions (Diagne et al. 2021;Gruber et al. 2021). As a support, we showed that management costs predominated on islands generally, with damage costs more common in mainland areas. Islands supported disproportionately high levels of native species endemism and extinction risk, often partly as a result of invasive rodents . The higher prevalence of management spending on islands may thus represent expenditures for local conservation purposes -particularly the costs invested in pre-invasion prevention and detection methods (Bodey et al, submitted in this issue).
While it is possible that the relative lack of management expenditure in mainland areas might indicate cost-e cient actions at local scales, the evident ongoing temporal increase in damage costs over time suggests this is unlikely. Notably, our results highlight the apparent complete lack of pre-invasion surveillance costs in these mainland areas.
Consistent and accurate accounting of the economic costs of rodents is therefore integral to coordinated, e cient and sustainable management of rodent invasions and their impacts. Furthermore, comprehensive estimates of the true costs of invasive rodents is essential to raising awareness (of both authorities and citizens) of rodents' impact, and obtaining community buy-in to control and prevention actions. Given the crucial importance of invasive rodent management as a priority for national governments, communicating the magnitude of these impacts is critical to creating a supportive legislative, political and societal environment which will implement long-term policies on rodent invasions (Novoa et al. 2017;Adamjy et al., 2020). Ultimately, this would help to design locally adapted -and thus sustainable -management strategies that account for the economic and societal realities (e.g. implementation of ecologically-based rodent management (EBRM) approaches with local communities; (Constant et al. 2020). This is particularly critical in low-and middle-income countries where economic resources are scarce, and societal concerns are dominated by food and health security (Crowley et al. 2017;Evans et al. 2018). Given the societal di culties and costs involved in minimizing the impacts of established invasive rodents, our results demonstrate the urgent, global need for increased policy development and effective measures to prevent further rodent invasions worldwide. Therefore, we encourage efforts to improve the e ciency of management actions through closer science-society interactions (Novoa et al. 2018), which should ultimately involve sustainable partnerships and interactions within/between local actors (biodiversity managers, funders and directly-impacted people, political leaders, socio-economic stakeholders) and scientists from different elds (e.g. economists, sociologists, biomedical and data scientists). Whether they are long term commensals of humans (rats, mice), invaders of speci c habitats (beavers, muskrats, coypus) or newly invasive from exotic pets (squirrels, dormice), invasive rodents remain relatively inconspicuous. Yet, they are particularly widespread and ubiquitous. We showed here that the small fraction of their impact that has been monetized is su cient to warrant much more focus on this invasive group. Online Resource 1 Data considered in this study on the economic costs of invasive alien rodents. The spreadsheets are the following: 'Descriptors' provides full de nition and details about the descriptive columns used in InvaCost as well as those added for the purposes of our analyses; 'Suitable subset' contains the raw cost entries pertaining to the order Rodentia in the original InvaCost database (version 3.0; complete database available at); 'Expanded subset' contains the annualized cost entries following data expansion through the invacost package (Leroy et al. 2020); 'Conservative subset' is the most robust subset of the 'Expanded subset' obtained after keeping only the observed ("Implementation" column) and high ("Method reliability" column) cost entries.
Online Resource 2 Distribution of cost entries and estimates recorded in our original subset according to their reliability (high versus low) and their implementation (potential versus observed). All details on the descriptive elds considered are provided in the Online Resource 1.
Online Resource 3 Temporal trends of the cost estimates of invasive alien rodents from our study. We considered (a) the original subset, (b) the conservative subset and (c) the non-conservative subset (see Fig. 1 and Online Resource 1 for further details on the subset and ltering steps). In (a), trend is described separately for potential and observed cost entries (see "Implementation" column; Online Resource 1). In (b) and (c), trends are described separately for damage, management and mixed costs (see "Type of cost merged" column; Online Resource 1). Costs are provided in 2017 US$ dollars. The horizontal dotted lines represent annual averages over the entire time period, solid bars represent 10year means and lled circles represent annual costs scaled by size to match the number of entries.
Online Resource 4 Summary of the cost distribution per invasive alien rodent taxon, impacted sector, geographic region and type of costs from the conservative subset used in our analyses. Costs are provided in 2017-equivalent US$ million. The number of annualized cost entries is provided in parenthesis. All details on the descriptive elds considered are provided in the Online Resource 1.
Declarations Figure 1 Work ow depicting the data collection and ltering process. Thirteen cost entries were excluded to limit dubious data (9 cost entries) and potential spatial overlaps (3 cost entries provided at the continental scale and 1 cost entry provided at the global scale) when generating the suitable subset. The expanded subset was obtained through the 'expansion' of the suitable subset using the 'invacost' R package (Leroy et al. 2020). The criteria used for generating the conservative subset were based on the descriptive elds of the InvaCost database (Online Resource 1, Tab Descriptors), i.e. the 'Implementation' (observed versus potential costs) and 'Method_reliability' (high versus low-reliability costs). The number of taxa includes both individual species and unde ned species aggregated at the genus level.

Figure 2
Temporal trends in rodent invasion costs considering a range of statistical models. Ordinary least squares (OLS), robust regression, generalised additive model (GAM), multivariate adaptive regression splines (MARS) and quantile regressions. Shaded areas are 95 % con dence intervals; points represent annual totals. The y axes are on a log10 scale, and are scaled separately among subplots. This is a list of supplementary les associated with this preprint. Click to download.