Alien species constitute one of the main threats to freshwater ecosystems, negatively impacting biodiversity, economy, biosecurity and ecosystem services. Predicting the arrival and spread of alien species is of paramount importance to prevent new introductions and control the expansion and establishment of already introduced species. We modelled the distribution of four freshwater invaders in Great Britain, using environmental and anthropogenic predictors, to help focus management actions. The species grouped different taxa including signal crayfish (Pacifastacus leniusculus), the marsh frog (Pelophylax ridibundus), the red-eared slider (Trachemys scripta) and the pike-perch (Sander lucioperca). The modelling approach accounted for methodological limitations and implemented two evaluations, a temporal evaluation using data corresponding to 70% of the oldest records to calibrate models and the remaining 30% for evaluation using various performance metrics (the common AUC, TSS and also null models) and an independent evaluation using the most recent range expansion of the species in the last six years. The distribution of the species was facilitated by multiple environmental and anthropogenic predictors. Road density was the second most important predictor of the occurrence of signal crayfish and red-eared slider preceded by the distance to ports and isothermality for each species respectively. Human population density was the most important predictor of marsh frog presence whereas pike-perch was mostly related to the proximity of boat ramps and precipitation regimes. Our distribution models were accurate and predicted the most recent range expansion of all of the species, highlighting their usefulness for preventing alien species spread and the value of using historical projections, usually available for non-native species, to calibrate and evaluate Species Distribution Models.

aquatic non-native species, human ecology, introduction pathways, management, species distribution models, forecasting

Freshwater ecosystems are among the most endangered ecosystems on the planet (Sala et al. 2000; Tickner et al. 2020), yet they tend to be neglected by conservation efforts compared to terrestrial or marine ecosystems (Darwall et al. 2011). Biological invasions are of special concern in freshwater ecosystems, threatening the native biodiversity by disrupting trophic links (Gallardo et al. 2016) or competing with species of commercial interest (Cardona et al. 2008), and also impacting the economy (Cuthbert et al. 2021) and human health (Tricarico et al. 2017; Mazza and Tricarico 2018). Information about the dispersal patterns and vectors of invasion is crucial to predict the invasion process of different species, for which appropriate models are essential (Barbet-Massin et al. 2018). Once an alien species has arrived in a new area or country, secondary dispersal may occur naturally or via human transport (Johnson et al. 2001; Vander Zanden et al. 2008), but for many organisms assessing the relative contribution of each of these factors is difficult (Huang et al. 2017). Therefore, understanding this post-introduction range expansion is critical to control the further spread of introduced species (Mallez and McCartney 2018; Robertson et al. 2020).

In addition to the importance of climate on the distribution of alien freshwater species (Rahel and Olden 2008), recent research has shown that anthropogenic factors can affect the range expansion more than environmental variables alone (Faurby and Araújo 2018; Chapman et al. 2020). For example, the number of invasive species tends to be higher outside protected areas than inside them, possibly reflecting the role of anthropogenic pressures (Gallardo et al. 2015; Moustakas et al. 2018). Multiple intentional introductions in distant localities are also common and can increase invasion success (Signorile et al. 2016; Huang et al. 2017), for example, most alien fishes in Great Britain have increased their numbers in the last decade (Copp et al. 2006) due to both accidental and deliberate releases (Zięba et al. 2010). A similar pattern exists for alien molluscs (Sousa et al. 2014) and pet turtles such as the red-eared slider, whose high establishment success has been aided by repeated introductions around the globe (Bomford et al. 2008). Therefore, combining anthropogenic and environmental types of variables in ecological models will help to obtain accurate predictions that support cost-effective management solutions which are usually challenging at large spatial scales (Glen et al. 2013).

Species Distribution Modelling (SDM) is a valuable tool that uses spatially explicit variables to explain and predict the range expansion of alien species and provide options for management actions (Guisan et al. 2013). However, SDM reliability to correctly predict the spread of the species has been questioned based on limitations such as lack of equilibrium of the species in the invaded ecosystems (Václavík and Meentemeyer 2012; Mainali et al. 2015), data restrictions and errors (Václavík and Meentemeyer 2009; Barbet‐Massin et al. 2012), shortcomings of the performance metrics (Jiménez‐Valverde 2012) or difficulties in the evaluation process (Warren et al. 2020). In fact, the evaluation step in SDM is key and requires independent distribution data which is, for most species rarely available (West et al. 2016), thus discrimination metrics are obtained through a data splitting approach [usually cross-validation (Helmstetter et al. 2021)] creating a training and testing dataset with randomly selected records from the total occurrence set. This type of evaluation has some limitations and has been criticised due to overestimation (Lobo et al. 2008). Thus, the desirable way to reliably test a model is by gathering independent data to evaluate predictions. Monitoring programs of alien species provide a great opportunity for a more independent evaluation by boosting data availability over space and time, helping to improve the evaluation approach (Barbet-Massin et al. 2018).

To overcome some of these limitations and investigate the range dynamics of alien species, we applied SDM including multiple predictors related to vectors of introduction and dispersal and correcting for potential error sources (Sillero and Barbosa 2021). We examined the distribution of four aquatic alien species from four different taxonomic groups (fish, reptiles, amphibians and crustaceans) deliberately introduced through different routes (i.e., ornamental trade, aquaculture, pet release). The aims were: i) to identify the main drivers of dispersal of alien freshwater species in Great Britain using detailed predictors that can give light into the invasion process and ii) to assess the power of SDMs to predict the invasion process using two types of evaluations including: a) the common cross-validation approach but using a temporal data splitting instead of random selection which is more realistic to recreate the invasion process and b) an independent data evaluation using the range shift of the species with the absolute newest records (during the last 6 years) to retrospectively predict current invasion patterns.

Our models were highly capable to predict the last range expansion of the species (last six years). Independent data for evaluation is desirable to measure the ability of the predicted areas to be occupied. Model calibration and evaluation 1 aimed to replicate the standard practice in SDM evaluation (cross-validation using 70% and 30% of the data for training and testing, respectively) but splitting datasets using the chronological information of the range expansion instead of a random selection. Our models provided better than random models but with relatively low values of performance. The reason for low values in model performance arose from the sorting bias correction. Spatial sorting bias is known to inflate evaluation metrics (Sillero and Barbosa 2021; Illanas et al. 2022) and generate model overestimation (Veloz 2009; Hijmans 2012). Therefore, sorting bias correction overall reduced model performance since uncorrected models showed a strong bias in their predicted species patterns (Hertzog et al. 2014; Freeman et al. 2022). Nevertheless, the most relevant evaluation in this study is the one based on the real range expansion of the species after 6 years (evaluation 2) showing that models´ predictions were highly accurate after independent data validation. This change in the predictive ability is in agreement with Hertzog et al. (2014) that reported average differences in the AUC between sorting bias uncorrected and corrected models of 0.4, and the model with the lowest AUC but that was bias corrected, obtained the best performance after independent field data evaluation.

The distribution of alien species is affected by both climatic and anthropogenic variables. Human population density can be an indicator of non‐native species propagule pressure (Copp et al. 2010), as confirmed by our results for marsh frog, pike-perch and red-eared slider occurrences, all favoured by high population density. Population density was the second most important predictor for the marsh frog distribution preceded by the distance to the introduction point, which indicates that the most populated areas not only promote their introduction but might also facilitate their establishment and expansion, for example via garden ponds (Vimercati et al. 2017), increasing the connectivity between populations in the absence of streams (Atobe et al. 2014). Population density as a predictor should be investigated in detail for management purposes, to identify the main points of entry or activities, as per se does not provide specific information on the activities or places that need to be regulated. Population density and the Human Footprint Index seem to shape the spatial distribution of biodiversity and also the presence of invasive species (Di Marco et al. 2013; Aronson et al. 2014; Gallardo et al. 2015; Dimitrakopoulos et al. 2017) which highlights the need to identify the specific vectors of secondary spread, such as flows of people, cars, trains or boats (Gilbert et al. 2004; Rodríguez-Rey et al. 2021) or defining spatial proxies of human activities promoting species invasion (Chapman et al. 2020). This can be achieved by prioritising risk assessment approaches that rely less on invasion history (e.g., historical sources of species) and focus more on environmental change and the consequences of expanding trade routes or aquatic activities, that increase source pools of alien species (Seebens et al. 2018). The release of animals to the wild is more frequent in urban than in rural areas (Olden et al. 2005), a good example being the red-slider, which was in our study particularly abundant close to cities with more than 50k inhabitants. Yet, although urban areas host many alien species, and act as foci from which they can spread and invade surrounding natural areas (McLean et al. 2017), most of the distribution patterns of the species in our study were unrelated to distance to cities.

Road density and proximity to garden centres favoured the spread of the red-slider suggesting that this species was introduced deliberately as pets (Copp et al. 2005) in the areas most accessible to people (Hulme 2009). Distance to garden centres has previously been found to promote the introduction of imported turtles, which increases propagule pressure (Duggan et al. 2006; Copp et al. 2010). Another species potentially favoured by road density was the signal crayfish, whose presence was also driven by climatic conditions. Its ability to adapt to multiple climatic conditions has allowed this species to adapt to areas with higher temperature and precipitation seasonality, as well as to large rivers at low altitudes, which tends to facilitate the invasion of freshwater species (Gebauer et al. 2018). However, altitude might limit its distribution due to high flows associated with steep gradients in headwaters. This species tends to be located far from ports pointing to the species’ tolerance to water salinity. Although ports are also located in estuarine waters, signal crayfish can only survive with salinity conditions below 14 PSU (Holdich et al. 1997) rarely common in British estuaries (van der Meer et al. 2016; Ownsworth et al. 2019).

Precipitation seasonality had a negative effect on pike-perch´s presence, which indicates that this species is unable to cope with variable flow regimens. In fact, for this species, seasonal weather patterns have been reported to affect spawning success, egg survival and post-hatching survival (Kanno et al. 2016) and changes in flow reduce fish recruitment (Dutterer et al. 2013). High precipitation in the warmest months benefited pike-perch’s presence which might be explained by the big size of the species, originally from large streams, that might be more affected than smaller fish by decreases in flow and habitat availability through an increase in competition (Rolls et al. 2012). Also, pike-perch prefers warm water which is why increasing temperature in the wettest months increased the presence of the species (Keskinen 2008). The influence of environmental drivers in the distribution of pike-perch is also reflected by its presence in less populated areas and far from garden centres, which might reflect a preference for more natural areas.

Most of the alien species we studied were more likely to be found closer to the places where they were first introduced, especially the marsh frog. This might be due to their limited ability to disperse due to climatic reasons. In this sense, further analysis is required to account for the type of dispersal followed by the different taxa. According to the importance of this predictor (i.e., “distance to the first records”), implementing management on areas surrounding the locations of the first introduction resulted highly important for all study species and supports the fact that effective early detection will facilitate the first management steps (Muha et al. 2017). Monitoring alien species over space and time provides data to develop time independent evaluations of the species´ models, which is one of the biggest limitations in SDM (Araújo et al. 2005) due to the dependency between training and testing data, that leads to prediction errors (Roberts et al. 2017; Liu et al. 2020). The modelling procedure implemented in this study used the spatial records of the species at earlier stages of invasion to calibrate the models and recent invasion records to evaluate them, can be applied with promising results (Barbet-Massin et al. 2018). In addition, we further conducted another evaluation using the range expansion in the last six years to realistically evaluate the predictive capacity of our models with independent data which constitutes a novel step in SDM for alien species. For all the species with a range expansion, the models were accurate in terms of AUC and correctly predicted the newly occupied localities, therefore, highlighting the usefulness of this approach.

Considering the most important anthropogenic predictors driving each species expansion together with the predictive risk maps (i.e., presented suitability maps, which help to identify those localities most vulnerable to be invaded next) can guide decision making to allocate resources and prioritising management actions to prevent the arrival of the species to new locations, especially to those places of conservation importance, with endangered species or in protected areas (Johovic et al. 2020), with biosecurity risks (Gallardo et al. 2022) or where high economic impacts are expected (Haubrock et al. 2021). Predicting species distributions to prioritise management is fundamental (Guillera‐Arroita et al. 2015; Wood et al. 2018) and non-native species monitoring in many regions promote the availability of records for a convenient evaluation using SDM. Future steps should aim for detecting more specific human activities at spatial scales and with a higher impact to expand alien species.

This study was funded by the call “Ayudas para la realización de proyectos de investigación de la Universidad de Alcalá” (PIUAH22/CC-054) to M.R-R, and Marie Sklodowska-Curie ITN (AQUAINVAD-ED; Grant no. 642197) to S.C. M.R-R was supported by “Maria Zambrano” fellowship from the Spanish Ministry of Universities and Next Generation-EU.

M.R-R. and C.G.L conceived the idea. M.R-R. compiled and curated the data and M.R-R. and C.G.L. analysed the data. All authors interpreted the outputs, contributed to manuscript writing, gave approval for publication and agree to be accountable for any question related to this work.

All data used is derived from public domain resources. A reference script is available as supporting information.

We are grateful to the anonymous reviewers for their constructive and helpful comments on this manuscript. This paper was a contribution to the RIP4FISH (PIUAH22/CC-054) project financed by the University of Alcalá.