Research Article |
Corresponding author: Anna Schertler ( anna.schertler@univie.ac.at ) Academic editor: Ingolf Kühn
© 2020 Anna Schertler, Wolfgang Rabitsch, Dietmar Moser, Johannes Wessely, Franz Essl.
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.
Citation:
Schertler A, Rabitsch W, Moser D, Wessely J, Essl F (2020) The potential current distribution of the coypu (Myocastor coypus) in Europe and climate change induced shifts in the near future. NeoBiota 58: 129-160. https://doi.org/10.3897/neobiota.58.33118
|
The coypu (Myocastor coypus) is a semi-aquatic rodent native to South America which has become invasive in Europe and other parts of the world. Although recently listed as species of European Union concern in the EU Invasive Alien Species Regulation, an analysis of the current European occurrence and of its potential current and future distribution was missing yet. We collected 24,232 coypu records (corresponding to 25,534 grid cells at 5 × 5 km) between 1980 and 2018 from a range of sources and 28 European countries and analysed them spatiotemporally, categorising them into persistence levels. Using logistic regression, we constructed consensus predictions across all persistence levels to depict the potential current distribution of the coypu in Europe and its change under four different climate scenarios for 2041–2060. From all presence grid cells, 45.5% showed at least early signs of establishment (records temporally covering a minimum of one generation length, i.e. 5 years), whereas 9.8% were considered as containing established populations (i.e. three generation lengths of continuous coverage). The mean temperature of the warmest quarter (bio10), mean diurnal temperature range (bio2) and the minimum temperature of the coldest month (bio6) were the most important of the analysed predictors. In total, 42.9% of the study area are classified as suitable under current climatic conditions, of which 72.6% are to current knowledge yet unoccupied; therefore, we show that the coypu has, by far, not yet reached all potentially suitable regions in Europe. Those cover most of temperate Europe (Atlantic, Continental and Pannonian biogeographic region), as well as the coastal regions of the Mediterranean and the Black Sea. A comparison of the suitable and occupied areas showed that none of the affected countries has reached saturation by now. Under climate change scenarios, suitable areas will slightly shift towards Northern regions, while a general decrease in suitability is predicted for Southern and Central Europe (overall decrease of suitable areas 2–8% depending on the scenario). Nevertheless, most regions that are currently suitable for coypus are likely to be so in the future. We highlight the need to further investigate upper temperature limits in order to properly interpret future climatic suitability for the coypu in Southern Europe. Based on our results, we identify regions that are most at risk for future invasions and provide management recommendations. We hope that this study will help to improve the allocation of efforts for future coypu research and contribute to harmonised management, which is essential to reduce negative impacts of the coypu and to prevent further spread in Europe.
biological invasions, climate change, consensus prediction, invasive alien species management, nutria, species distribution modelling, vertebrate
Invasive alien species, i.e. species introduced to areas outside their native range that have become successfully established, spread and cause substantial impacts on the new environment (
One prominent example, even included in the list of “100 of the World’s Worst Invasive Alien Species” (
Negative impacts of the coypu are mainly due to its burrowing activity and feeding behaviour and include undermining of flood protection structures, such as river banks and dykes and therefore increased risk of floods, as well as agricultural damage, mainly on corn and sugar beet (
The coypu is an opportunistic herbivore, preferably inhabiting slow-flowing or standing water bodies that are rich in hydrophytes, reeds and riparian vegetation, as well as wetland areas and swamps in lowlands (
Nowadays, the coypu is established in many European regions (
However, successful management requires an adequate understanding of the ecology and behaviour of the targeted species (
In the light of the urgent need of a harmonised coypu management, here we provide such an assessment for Europe. Specifically, we reconstruct the recent spread within the last decades and the current distribution of the coypu and group the occurrence data into different persistence-categories to identify regions that are suitable for permanent occurrence of coypus. Further, by using a consensus approach, we predict its potential current distribution and analyse to what extent suitable regions are not yet invaded. Finally, we investigate which climatic, land cover and socioeconomic variables influence coypu occurrence and model the potential future distribution under four different climate change scenarios. Based on our results, we identify regions that are most at risk for future invasions and provide management recommendations.
The study region includes most parts of the European mainland and the larger islands (excluding only European Russia, Ukraine, Belarus and Cyprus) (Fig.
Coypu occurrence records from 1980 to 2018 in Europe. The decade-wise accumulation of records is depicted on the left side, with records of the respective decade in black and records of previous decades shown in grey. On the summary map (right side), records in Great Britain are displayed in red, as the coypu is officially eradicated (see
The raw occurrence data were prepared and quality-checked prior to analyses. If a source described the occurrence of coypu over several decades, the record was split into one record per decade. Only records that contained geographic information and approximate sampling date were considered in this study. Records that were lacking coordinates, but contained an unambiguous locality description, were georeferenced, either using the point-radius method (estimating coordinates and an uncertainty radius according to the precision of the locality description) or the shape method (assigning a geographic shape that represents the uncertainty) (
For further analysis, records that were missing essential information (i.e. no georeference), putative duplicates (i.e. records with same year and coordinates or locality description), as well as records exceeding an uncertainty radius of 10 km in areas where more accurate records were available, were discarded. This resulted in a final dataset consisting of 24,232 coypu records between 1980 and 2018 across 28 European countries containing year, uncertainty estimate and coordinates (Fig.
We transformed those presence records (inclusive uncertainty buffer) to a grid of 5 × 5 km resolution (temporal resolution: one year) to reduce the effect of pseudo-replication (i.e. artificial inflation of the sample size due to intensively-sampled regions or non-detected duplicates in different datasets). Grid cells that showed only marginal overlap with buffered presences (< 2.5% of the grid cell area) were not defined as presences, to avoid area inflation. After discarding those, 25,534 grid cells (about 12.6% of all grid cells throughout the study area) were defined as presence grid cells and were used for further analysis (see workflow scheme, Fig.
Workflow used for species distribution modelling. Presence data was transformed to a raster grid (5 × 5 km), spatiotemporally analysed, accordingly grouped into five levels of persistence and then combined with environmental data and generated absences (using four different sampling strategies) to fit logistic regression models. The absence sampling strategy that derived the best evaluation measures was used for predicting suitability under current (1979–2013) and future climatic conditions after going through a model selection and averaging procedure.
On this basis, we created sub-datasets by stepwise exclusion of the lowest level of persistence (those sub-datasets are hereafter called persistence levels and abbreviated as indicated by the bold words: all grid cells = 25,534, multiple records per grid cell = 15,078, multiple records per grid cell that cover at least one GL = 11,619, multiple records per grid cell that cover at least two GL = 5,145, established populations only = 2,505).
Environmental predictors included bioclimatic variables from the CHELSA database (
Environmental predictor variables. All predictors were rescaled to a 5 × 5 km raster resolution (bilinear interpolation) and standardised (scaled to a mean of 0 and a standard deviation of 1).
Predictor | Description | Temporal coverage | Source |
---|---|---|---|
Bio2 | Mean Diurnal Range [°C] | 1979–2013 | CHELSA (Karger et al. 2017) |
Bio6 | Min Temperature of Coldest Month [°C*10] | ||
Bio10 | Mean Temperature of Warmest Quarter [°C*10] | ||
Bio15 | Precipitation Seasonality [CV] | ||
Bio17 | Precipitation of Driest Quarter [mm/quarter] | ||
Hilliness | Std. Dev. of m a.s.l./cell | 2000 | EEA |
Pop. density | Mean human population density [inhabitants/km²] log-transformed | 2011 | GEOSTAT v.2.0.1. / Eurostat, EFGS |
Distance Settlement | Euclidean distance to the next grid cell containing artificial surfaces | 2012 | CORINE LANDCOVER, vers. 18.5.1 |
Agriculture | Agricultural surfaces [% counts/cell] | ||
Wetlands | Wetlands surfaces [% counts/cell] | ||
Waterbodies | Water bodies’ surfaces [% counts/cell] | ||
Shores | Total shoreline (rivers and lakes) [m/ha] | 1990–2006 | ECRINS v.1.1 / EEA |
Additionally, we used qualitative information on biogeographic region (
Two IPCC (Intergovernmental Panel on Climate Change) climate change scenarios were selected to model coypu response to a changing climate by the mid-21st century (2041–2060). One represents medium (RCP 4.5; Representative Concentration Pathway) and one represents severe climate change (RCP 8.5) by depicting the different approximate radiative forcing in comparison to the pre-industrial state (i.e. + 4.5 and + 8.5 W/m²) (
To model the range of potential distribution under current climate and under climate change, we used logistic regression, a generalised linear modelling (GLM) technique that is widely used for predicting species distributions (
We combined each of the four different absence sampling strategies (background full, background restricted, pseudo-absence full, pseudo-absence restricted; see Suppl. material
As pseudo-absences which were generated across the whole study area extent (“pseudo-absence full”) consistently derived the best evaluation values (Suppl. material
We assessed the goodness of fit for the full models (including linear and quadratic terms for all variables) of all datasets and of 25 random subsets per persistence level. Those were created by drawing random presence subsamples that equal the size of the according persistence level to check for sampling size effects that might occur. For the model evaluation, we compared a set of commonly-used measures (
Variable importance was measured for each predictor by evaluating the mean drop in explained deviance caused by removal of the respective predictor. Finally, true positives and omission errors were mapped to reveal sensitivity issues and spatial patterns in model performance (Suppl. material
The full models’ quality notably increased with increasing level of persistence and this effect could be clearly distinguished from sample size effects, when comparing the evaluation measures with those of the random subsets (Fig.
Comparison of model performance for the persistence levels (with pseudo-absences for the full extent) versus random subsets. The boxplots show the results of the fivefold-cross validation for the persistence levels (blue) and the according random subsets (orange).
We assumed grid cells that contain long-term occurrences to be more informative than those where coypu occurrence was only registered once. Both approaches (using all data vs. subsets) in its extremes may incorporate biases (i.e. all presence data will more likely include non-persistent occurrences and false identifications, whereas grid cells that show long-term occurrence might underestimate the area of a still-spreading alien species and comprise historical effects of propagule pressure due to regional differences in fur farming intensity, as well as effects of uneven data availability across regions). To balance those possible biases and reduce uncertainty, we combined the resulting predictions of probability of occurrence for all persistence levels and created a consensus prediction by simply calculating the mean probability of occurrence per grid cell, depicting the overall agreement of the averaged models.
Further, consensus forecasts for all climate change scenarios were computed. The change in probability of occurrence was assessed by comparing the number of cells that were classified as suitable under current climate and under climate change scenarios and by subtracting the probabilities of occurrence of current from future predictions. To obtain the agreement between binary models, we used the sum of predicted presences per cell across all averaged models, with a high value meaning high agreement.
Finally, we used the resulting predictions to define priority regions for surveillance and management by creating a risk map. Grid cells that 1) show high probability of occurrence in the consensus prediction and 2) areas adjacent to already known recent occurrences, are deemed to be particularly susceptible to invasion by coypus, due to short colonisation distances. Thus, to incorporate dispersal constraints and to account for proximity to known occurrence, a weighting matrix was computed, by summing up weighted inverse Euclidean distance classes per decade for each cell (Suppl. material
Statistical analyses were conducted and maps were produced using ArcGIS 10.5.1 (
In total, 24,232 coypu presence records (corresponding to 25,534 grid cells at 5 × 5 km) were collected across 28 countries. The spatiotemporal analysis of presence grid cells shows centres of documented long-term occurrence in Czech Republic, France, Germany, Italy and the Netherlands. Of all presence grid cells, 45.5% (corresponding to 20 countries) show at least early signs of establishment (i.e. had multiple records that covered one generation length as a minimum; of those 20.1% have been covered by at least two generation lengths and 9.8% of the grid cells (corresponding to 10 countries) show spatially-explicit evidence for long-term persistence (i.e. established populations) with coypus being present over a period of at least 15 years (Fig.
Persistence levels of presence grid cells as derived from the spatiotemporal analysis. Each grid cell that intersects at least one record of coypu presence between 1980 and 2018 is coloured according to the maximum derived persistence level: 1) single record, 2) multiple records, 3) one generation length (multiple records covering at least 5 years), 4) two generation lengths (multiple records covering at least 10 years), 5) established (multiple records covering at least 15 years). One generation length is assumed to be 5 years, following Ojeda et al. (2016).
Model quality increased with higher levels of persistence, with mean AUC values ranging from 0.90 (all) to 0.96 (established) indicating excellent discrimination ability across all averaged models and TSS values ranging from 0.61 to 0.79 which can be interpreted as good to excellent agreement between training and test data (
Evaluation statistics of the averaged models for all five levels of persistence. For computation of the TSS, the sensitivity was set to 0.95.
AUC | D² adj [%] | TSS (Sens = 0.95) | Specificity | |
---|---|---|---|---|
all | 0.90 | 43.0 | 0.61 | 0.65 |
multiple | 0.92 | 50.3 | 0.68 | 0.73 |
one GL | 0.93 | 53.1 | 0.71 | 0.76 |
two GL | 0.95 | 58.6 | 0.76 | 0.81 |
established | 0.96 | 57.5 | 0.79 | 0.84 |
Between two to four top models were averaged for the persistence levels, with the full model always being included. Only land cover variables were excluded (‘shores’, ‘water-bodies’) and none of them was excluded across all persistence levels (for model weightings and ΔAICc, see Suppl. material
Importance of predictor variables for the averaged final models. The importance is measured as the mean drop in explained deviance (D²) upon removal of the respective predictor. For descriptions of the predictor variables, see Table
Plotting of the omission errors of the binary consensus prediction revealed that those mostly occurred at the range margins of the currently known European distribution, especially towards Southern Europe and mountainous areas (Suppl. material
The consensus map for current climatic conditions shows that, currently, large parts of Europe have a high probability of coypu occurrence (Fig.
Consensus predictions of the probability of occurrence across the study area under current climatic conditions (years 1979–2013). a Mean probability of occurrence across the final averaged models of all persistence levels. b Binary classification of suitable and unsuitable grid cells after applying a threshold corresponding to 0.95 sensitivity (= 0.16).
All four climate change scenarios show substantial shifts in predicted habitat suitability until the mid-21st century (2041–2060) (Fig.
Future predictions. Agreement between averaged models for projected probability of occurrence in the mid-21st century under two climate change scenarios (medium climate change: RCP 4.5; severe climate change: RCP 8.5) combined with two different global circulation models (HadGEM1-A0, CESM1-BGC, displayed as number of models predicting presence (left side) and net change in occurrence probability compared to the current climatic situation (right side).
The risk map (Fig.
Risk map, highlighting regions potentially prone to invasion, i.e. with high probability of occurrence under current climate and adjacent to known recent occurrences of coypu. Presence grid cells are shown in black. Great Britain was excluded as the coypu is officially eradicated (Gosling & Baker, 1989).
This study confirms and substantially expands the overview of
The country-wise percentage of suitable grid cells under current climate (brown bars) and grid cells containing presences (red bars). Countries are marked according to the maximum persistence level. Countries with no suitable areas and occurrences are not shown, as well as microstates not covering a whole grid cell. Alphabetically ordered country abbreviations with corresponding percentages of suitable grid cells and occupied cells in parentheses: AL : Albania (4.7/0.2), AT : Austria (30.0/6.3), BA : Bosnia and Herzegovina (48.2/-), BE : Belgium (99.9/15.6), BG : Bulgaria (63.3/2.8), CH : Switzerland (43.2/3.9), CZ : Czech Republic (50.5/26.0), DE : Germany (95.7/61.0), DK : Denmark (70.1/7.7), ES : Spain (17.3/0.6), FI : Finland (-/0.1), FR : France (89.7/55.2), GR : Greece (5.5/2.8), HR : Croatia (89.6/0.8), HU : Hungary (93.6/1.8), IE : Ireland (39.5/2.5), IT : Italy (55.4/10.0), LI : Liechtenstein (33.3/-), LT : Lithuania (29.3/-), LU : Luxembourg (100/19.8), LV : Latvia (4.5/-), ME : Montenegro (14.7/1.1), MK : Macedonia (16.5/2.2), NL : Netherlands (98.3/35.5), NO : Norway (0.8/-), PL : Poland (79.8/0.7), PT : Portugal (12.6/-), RO : Romania (38.6/0.2), RS : Serbia (63.2/0.5), SE : Sweden (6.0/<0.1), SI : Slovenia (74.1/4.1), SK : Slovakia (53.2/6.1), TR : Turkey* (32.4/2.8), UK : United Kingdom (50.3/6.4), XK : Kosovo (64.2/-).*) only the area of the European part of Turkey is considered.
All four climate change scenarios used in this study predicted a slight to moderate decrease of suitable area (from 42.9% under current climate to between 34.7% and 40.9%). This decline is caused by a loss of suitable habitats in the southern parts of Europe, which is not fully compensated for by increasing suitability at higher latitudes (Fig.
The recent occurrence of the coypu in Ireland caused the first Species Alert issued by a European Union Member State under the EU Regulation on Invasive Alien Species. Although those areas were only partly predicted correctly, a considerable area of Ireland is classified as suitable by our predictions. In line with our study,
While our predictions classify most of the Atlantic, Continental, Black Sea and Pannonian biogeographic regions as suitable, this is not the case for the Alpine biogeographic region. Therefore, this study is in agreement with others that have classified the coypu as a typical lowland species and implies that mountain regions act as effective dispersal barriers on a regional scale (
Our results highlight the importance of temperature-related climatic variables, such as the mean temperature of the warmest quarter, the mean diurnal temperature range and the minimum temperature of the coldest month as being essential in shaping habitat suitability for coypus (Fig.
Currently, urban coypu populations, fed by humans and profiting from mild urban climate and, in some cases, the thermal pollution of rivers, clearly demonstrate the consequences of high reproduction rates coupled with lowered mortality and enhanced resource availability for population densities (
Although on a continental scale, climatic aspects are clearly of higher importance (
The assumption of an equilibrium between a population and its environment is typically violated during biological invasions, due to ongoing dispersal.
The SDMs performed well (Table
The majority of grid cells deemed suitable for coypu under current climate or climate change are not yet colonised. Our results illustrate the urgent need to not only improve management measures in areas with persisting populations, but also find strategies to prevent or reduce further spread as the costs of early intervention are much smaller than control of established populations (
Baroch and Hafner (2002) argue that, in the case of low population densities, impacts of coypus in general are rather minor. Given that the coypu is already fairly widely distributed, management that minimises population density and therefore negative economic and environmental impacts should be the aim. As there are several hotspots of coypu occurrence covering more than one country, there is a need for international collaboration to coordinate control measures on a metapopulation scale and prevent compensatory re-invasion from adjoining populations (
Accounting for coypu in hunting laws would allow integrating it as a wildlife resource and harvesting coypu for its meat and fur. Meat of wild coypus was shown to be low in fat and cholesterol, while rich in proteins (
Aside from direct control measures, another aspect of coypu management is the facilitation of winter survival and rise of reproduction rates by providing additional food sources. Wildlife feeding in or nearby settlements can induce rapid increases of coypu populations. Urban feeding sites and easily accessible agricultural areas may suffer from high coypu abundances (
It is well-established knowledge that the coypu causes substantial economic and environmental damage when occurring at high densities. Although cool-temperate climates were believed to keep coypus at low densities, in many parts of Europe numbers have increased strongly during the last decades (
Our study shows that the coypu has, by far, not yet reached all potentially suitable regions in Europe and further highlights the importance of clarifying its response to increasing temperatures and arid conditions as they are likely to increasingly occur in the near future under climate change. However, one must consider the shortcomings of predictions that are made on the basis of opportunistic records from various sources and of differing data quality. Sampling effort differs spatiotemporally across the study area and, although we considered the violated assumption of an equilibrium for taxa undergoing an invasion process (
We are especially grateful for the provision of coypu occurrence data across Europe by a large number of colleagues and institutions. The first author would like to thank all correspondents throughout Europe for insightful conversations and the provision of information on coypus in their countries. We are thankful for the constructive comments by Sandro Bertolino and Martin Damus. Finally, we highly appreciate funding by the Austrian Science Foundation FWF (Pr.nr.: I3757-B29).
Supplementary materials
Data type: tables and figures
Explanation note: Figure S1. Pairwise Pearson’s correlation coefficients for predictor variables. Figure S2. Absence sampling strategies. Figure S3. Predictive ability of the consensus prediction under current climatic conditions (1979–2013). Figure S4. Calculation of the weighting factor for the coypu invasion risk map. Table S1. Presence data sources included in this study, listed by country and the total number of records per country after data cleaning. Table S2. Top models (ΔAIC < 4) that have been used for averaging within the persistence levels and their according weight.