Integrating climate scenarios and advanced modeling to predict freshwater fish invasions: insights from Carassius species in Iran

Abstract

Freshwater ecosystems are increasingly imperiled by the dual pressures of biological invasions and climate change, necessitating robust predictive frameworks for effective management. This study integrates advanced ensemble machine learning (EML) within a species distribution modeling (SDM) framework to assess the current and future invasion risk of Carassius species (C. auratus, C. gibelio, and C. langsdorfii) across Iranian inland waters. A comprehensive dataset of 486 occurrence records was analyzed alongside eight rigorously selected environmental predictors encompassing climatic, topographical, hydrological, and anthropogenic variables. The BIOMOD2 R package facilitated the construction of an EML-based SDM, leveraging six algorithms weighted by AUC to maximize predictive accuracy. Model performance, evaluated via AUC and true skill statistic (TSS), demonstrated high discriminatory power. Projections under two CMIP6 climate scenarios (SSP 126 and SSP 585) reveal significant potential for range expansion and spatial redistribution of Carassius species, particularly under high-emission trajectories, highlighting increased invasion risks in ecologically sensitive basins. Variable importance analysis underscored the primacy of temperature, precipitation, terrain ruggedness, and human footprint in shaping invasion potential. Additionally, using kernel density estimation (KDE) analysis, the Caspian basin emerged as a critical invasion region for Carassius populations. These findings underscore the urgent need for targeted monitoring and management strategies and demonstrate the utility of EML-SDMs in anticipating biological invasions under global change. The integrative approach presented here provides a scalable framework for proactive biodiversity conservation and policy development in freshwater systems facing multifaceted anthropogenic threats and provides a replicable framework for forecasting biological invasions in other vulnerable freshwater systems.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-21958-8.

Introduction

Freshwater ecosystems play a crucial role in various aspects of human society. They serve as key focal points within the landscape and offer numerous cultural and ecological benefits by harbouring diverse biological life^1–4. The increasing challenge of balancing the water needs of a growing human population with maintaining ecological health is a primary reason why freshwater organisms are amongst the most vulnerable fauna worldwide^5–7. In recent years, non-native invasive fish species have emerged as a significant threat to biodiversity in freshwater environments, exacerbating the risks posed by habitat loss, fragmentation, hydrologic changes, climate variability, overexploitation, and pollution^8–10. This threat is compounded by the fact that not all introduced fish species become established; however, those that do can have significant ecological, evolutionary, and economic implications for their new ecosystems^11–13.

A key challenge in predicting habitats for non-native species is identifying factors differentiating successful invasions from unsuccessful ones, including environmental conditions and biotic interactions^14–16. For a species to establish, two conditions must be met: (i) suitable environmental conditions for survival and reproduction and (ii) manageable biotic interactions, such as competition and predation¹⁷. In some cases, using only environmental variables may suffice for predicting species distributions, especially when strong biotic interactions are absent^18–21. The direct response of freshwater fishes to many biological, physical, and chemical factors represents a signal correlated to the alteration or shifting in aquatic ecosystems, such as a climate shift^22,23. Climate change has been shown to affect freshwater biodiversity by altering species extinction rates, causing range shifts in species distribution, and improving the invasion success of non-native species^24,25. Therefore, understanding the factors and then using the new approach to determine what contributes to a species’ invasiveness is vital for effective management and conservation efforts.

Scientific risk assessments often underpin decision-making about the control and management of invasive non-native species, and species distribution models (SDMs) are increasingly seen as valuable tools for this^26–28. SDMs are powerful tools used to describe species’ suitability habitats and spatial occurrences and to address many ecological questions, including identifying sites with a high likelihood of observing rare/cryptic species^29,30, conservation strategies^31,32 or plans to control biological invasion under current and future climate scenarios³³. It has been firmly documented that the SDM approach could be a valuable proactive tool for distinguishing invasion potential for non-native species^34–37. Many statistical methods and algorithms are available to model the spatial distribution of a target species in the framework of species distribution models. Ensemble machine learning (EML) for species distribution modelling (SDM) is an advanced approach that uses multiple modelling techniques or algorithms to improve the accuracy and robustness of predictions regarding the geographical distribution of species^38–40. Instead of relying on a single model, ensemble methods combine the predictions from several models modelling how species distributions may shift under various climate scenarios.

The crucian carps, belonging to the genus Carassius, are freshwater fish species that are distributed both naturally and through human introduction across Eurasia^41–43. Their range extends from Portugal in the west to Japan in the east and from Siberia in the north to Vietnam in the south. Occurrences have been reported in regions beyond this landmass, including Australia and North America^44–50. Among these introduced species, members of the genus Carassius, which includes the Carassius auratus (common name: Goldfish), Carassius gibelio (common name: Gibel carp), and Carassius langsdorfii (common name: Ginbuna) have been identified as particularly non-native and invasive in various regions, including Iran^41,51,52. The ecological consequences of introducing species can be profound, leading to the displacement of native fish species, alterations in habitat structure, and disruptions in local food webs.

Given the extraordinary richness of fish species in Iran’s inland waters^53–56, and the increasing pressures from anthropogenic activities, understanding the factors and then using the new approach, like EML, to determine what contributes to the invasiveness of species such as Carassius, is vital for effective management and conservation efforts. Introducing non-native fish species into Iran’s inland waters has resulted in considerable ecological disturbances, posing a threat to the native biodiversity within the region’s aquatic ecosystems^57,58. These invasive species compete with indigenous fish for essential resources, frequently resulting in declines in native populations and disrupting the equilibrium of local food webs^59,60. Despite increasing awareness regarding these invasions, there is a notable deficiency in effective methodologies for predicting the potential invasiveness of newly introduced species in Iranian basins. This gap in predictive capacity impedes management strategies and highlights the critical need for research and monitoring efforts to mitigate the impacts of invasive fish species.

Specifically, in the current study, we explore an SDM framework combined with EML algorithms to (i) delineate the extant climatic niche and geographic distribution of three Carassius species in Iran, (ii) assess prospective alterations and invasion in their spatial occupancy in response to diverse projected climate change scenarios, and (iii) generate the inaugural species richness map for Carassius, thereby identifying critical regions warranting prioritized conservation interventions.

Materials and methods

Sampling area and Carassius species occurrences

This research focuses on Iran, which is situated in the Middle East. Iran boasts a stunningly diverse landscape stretching from the Caspian Sea’s northern shores to the Persian Gulf’s southern reaches^61,62. Covering about 1,648,195 km², it offers a wide variety of terrains^63,64. The country’s elevation varies significantly; while some areas sit at sea level along the coast, others rise dramatically, with peaks in the Alborz mountain range reaching over 5,600 m (Fig. 1). Iran’s average elevation is around 1,200 m above sea level, showcasing its rich and varied geography^65,66.

Fig. 1.

Sampling area and spatial distribution of the Carassius species occurrence points in Iran. the authors used ArcGIS software (Ver. 10.6) (https://desktop.arcgis.com/ zh- cn/quick- start guides/10.6/) to perform spatial analysis and create color-coded maps.

Between 2016 and 2025, the research team systematically collected and documented a total of 486 occurrences of Carassius species utilizing a variety of methodologies. These included direct field surveys (245 presence points), data sourced from the Global Biodiversity Information Facility (GBIF; https://www.gbif.org/) website (106 presence points), Brian Coad’s Ichthyology Site (23 presence points), and relevant published literature (112 presence points). To mitigate the uncertainty associated with the 486 presence points’ geographical locations, we verified and rectified the locations using the Google Earth Engine (https://earthengine.google.com/). To mitigate potential model performance biases that may arise from imbalances in presence and absence data (pseudo-absence points), we generated pseudo-absence points following the methodology delineated by Dar et al. (2021). Specifically, we produced 486 pseudo-absence points to correspond with the number of presence records^67,68. The study area was delineated through the construction of a convex hull that encompassed all sampled location points, incorporating a buffer zone of 5 km. This convex hull was generated utilizing the R package rangemap⁶⁹. Subsequently, we employed the CoordinateCleaner package⁷⁰ within R to eliminate records characterized by inaccurate spatial data.

Spatial thinning

A spatial thinning approach was applied to the occurrence data to address potential sampling bias, following the methodology outlined by Steen et al. (2021)⁷¹. The methodology involved a process where, for each species, only one occurrence record is kept within a specified minimum distance. This thinning distance is chosen to ensure that the remaining data points are more evenly distributed, providing a more accurate representation of the species’ true range. The specific thinning distances mentioned for the three species C. auratus at 1 km, C. gibelio at 1 km, and C. langsdorfii at 3 km indicate that for each species, any two remaining records will be at least that far apart. For example, for C. langsdorfii, if two records were originally found 2 km apart, one of them would be removed. This process results in a dataset that is less biased and more suitable for creating reliable species distribution models. The different distances used for each species reflect their unique ecological characteristics and distribution patterns. This differential application reflects the observed patterns in the data, where the minimum distances for presences were consistently less than those for absences, indicating that absences were both more frequent and more widely dispersed than presences. The thinning algorithm was executed over 50 iterations for each species, with the iteration yielding the highest standard deviation of longitude coordinates being selected for further analysis. The spatial thinning procedure was implemented using the R package spThin⁷².

Environmental variable selection

The environmental variables utilized in this study encompassed four categories of constraints influencing the distribution of Carassius species: climatic, topographical, hydrological, and anthropogenic factors (Table S1). All environmental data were projected into the WGS84 coordinate reference system with a resolution of 30s. To enhance the rigor of our analysis, we procured gridded data for bioclimatic variables (Bio1-Bio19) from the WorldClim Global Climate Database, which is available at https://www.worldclim.org. To examine the topographical influences on species distribution, we employed slope measurements alongside the Terrain Ruggedness Index (TRI), adhering to the methodological framework established (Riley et al. 1999). To access hydrological data related to rivers, including slope layers and river height and width, we used Global Runoff Data Centre (GRDC) (https://www.un-spider.org/links-and-resources/data-sources/global-runoff-data-centre-grdc-wmo). For the assessment of anthropogenic factors, a human settlement layer pertinent to the study area was sourced from the Wildlife Conservation Society (https://wcshumanfootprint.org/data-access).This settlement data was converted into raster format utilizing ArcGIS 10.6⁷³. To effectively integrate the ramifications of human activities into our modeling framework, we incorporated variables such as human population density, accessibility, and land transformation, as delineated by the human footprint surface developed⁷⁴. This comprehensive surface enabled us to quantitatively assess the anthropogenic activities’ environmental impacts.

All geospatial layers were standardized to align with the Universal Transverse Mercator (UTM) zone projection and were resampled to achieve a consistent spatial resolution of 1 km². A variance inflation factor (VIF) analysis assessed multicollinearity among the environmental variables⁷⁵. We established a threshold of VIF = 10, retaining only those environmental variables with VIF values below this threshold for inclusion in the modeling framework. As a result, eight variables were retained for inclusion in our predictive model: Annual Mean Temperature (Bio 1), Max Temperature of Warmest Month (Bio 5), Temperature Annual Range (Bio 7), Mean Temperature of Driest Quarter (Bio 9), Annual Precipitation (Bio 12), Precipitation of Driest Month (Bio14), Terrain Ruggedness Index (TRI), and Human Footprint (HF).

Modeling framework

We implemented a multi-model ensemble methodology in accordance with the framework established by Thuiller et al. 2009. The analytical process was conducted using the BIOMOD2 package⁷⁷ within R version 4.4.1⁷⁸. This modeling framework synergistically integrates outputs from various algorithms to reduce the uncertainties inherent to individual modeling techniques. By amalgamating these diverse results, our objective was to enhance the accuracy and reliability of our predictive outcomes^76,77. The ensemble model was formulated by calculating a weighted average of the area under the curve (AUC) values generated by an array of machine learning algorithms, including; MARS (Multivariate Adaptive Regression Splines; Leathwick et al., 2011⁷⁹), GLM (Generalized Linear Model⁸⁰, GBM (Gradient Boosting Machines⁸¹, RF (Random Forest⁸², ANN (Artificial Neural Network⁸³, GAM (Generalized Additive Models⁸⁴; and FDA (Functional Data Analysis⁸⁵. Then, we employed ten replicates for each model to ensure robustness in our analysis.

Model selection

Our study employed two prevalence-independent metrics for evaluating species distribution models based on environmental variables: the area under the receiver operating characteristic curve (AUC-ROC) and the true skill statistic (TSS). These metrics were selected for their appropriateness in assessing presence-absence predictions. AUC is a threshold-independent measure that effectively gauges the model’s discriminative capability between presence and absence⁸⁶. In contrast, TSS, which relies on a defined threshold, evaluates both presence and absence predictions, making it particularly advantageous for unbalanced datasets. This combination allows for a comprehensive assessment of model performance, as AUC quantifies discrimination while TSS accounts for false predictions, thus aligning closely with the characteristics of our dataset. While we acknowledge the inherent limitations of AUC and TSS, we emphasize their utility in evaluating overall model performance rather than focusing on individual prediction nuances. To derive the ensemble model from particular models, we applied the weighted average method outlined by⁷⁶, which involved calculating the weighted average of all individual model outputs. This consensus approach significantly enhances predictive reliability compared to particular models and other consensus methods⁸⁷. Additionally, to assess the impact of each environmental variable on the distribution model’s performance, we calculated average scores across all models, enabling us to determine the overall importance of each variable in shaping species distributions.

Model prediction: invasive under future climate conditions

In our investigation of future projections, we analyzed two greenhouse gas concentration scenarios, referred to as Shared Socio-economic Pathways (SSPs), in accordance with the latest Coupled Model Intercomparison Project Phase 6 (CMIP6) framework. Our focus was on a low radiative forcing scenario (SSP 126) and a high radiative forcing scenario (SSP 585) as delineated by the Intergovernmental Panel on Climate Change (IPCC, 2023)⁸⁸. These SSPs offer a structured framework for understanding potential climate change trajectories, illustrating the anticipated impacts based on representative concentration pathways. They are predicated on estimated radiative forcing values for the year 2100 relative to preindustrial levels, providing essential insights into the expected severity and variability of future climate conditions⁸⁹. To assess the potential impacts of these climate change scenarios on suitable habitat conditions for three non-native Carassius species, we utilized bioclimatic data derived from three global circulation models (GCMs): BCC-CSM2-MR, MIROC6, and MRI-ESM2-0. As highlighted by⁹⁰, the use of multiple GCMs is crucial for mitigating uncertainty in forecasting future scenarios. This approach is particularly relevant given that the variability among different GCMs for a specific SSP can often surpass the differences observed among various SSPs within a single GCM. We specifically extracted bioclimatic parameters corresponding to both SSPs for 2041–2060. These data were sourced from the WorldClim-Global Climate Database (www.worldclim.org/data/cmip6/cmip6_clim30s.html), which provides a comprehensive array of climatic variables vital for ecological modeling. For each SSP, we calculated the average predicted probability of species occurrence across the three GCMs to reduce uncertainty associated with individual GCM outputs^77,90,91. Furthermore, we generated uncertainty maps for each prediction by employing the standard deviation of model outputs among the GCMs^92,93. By extrapolating our existing niche models to include these future climate scenarios, we evaluated potential changes in the spatial distribution of Carassius species.

Carassius hotspot map: determining the risk region for conservation assessments

Kernel Density Estimation (KDE) is a fundamental nonparametric statistical technique employed to estimate the probability density function of a random variable based on a finite sample of data⁹⁴. This method is particularly adept at analyzing point-based or line-based datasets, enabling researchers to visualize the distribution of occurrences across a specified geographical area. In our investigation, we utilized the sfhotspot package⁹⁵ to identify hotspot areas based on the current distribution of Carassius species. The approach employed is the region method, which overlays a grid or cell network onto the species’ presence points. Then selected an optimal radius to delineate the occurrence range for each observation point. For this analysis, we established a cell network with dimensions of 10 × 10 km while determining the optimal radius for each point to be 1 km. Within this framework, a hypothetical circle is constructed around each observation point based on the specified radius. The KDE algorithm then computes the density within this circular area, applying greater weight to points closer to the circle’s center. This weighting mechanism ensures that observations in proximity contribute more significantly to the density estimate than those located further away. The final density value for each cell in the grid is obtained by aggregating the contributions from all overlapping circles within that cell. Consequently, areas where multiple circles intersect will display elevated density values, thereby facilitating the identification of hotspots for species occurrence.

Results

Distribution of the Carassius species under current conditions

In Fig. 2, we delineate the potential climatic niche for each species of Carassius under prevailing climatic conditions. Notably, all occurrence records for these species fall within the designated suitable range, affirming that our models accurately encapsulate their current distribution patterns. The area under the curve (AUC) and true skill statistic (TSS) values serve as metrics for evaluating the predictive performance of our habitat suitability models. Specifically, AUC values for the Carassius species range from 0.81 to an exceptional 1, underscoring a robust predictive capability. Correspondingly, TSS values span from 0.71 to 0.93, signifying that these models demonstrate good to excellent predictive proficiency (refer to Table S2). Further examination of the models elucidated the relative significance of various climatic variables in shaping habitat suitability. Among these variables, three emerged as particularly salient: Bio 1 (Annual Mean Temperature), Bio 14 (Precipitation of Driest Month), and HF (Human footprint) (see Table 1). These variables were instrumental in informing the habitat predictions for the Carassius species. In our comparative observed analysis of different modeling algorithms, GBM, RF, and FDA all exhibited praiseworthy performance, with AUC and TSS values surpassing 0.8 (refer to Table S2). Projections indicate a substantial future increase in this suitable habitat, particularly for C. auratus, as illustrated in Table S3 and S4.

Fig. 2.

The current spatial distribution of climatically suitable habitats for three species of Carassius fish. The map was generated using ArcGIS 10.6 software (URL: https://www.arcgis.com/index.html).

Table 1.

Key Climatic variables for predicting habitat suitability. The percentage contributions of these significant variables are provided. bio 1: annual mean temperature; bio 5: max temperature of warmest Month; bio 7: temperature annual Range; bio 9: mean temperature of driest Quarter; bio 12: annual precipitation; bio 14: precipitation of driest Month; TRI: terrain ruggedness Index, and HF: human Footprint.

	Bio 1	Bio 5	Bio 7	Bio 9	Bio 12	Bio 14	TRI	HF
C. auratus	41.53	24.14	-	-	2.14	7.18	-	20.02
C. gibelio	4.03	-	-	49.19	-	19.14	2.34	23.68
C. langsdorfii	41.19	-	3.63	34.16	-	11.67	-	8.13

Projected distribution of Carassius species under future conditions

Analysis of the effects of climate change on the range dynamics of three non-native Carassius freshwater fish species reveals significant shifts in suitable habitat areas by the mid-21st century. Using three global circulation models (GCMs)-BCC-CSM2-MR, MIROC6, and MRI-ESM2-0 projections were made for the period 2041–2060 under two SSP 126 (low emissions) and SSP 585 (high emissions). The three GCMs generally agree on the change direction (significant expansion), though the extent of range win differs between models and scenarios (Tables S3, S4 and Fig S1, S2, S3, S4, S5, S6, S7, S8). The average simulation results of the projected distribution of suitable habitat areas for each Carassius species under anticipated future climate conditions reveal significant trends. In the context of the optimistic scenario delineated by SSP 126, it is forecasted that all three species will experience an expansion of their predicted geographic ranges during the temporal window spanning from 2041 to 2060; however, the magnitude of the win varies among species. Specifically, C. langsdorfii exhibiting a modest increase of approximately 6.07%, while C. auratus is anticipated to undergo a more substantial augmentation of its range, estimated at 38.85% (refer to Table S3 and S4; Fig. 3). Under the pessimistic scenario characterized by SSP 585, a similar trend of range expansion is expected for all three species during the same period. Notably, the predicted increases in their geographic distributions are forecasted to be more pronounced in this scenario, with C. langsdorfii showing an increase of around 28.85%, while C. auratus is projected to experience a remarkable expansion of up to 60.48%, indicating increased suitability to climate change for all Carassius species (Table S3 and S4; Fig. 4). However, based on the prediction of future maps, the distributional patterns and potential range expansion of three Carassius species; C. auratus, C. gibelio, and C. langsdorfii are projected to diverge. The species C. auratus and C. gibelio will primarily be distributed in and are likely to expand into coastal basins and watersheds. In contrast, C. langsdorfii is projected to find more suitable habitats and potential range expansion in the central basin and mountainous areas.

Fig. 3.

Average dynamic changes in habitat suitability for three Carassius species from 2041 to 2060 under the SSP 126 future climate scenario. The map was generated using ArcGIS 10.6 software (URL: https://www.arcgis.com/index.html).

Fig. 4.

Average dynamic changes in habitat suitability for three Carassius species from 2041 to 2060 under the SSP 585 future climate scenario. The map was generated using ArcGIS 10.6 software (URL: https://www.arcgis.com/index.html).

Hotspot map

The map illustrates the spatial intensity of three non-native Carassius species across Iran, using a kernel density function to identify hotspots of occurrence.

Primary Hotspot: The most prominent concentration of non-native Carassius is along the southern Caspian Sea basin, particularly in the provinces adjacent to the Caspian Sea. This area is marked by deep blue and magenta colors, signifying the highest kernel density values. This suggests that the Caspian basin is the main invasion core for these species, likely due to favorable environmental conditions and human-mediated introductions.

Secondary Hotspots: Additional, though less intense, hotspots are observed in: The Urmia basin in the northwest, the Zohre basin in the southwest, the Harirud basin in the northeast and scattered low-density occurrences in the Tigris, Sistan, and Mashkid basins (Fig. 5).

Fig. 5.

Hot range of biodiversity of Carassius species using kernel analysis in their Iran distribution range. The map was generated using ArcGIS 10.6 software (URL: https://www.arcgis.com/index.html).

Discussion

This study significantly contributes to the field of invasion ecology by employing climate scenarios alongside ensemble machine learning (EML) techniques to predict the potential distribution of invasive Carassius species in Iran’s freshwater ecosystems. This integrative approach effectively addresses critical methodological gaps in forecasting biological invasions, especially in regions characterized by high biodiversity and escalating human pressures.

The results highlight the importance of multi-faceted modeling in comprehending and managing aquatic invasions. By integrating climatic, topographical, hydrological, and anthropogenic factors, the research offered a comprehensive understanding of the elements driving the invasiveness of Carassius species. The selection of variables such as annual mean temperature, temperature extremes, precipitation patterns, terrain ruggedness, and human impact illustrates the complex nature of invasion success, resonating with broader ecological theories emphasizing the interaction between abiotic conditions and human-mediated dispersal mechanisms^8,96. The application of EML within the BIOMOD2 framework marks a significant methodological advancement, as ensemble models have demonstrated superior predictive accuracy and robustness compared to single-algorithm approaches^97,98. This ensemble strategy reduces individual model biases by utilizing algorithms with diverse assumptions and strengths, enhancing reliability, an essential aspect for management decisions in contexts with limited data.

Our study demonstrates methodological rigor through spatial thinning of occurrence data, careful selection and standardization of environmental predictors, and robust evaluation metrics (AUC, TSS). Using pseudo-absence points and multiple model replicates further strengthens the inference by addressing common pitfalls in presence-only data and sampling bias⁹⁹. However, several limitations warrant consideration. First, while the ensemble approach improves predictive performance, the reliance on correlative SDMs may still overlook key biotic interactions, such as competition and predation, that can modulate invasion outcomes¹⁰⁰. Although the authors note that environmental variables may suffice without strong biotic interactions, future work could benefit from integrating mechanistic or hybrid models that explicitly account for species interactions and life-history traits. Second, the spatial and temporal resolution of environmental data and the uncertainty inherent in climate projections introduce potential sources of error^101,102. Downscaling climate data and incorporating local-scale habitat features could refine predictions, particularly in heterogeneous landscapes like Iran.

Climate change and invasion risk

The study’s incorporation of future climate scenarios (SSP 126 and SSP 585) is particularly noteworthy, given the increasing evidence that climate change heightens the risk of aquatic invasions^103–105. It demonstrates climate change’s influence on the distribution of three non-native Carassius species in Iran’s freshwater ecosystems. The modeling results indicate potential spatial shifts in Carassius species, suggesting that the current distribution maps indicate suitable habitats for all species, which may expand over time (Figs. 3 and 4). According to Carosi et al. (2019)¹⁰⁶, species may face four possible outcomes in response to climate change: reduction, expansion, a combination of both, or stability. Notably, all Carassius species are predicted to pursue expansion scenarios. The output maps generated under the SSP scenarios suggest that climate change will significantly impact the distribution of all Carassius species, demonstrating a pronounced tendency for expansion over time. Liu et al. (2024) studied the effects of climate change on C. auratus in the Qinghai-Tibet Plateau using the Maxent model. Consistent with our results, they also identified the effects of climate change as a factor in the success of the expansion of this species. Also, in alignment with our findings, Mamun et al. (2018)¹⁰⁷ projected future climate change effects on the invasive Largemouth Bass in the Korean peninsula for 2050 and 2100. Their results indicate that the impact on this invasive species is expected to increase significantly by those years.

The modeling results demonstrate that regions characterized by high human population density exhibit an increased likelihood of future Carassius distribution. Bagheri & Soltani, (2023)¹⁰⁸ believed that population density in northern cities of Iran is almost higher than in other regions. This trend can be attributed to heightened human activities within these areas, including trade and travel. Aligning with our result, Jia (2019)¹⁰⁹ showed how environmental factors and human activity influence life-history traits in invasive Carassius auratus within Tibet’s Yarlung Zangbo River, as their findings suggest C. auratus may expand further under global ecological changes and that the expansion is greater in areas with high human population density. Another of our results showed that rising temperatures and the warming of aquatic environments likely influence the expansion rate of Carassius into new distribution areas. The findings indicate that climate change is projected to favor thesespecies, with temperature emerging as the most significant factor in the predictive models. Supporting these observations, Hong et al. (2022)¹¹⁰ utilized various analytical tools to assess anticipated range shifts for two invasive fish species under future climate scenarios, as their study revealed that such range shifts are expected to increase due to climate change, with temperature being the primary driver behind the expansion of these alien fish. Moreover, it has been suggested that the proliferation of invasive species may heighten the vulnerability of native species^111,112, potentially leading to a decline in native species diversity and compromising the sustainability of indigenous fish populations.

These findings support the hypothesis that climate warming promotes the establishment of thermophilic non-native species while increasing the vulnerability of native species by changing habitat structures and resource availability¹¹³. Furthermore, the explicit incorporation of anthropogenic factors, such as the human footprint, underscores the combined influence of climate and human activities on invasion dynamics. This is especially pertinent in Iran, where swift land-use changes, water extraction, and habitat fragmentation intersect with climate-induced hydrological shifts, creating new opportunities for establishing and spreading invasive species. While our study provides valuable insights into the direct climatic impacts on species distribution, it’s crucial for future research to incorporate the dynamic nature of the human footprint, as human population and land-use changes are expected to co-vary with climate to shape future habitats.

Invasion hotspots density

The KDE approach is instrumental in elucidating the spatial distribution of invasive populations, allowing researchers to identify both core areas of high density and peripheral regions where establishment is occurring. By employing KDE, the study maps the presence of these invasive species and assesses their intensity and spatial clustering, which are essential for effective management strategies¹¹⁴. Identifying invasion hotspots, particularly within the Caspian basin, underscores the significance of localized ecological dynamics and anthropogenic factors contributing to the spread of Carassius species. The designation of the Caspian basin as the primary epicenter for invasions underscores the complex interactions between environmental factors and anthropogenic activities, including aquaculture and water transfers. These dynamics create pathways that facilitate the introduction and establishment of non-native species, highlighting the necessity for coordinated management strategies at the regional level. In the Iranian Caspian basin, the interplay between environmental conditions and human activities, such as aquaculture practices and water diversion, has significantly contributed to the invasion of freshwater species like Carassius species^41,52. Alterations primarily drive this invasion to aquatic habitats from dam construction, water extraction, and irrigation practices, which collectively modify hydrological patterns, thermal regimes, and habitat structures. Such changes foster conditions conducive to the establishment and proliferation of invasive species.

Additionally, introducing exotic fish for aquaculture and ornamental purposes, alongside environmental degradation and shifts in water quality, has expedited the invasion process and facilitated the co-introduction of parasites, as these factors have adversely affected native fish populations in the region from an ecological perspective, understanding the spatial patterns of invasions is crucial for prioritizing conservation actions. Identifying high-density areas allows resource managers to allocate efforts effectively, targeting regions most at risk of further ecological degradation due to invasive species¹¹⁵. This proactive approach is vital for mitigating potential impacts on native biodiversity and ecosystem services¹¹⁶.

Furthermore, the recognition of secondary hotspots in adjacent basins indicates additional regions where early intervention could be instrumental in preventing further spread. This aligns with invasion ecology principles that advocate for timely responses to emerging threats. Early detection and rapid response frameworks can significantly enhance management outcomes by curbing the establishment of invasive populations before they become entrenched¹¹⁷. Establishing long-term ecological monitoring can provide critical data on population dynamics, ecological interactions, and responses to management interventions¹¹⁸. Such data are essential for adaptive management practices that can evolve in response to changing environmental conditions and invasion patterns¹¹⁹.

Kernel density estimation in this context enriches our understanding of invasion dynamics and provides a robust framework for developing targeted conservation and biosecurity strategies. The insights gained from this analysis can inform policy decisions and management practices aimed at safeguarding native ecosystems from the impacts of biological invasions, particularly in regions like the Caspian basin, where non-native Carassius species are most prevalent. Future research should continue to explore the interactions between invasive species and their environments and the efficacy of different management approaches in mitigating their impacts.

Finally, it is important to note that our study used Kernel Density Estimation (KDE) to create a map of species hotspots based on presence-only occurrence points. We recognize that using this type of data can raise a concern: are the hotspots on the map genuine areas of invasive species concentration, or do they reflect where authors have focused their survey efforts?

We are confident that our map represents true invasion patterns rather than just survey effort. The hotspots we identified, particularly in the Caspian Sea basin, align with known ecological and human factors that favor the spread of invasive species, such as suitable environmental conditions and human activities like aquaculture and water transfer. This strong correlation between the identified hotspots and known drivers of invasion strengthens our conclusion that these are real hotspots of invasive species, and not merely a reflection of the spatial distribution of our sampling.

Broader context and future directions

The study’s focus on Carassius species is particularly relevant in light of their well-documented ecological consequences, which include the displacement of native species and significant disruptions to food webs in regions where they have become established¹²⁰. These impacts underscore the urgency of understanding and managing the spread of these invasive fish. By generating spatially explicit risk maps that project current and future climate scenarios, this research provides valuable insights that conservation practitioners and policymakers can utilize. Identifying invasion hotspots is crucial for informing targeted monitoring efforts, facilitating early detection of invasive populations, and enabling rapid response strategies. Such proactive measures are essential for enhancing the effectiveness of biosecurity interventions to mitigate the adverse effects of Carassius species invasions.

To build upon these findings, future research should prioritize integrating socio-economic scenarios to help contextualize the ecological data within human dimensions. Engaging stakeholders in the research process is vital to ensure that management strategies are scientifically sound, socially acceptable, and economically viable. Additionally, incorporating adaptive management frameworks will allow for the operationalization of model outputs, ensuring that strategies remain flexible and responsive to changing conditions. Moreover, longitudinal monitoring and experimental validation of the predicted distributions of Carassius species will be critical to refining risk assessments. This ongoing evaluation will enhance the robustness and reliability of models, improving their transferability across different taxa and geographical regions. By addressing these aspects, future studies can significantly contribute to a more comprehensive understanding of invasion dynamics and facilitate more effective management practices in the face of ongoing environmental change.

Conclusion

Current study exemplifies the value of combining advanced modeling techniques SDM with climate scenarios to anticipate and manage biological invasions in freshwater systems, particularly in the context of three Carassius species invasions. The research employs sophisticated spatial modeling approaches such as EML and offers a nuanced understanding of how these invasive species will likely respond to changing climate conditions. Integrating predictive climate models allows for identifying potential invasion hotspots, crucial for prioritizing conservation efforts and resource allocation. As environmental conditions evolve, these integrative frameworks ongoing refinement and application will be pivotal in safeguarding freshwater biodiversity in Iran and other regions facing similar challenges. Future research should focus on longitudinal monitoring and experimental validation of model predictions, which will enhance the reliability and applicability of these tools across different taxa and geographical contexts. By addressing these aspects, the study paves the way for more effective management practices that can adapt to the dynamic nature of biological invasions in an era of rapid climate change.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

H.K: Writing – review & editing, Writing – original draft, Methodology, Formal analysis, Conceptualization, Data curation, Visualization. A. S. N: Writing – review & editing, Conceptualization, Methodology. M. K. B: Writing – review & editing, Conceptualization, Methodology. K.J: Supervision, Methodology, Writing – original draft. A. A: Supervision, Methodology, Writing – original draft, Conceptualization.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Data availability

The data supporting this study’s findings are available from the corresponding author upon reasonable request.

Declarations

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.