Introduction of alien and invasive species poses a serious threat to native biodiversity, ecosystem integrity, and agricultural productivity, which in turn results in large economic losses worldwide (Early et al., 2016; Díaz et al., 2019). Invasive species cause trophic imbalance and decrease species resilience (Weiskopf et al., 2020). Moreover, global climate change aggravates the risk of alien and invasive plant species and expands their range (Weiskopf et al., 2020). Earlier studies report that many alien and invasive species have traits that increase invasion success and enable them to readily adapt to climate change. These traits include greater tolerance to increased temperatures, intensive growth under elevated carbon dioxide levels, and geographic latitude-dependent phenotypic plasticity (Molina-Montenegro and Naya, 2012). In addition, extreme climatic events, such as prolonged heat waves, excessive drought, and heavy precipitation, facilitate plant invasion and decrease native species resistance to biotic stresses (Diez et al., 2012). For almost two decades, several researchers have anticipated that climate change will enhance the spread of alien and invasive species. However, actual evidence for climate change-facilitated alien species invasiveness is limited and poorly synthesized (Hulme, 2017), which inhibits the proactive prevention and management of alien and invasive species in the ecosystems they affect (Bonebrake et al., 2018).

Human-induced factors, such as anthropogenic land cover change, urbanization, and the construction of transportation corridors, increase invasion risk of alien plants by promoting disturbances and seed dispersal, which in turn create new habitats that enable invasive weeds to thrive, while inhibiting native species (Vila and Ibáñez, 2011; Wang et al., 2016; Rew et al., 2018). Therefore, along with climate change, other environmental variables, such as land cover change, should be considered in species distribution modeling to more effectively identify areas that are at high risk of invasion by unwanted plant species.

The rapidly growing human population accelerates economic development, international trade, and tourism, which are factors that increase the introduction of alien and invasive species (Lin et al., 2011; Seebens et al., 2018). South Korea is the world’s 12th largest importer of goods, including agriculture products, leading to the introduction of alien and invasive species (UN, 2020). Until 2016, 427 alien and invasive plant taxa were intentionally or unintentionally introduced to South Korea, resulting in their invasion of agricultural fields, orchards, and pastures (NIE, 2017; Young et al., 2017). The Ministry of Environment, South Korea listed 16 alien and invasive plants as ecosystem disturbing alien plant species (EDAPS) based on their noxious characteristics and high invasion potential to the ecosystem (Young et al., 2017; MoE, 2022). Alien and invasive species, including EDAPS, cause the loss of approximately 22.6 billion Korean Won per year (MoE, 2014). Because of the harmful effects of invasive plant species on the country’s economy, it is necessary to profile alien and invasive species, and identify high invasion risk areas. Doing so will enable the adoption of cost-effective control and management strategies to minimize the threats of EDAPS invasion in South Korea. Moreover, implementing tools that can forecast potential changes in invasive species distribution and accurately assess future invasion risks is required.

The problems caused by invasive species prompted us to study invasive weeds in South Korea under future climate change scenarios, to guide the development of measures to control their spread (Adhikari et al., 2021; Hong et al., 2021). In one study, 16 invasive weeds from the southern region (<36° latitude) were evaluated. These 16 weed species were estimated to expand their range and habitat to the northern region of the country in large scale over 145.19% in year 2070 (Hong et al., 2021). In another study, ten alien plant species that were intentionally introduced for horticulture, pasture improvement, and soil erosion control were assessed for their spatial invasion risk. Under future climate change scenarios, intentionally introduced species are predicted to spread to non-targeted ecosystems and occupy approximately 86.21% area of the country, which falls under the moderate- to high-risk category (Adhikari et al., 2021). Both of our previous studies identify rapid environmental changes that carry an increasing risk of alien weed invasion. These reports indicate that the rise in invasive weeds could seriously threaten natural ecosystem integrity and cropland productivity in South Korea.

Risk assessment is an essential step before initiating procedures for the control and eradication of alien and invasive species. Numerous risk assessment tools are used to quantify invasion risk of alien plant species (Morse et al., 2004; Robertson et al., 2021). Species distribution models (SDMs), for example, are widely used tools to assess the invasion risk of alien plants, and aid in designing conservation policies for their control and eradication (Bradley et al., 2010; Dullinger et al., 2017; Adhikari et al., 2021). SDMs resolve the ecological niche of plant species by statistical regression, machine learning, and spatial interpolation (Elith and Leathwick, 2009; Li and Wang, 2013; Zurell et al., 2020). However, the large number of SDM algorithms make it difficult to select the best algorithm and methodology (Elith et al., 2010). Moreover, a single SDM may not always provide accurate predictions for risk assessment, and there is no modeling technique that consistently performs better in predicting invasiveness across species, regions, and applications (Elith et al., 2006; Koo et al., 2017). Evaluating many models and selecting the best one based on its predictive performance allow the acquisition of more reliable predictions (Mainali et al., 2015; Koo et al., 2017; Zurell et al., 2020; Ahmed et al., 2021). Therefore, we adopted this concept in the risk assessment of EDAPS in South Korea.

The main objective of this study was to conduct a risk assessment of EDAPS present in South Korea. The specific objectives were as follows: (1) Selection of the best algorithm between two statistical regression-based models, namely, the generalized linear model (GLM) and multivariate adaptive regression splines (MARS), and among three machine learning models, namely, artificial neural network (ANN), maximum entropy (MaxEnt), and random forest (RF). Algorithms were assessed using established the model calibration techniques, area under the curve (AUC) value of receiver operating curve (ROC) (Pearson, 2010), true skill statistics (TSS) (Allouche et al., 2006), and Kappa scores (McHugh, 2012). (2) Estimate the potential distribution area of 12 EDAPS and assess invasion risk using the best SDM algorithm under the current and future climate change scenarios, representative concentration pathways (RCP) 4.5 and RCP 8.5, and land cover changes in South Korea.

The following results were obtained in this study. (1) RF was selected as the best model among the five SDMs based on the average AUC, TSS, and Kappa scores. (2) The RF model estimated invasion risk to be high in the coastal areas of the northwestern and southern regions, and some large cities of South Korea, such as Incheon, Seoul, and Busan, under the current climate. In the future climate change scenarios, invasion risk was predicted to increase across the whole country except for the high-altitude northeastern region of Gangwon province (Figure 4A). (3) The extreme future climate change scenario presented by RCP 8.5 revealed that EDAPS distribution fell under the high- or very high-risk category, which corresponded to a large proportion of the country’s area.

The five SDMs used in this study showed different prediction results of EDAPS while using same environmental variables and species occurrence data, which indicates the model uncertainty (Wenger et al., 2013). Accounting for this model uncertainty is critical for effective conservation actions. Therefore, we quantified model uncertainty by comparing predictive performances of each model using three standard model evaluation metrices (AUC, TSS, and Kappa). A quantitative evaluation of model performance establishes the model’s suitability for specific applications and identifies components of the model that require further improvement (Koo et al., 2017). Furthermore, evaluating model performance can be used to compare different modeling approaches, and to study how different data and/or species characteristics affect the accuracy of predicted maps created by the model (Allouche et al., 2006). The AUC value shows the predictive capacity of a model using training and test datasets (Thuiller et al., 2005) to distinguish observed occurrences from background data (Thuiller et al., 2005). The AUC is unaffected by dataset size (prevalence); however, it has limitations because it equally weighs commission and omission errors, avoids true probability values, and is based on geographical extent (Lobo et al., 2008). AUC values are inflated when the geographical range is expanded beyond the current distribution range, resulting in overestimation of AUC. This in turn causes misinterpretation during model evaluation (Lobo et al., 2008). Because of the uncertainties with AUC metrics, TSS and Kappa were used as alternative criteria for validating model performance. TSS correlates with sensitivity, which corresponds to observed presence that is precisely estimated, and specificity, which means observed absence that is precisely estimated. Sensitivity and specificity are recognized as omission and commission errors, respectively (Pearson, 2010). Kappa values describe the overall accuracy of model performance with the assumption that accuracy occurs by chance (Allouche et al., 2006). In this study, we used average scores of AUC, TSS, and Kappa (n = 12) for evaluating model performance and selecting a best model. Similar to our findings, other studies used these evaluation metrics for model calibration, validation, comparison, and selection of a best model (Li and Wang, 2013; Mainali et al., 2015; Koo et al., 2017). Here, RF was the most accurate model in predicting the distribution of all EDAPS, while ANN was the worst performing model (Figure 3). Consistent with results presented in this study, the superior performance of RF was reported by previous SDM studies (Attorre et al., 2011; Mainali et al., 2015).

Our studies showed that current invasion was predicted to be highest in the northwestern region, where Gyeonggi province and Seoul city are located. High invasion risk also partly occurred in the southern and southeastern regions, where Jeollanam-do, Jeju-si, and Gyeongsangnam-do provinces, as well as Busan city, are located (Figure 4A). However, the central region of the country, which includes Chungcheongbuk-do province, Daejeon and Sejong cities, and Gangwon province in the northeastern region, exhibited low risk of invasion. These results showed that EDAPS invasion risks initiated along ocean routes of the northwestern, southern, and southeastern regions, and primarily colonized the harbor areas of these regions and surrounding islands. Similar predictions were uncovered by Adhikari et al. (2019, 2021) and Hong et al. (2021). Invasion is typically brought about by global trade and tourism, in which invasive plants enter through harbors and airports intentionally or unintentionally (Hellmann et al., 2008). The northwestern, southern, and southeastern regions of South Korea have many international harbors (e.g., Incheon, Pyeongtaek, Yeosu, Mokpo, and Busan) and airports (e.g., Incheon, Busan, and Muan). Therefore, invasive plants are introduced through these harbors and airports, and colonize agricultural fields, urban lands, and wild ecosystems (Early et al., 2016). Moreover, densely populated areas and disturbed lands are characteristic of big cities, such as Seoul, Incheon, Busan, and Daegu, and Gyeonggi-do province. Therefore, invasion risk could be more severe in these areas. Urban development, industrialization, construction of road networks, and establishment of recreational parks lead to disturbed lands, thereby facilitating the growth of EDAPS (Hellmann et al., 2008; Adhikari et al., 2020a, 2021). Introduced invasive species could then disperse across the country through land, air, and water transportation systems (Eminniyaz et al., 2013).

Under the future climate change and anthropogenic land cover change, large areas of the country were predicted to be under the high-risk (27.86%) and very high-risk (60.43%) categories of invasion by 2070. Furthermore, it was estimated that these types of invasion events will occur in the northwestern, western, and southern regions of the country, particularly in the provinces of Jeollanam, Jeollabuk-do, and Gyeonggi (Figure 4A). These predictions are reinforced by the fact that the RF model selected for risk assessment is a more robust classifier than other SDMs, and it has a tendency to emphasize differences between grid cells (Mainali et al., 2015). Furthermore, RF has a flexible fitting process, making it very efficient in modeling complex responses (Berk, 2008). Climate change generates favorable conditions for introducing new invasive species into ecosystems. It does so by making the habitat more suitable for the invading species, while changing native species distribution and abundance (Hellmann et al., 2008). Moreover, climate change alters competitive interactions among species, making native groups more susceptible to colonization by new and established invaders (Finch et al., 2021). Extreme climate (i.e., very wet and dry) favors establishment and range expansion of invasive species (Hellmann et al., 2008; Adhikari et al., 2019). Unfortunately, the average temperature in year 2100 is predicted to be 5.7°C higher than the current average temperature (16°C) in South Korea (KMA, 2021). Increasing temperature may play roles in expansion of transportation networks, urbanization, and industrialization. Under these circumstances, finding the most accurate model is required to mitigate potential invasion risks of EDAPS and/or any invasive weeds and plants.

In this study, we investigated invasion risk of 12 EDAPS in South Korea using SDMs. SDMs are empirical tools for predicting species spatial distribution across geographical space and time by determining correlations between species occurrence and environmental parameters (Elith and Leathwick, 2009; Franklin, 2013; Zurell et al., 2020). SDMs are used extensively in landscape ecology, conservation biology, wildlife management, and studies of climate change (Li and Wang, 2013). SDMs are used synonymously with ecological niche species distribution, habitat suitability, and climatic envelope models (Zurell et al., 2020). We created invasion risk maps for South Korea and categorized them into low risk, moderate risk, high risk, and very high risk under the current and future climate change and land cover change. To the best of our knowledge, this is the first study assessing invasive and alien plants using five SDMs. Furthermore, the best model for alien and invasive plant risk assessment in South Korea was identified. Through this model, we expect to cover research gaps in risk assessment of alien and invasive plants in South Korea. The current study sheds light on prediction modeling of distribution and invasion risk that can be applied to other invasive weed and plant species. This study could inform the central and federal governments and society about areas at high risk of invasion, guide invasive species management, and secure economic resources (Renteria et al., 2021).

Ecosystem disturbing alien plant species invasion risk may not only depend on climatic factors and land cover change, but also on other factors, such as biogeographic origin, and biological and physiological traits (Ni et al., 2021). Most of the EDAPS studied here originated from tropical regions of South America and temperate regions of North America and Europe. They were indiscriminately introduced into South Korea except for S. pilosum and S. altissima, which were imported as ornamental plants and as a source of nectar (Table 1). The introduced EDAPS then colonized disturbed lands, such as parks, roadsides, and coastlines, and eventually became fully adapted. Such plants adapt better to climatic variability than native species (Barral, 2019). The warming climate correlates positively with seed germination, and daytime temperature fluctuations may stimulate seed germination (Chen et al., 2017). In addition, species specific biological traits, such as growth rate, reproduction rate, and seed mass, correlate with invasiveness (Carboni et al., 2016). Furthermore, physiological traits, such as photosynthetic rate, efficiency of resource utilization, and competition, could play important roles in the success of invasive plant species (Ni et al., 2021). Moreover, natural phenomena, including wind, tides, and flood, and animals, such as birds and mammals, could facilitate EDAPS seed dispersal. Further studies are required to determine if similar phenomena influence EDAPS invasion risks in South Korea. Therefore, in addition to climatic and anthropogenic factors, biological and physiological traits of invasive plants, and methods of seed dispersal, need to be incorporated in models to obtain more accurate predictions of invasiveness.

This study compared the predictive performance of five SDMs based on three evaluation metrics. These metrics are widely used for calibration and validation of model performance. By comparing the predictive performance and outputs of the SDMs, we concluded that RF is the best modeling method among the five SDMs for forecasting EDAPS invasion risk. Climate change is likely to favor the spread of EDAPS and increase invasion risk across South Korea. This study suggested that comparison of many models with the same species and location is an ideal approach to determine the best model, and interpret the accuracy and coherence of the modeling results, for predicting the impact of climate change on species invasiveness. Predicting current and future EDAPS invasion risk areas could provide valuable information for the development of effective control and management strategies to minimize the threat of invasive plant species at the regional and local levels. Based on the results of risk analysis presented in this study, EDAPS detection and eradication activities are required in South Korea’s northwestern, southern, and southeastern regions, particularly focusing on the cities of Seoul and Busan and the provinces of Incheon, Chungcheongnam, and Jeollanam. These areas are likely to be the source for future EDAPS range expansion into new geographical areas.

The original contributions presented in this study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding authors.

PdA: conceptualization, software, formal analysis, and writing–original draft preparation. PdA and YL: methodology and validation. PdA, YL, and Y-SP: investigation. PdA and PbA: literature review and data curation. PdA and SH: resources. PdA and Y-SP: review and editing. Y-SP and SH: supervision. SH: project administration and funding acquisition. All authors read and approved the published version of the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.