Species distribution models (SDMs)

To estimate the potential effects of climate change on insect pollinators in the Republic of Korea, we used SDMs. These models require two categories of dependent and independent data, as follows: (i) presence and absence points of species, as dependent data; and (ii) predictor variables like climatic data, as independent data (Rahimi & Jung 2024b).

Occurrence data and data cleaning

This research employed presence-only data from the Global Biodiversity Information Facility (GBIF, https://www.gbif.org) to model the distribution of insect pollinators in the Republic of Korea. The taxa studied included bees, butterflies, moths and hoverflies. The adequacy of occurrence data for species distribution modeling hinges on multiple factors, including model complexity, data quality, species characteristics, research objectives and scale (Bean et al. 2012; Rahimi & Jung 2024b). Although there is no fixed rule, it is generally advisable to have a larger sample size with more occurrence data points for more robust modeling outcomes. Although a rough guideline suggests aiming for a minimum of 30–50 presence points, this requirement can vary significantly depending on the specifics of each study (Bean et al. 2012; Stockwell & Peterson 2002). In our research, we established a minimum requirement of 30 data points to ensure precise model outcomes. It is worth noting that the field of SDMs encompasses studies with varying quantities of occurrence data, ranging from very few to several thousand data points (Orr et al. 2021; Williams et al. 2009). To ensure data accuracy, a meticulous examination of presence points is required, aiming to eliminate duplicate records and those positioned beyond the boundary of the study area. The package rgbif, a convenient R interface to access biodiversity data from GBIF, played a key role in data manipulation (R Foundation for Statistical Computing, Vienna, Austria). This package (Chamberlain et al. 2022) enables users to query and retrieve biodiversity data directly within the R environment. We also implemented a minimum distance of 10 km between occurrence data points to mitigate potential spatial bias.

Table 1 provides a breakdown of the taxa studied, along with their respective families and the number of species per family. For bees, we refined our data set to a total of seven species distributed across three families. Butterflies exhibited a higher diversity, with a total of 47 species across five families. The family Nymphalidae had the highest number of species, with 20, followed by the family Lycaenidae, with 10 species. Moths showed the highest diversity among the taxa studied, with a total of 133 species across five families. The family Geometridae was the most diverse, with 62 species, followed by Noctuidae with 53 species. Lastly, hoverflies exhibited a total of 19 species within the family Syrphidae. In total, we compiled data for 206 species spanning all taxa, forming a comprehensive data set for modeling insect pollinator distribution in the Republic of Korea.

Environmental variables

Bioclimatic layers, used as predictor variables, were also downloaded from the WorldClim database (https://www.worldclim.org) using the R package geodata (Sillero et al. 2023), which includes climate information up to the year 2000. The bioclimatic data in the WorldClim database includes 11 temperature variables and eight precipitation variables, which have a high correlation with each other, and therefore it is not recommended to use all these variables in species distribution modeling. For this purpose, we employed the package usdm (Naimi 2017) to eliminate highly correlated variables from the data set using a stepwise approach relying on the variance inflation factor (VIF). The remaining variables include isothermally (BIO 3), temperature annual range (BIO 7), mean temperature of wettest quarter (BIO 8), annual precipitation (BIO 12), precipitation of wettest month (BIO 13) and precipitation of driest month (BIO 14).

Model fitting

In this study, we employed the R package FLEXSDM (Velazco et al. 2022) to model the distribution of the species under investigation. To accomplish this, we utilized the MaxEnt model to generate climate suitability maps based on both presence data and climate information. The MaxEnt model has been extensively utilized in various studies to predict species distribution and assess the impacts of climate change, particularly on insects (Abrol 2006; Aguirre-Gutiérrez et al. 2013; Filazzola et al. 2020; Miličić et al. 2018b; Rahimi & Jung 2024b; Tabor & Koch 2021). Additionally, we created 5000 random pseudo-absence points to supplement the presence data for modeling with the MaxEnt model across each continent. Subsequently, we employed the climate change scenarios outlined in SSP245 (shared socio-economic pathway) and SSP585 (shared socio-economic pathway) in 2070, where SSP245 represents a scenario with intermediate challenges and moderate emissions. It involves a pathway where CO₂ emissions peak around 2040 and then gradually decline. Temperature increases are expected to be more moderate compared with SSP585. SSP585 represents a pessimistic climate change scenario, characterized by a tripling of CO₂ emissions by 2075 and a projected temperature increase of 4.4 degrees by 2070 (O'Neill et al. 2016).

Furthermore, we categorized the climate suitability map into two classes: low and high suitability, with the latter encompassing values exceeding 0.6 (Rahimi et al. 2021; Rahimi & Jung 2024b; Wang et al. 2018). Consequently, we exclusively assessed range changes within the high-suitability class. Following this, we calculated the percentage of change using the following formula:

Percentage of change = [(number of cells in the future under the suitable class) – (number of cells in the current under the suitable class)/(number of cells in the current under the suitable class)] × 100.

Model assessment

To assess model performance, we employed three metrics: the inverse mean absolute error (IMAE); the area under the receiver operating characteristic curve (AUC); and the Boyce statistic (BOYCE), utilizing the R package FLEXSDM (Velazco et al. 2022). The range of AUC values between 0.7 and 0.9 is considered acceptable, and values above 0.9 are considered excellent results, which means that the model has estimated a very good prediction. IMAE, calculated as one minus the mean absolute error, signifies higher accuracy with increasing values. The Boyce index evaluates model effectiveness across various suitability levels, ranging from −1 to 1. Values nearing 1 indicate strong alignment between observed and predicted probabilities, whereas values close to 0 suggest performance akin to random chance. Additionally, values below 0 denote poor predictive power. Cross-validation using a fivefold methodology was employed to gauge model performance (Fielding & Bell 1997).

Model assessment

Table 2 summarizes the model validation metrics for four taxa (bees, butterflies, hoverflies and moths) using the MaxEnt algorithm. The metrics include AUC, BOYCE and IMAE, with standard deviations presented in parentheses. All groups of pollinators showed similar and moderate discrimination abilities. Bees and hoverflies had AUC values ranging from 0.78 to 0.76, whereas butterflies and moths had AUC values ranging between 0.77 and 0.75. Strong alignment between observed and predicted probabilities was indicated by high BOYCE values ranging from 0.97 to 0.93 for bees, butterflies and hoverflies, and 0.95 for moths. Relatively low error rates in model predictions were observed across all taxa, with IMAE values ranging from 0.68 to 0.62.

Current and potential future distributions

Tables 3 and 4 offer insights into the projected impacts of climate change on the distribution of insect taxa, specifically bees, butterflies, hoverflies and moths, under two different scenarios: SSP245 and SSP585, for the year 2070. These tables provide the average percentage of changes in distribution area of the high suitability class, along with the number of species for which the distribution range is expected to increase or decrease. Starting with bees, under the SSP245 scenario in Table 3, two species were projected to experience an increase in distribution range, with an average increase of 24.3%, whereas five species were expected to decrease, with an average decrease of 61.3%. However, under the more severe SSP585 scenario in Table 4, only one species was projected to increase, with a substantial average increase of 161%, whereas six species were anticipated to decrease, with an average decrease of 73.2%. This indicated a higher magnitude of change under the SSP585 scenario compared with the SSP245 scenario, both in terms of increase and decrease.

For butterflies, hoverflies and moths, a similar trend was observed. In Table 3, under the SSP245 scenario, there were increases in distribution range for 12 butterfly species (average increase of 144.6%) and six hoverfly species (average increase of 75.7%), whereas there were decreases for 35 butterfly species (average decrease of 86.1%) and 13 hoverfly species (average decrease of 68.8%). However, under the SSP585 scenario in Table 4, the number of species experiencing an increase in distribution range was higher, with 15 butterfly species (average increase of 157%) and ten hoverfly species (average increase of 117.9%) predicted to increase their distribution ranges. Conversely, the number of species experiencing a decrease also rose, with 32 butterfly species (average decrease of 89.7%) and nine hoverfly species (average decrease of 90.8%) predicted to decrease their distribution ranges.

For moths, the pattern was consistent with other taxa. In Table 3 (SSP245), 37 species were projected to increase in distribution range (average increase of 131%), whereas 96 species were expected to decrease (average decrease of 90.8%). Under the SSP585 scenario in Table 4, the number of species with an increase in distribution rises to 38 (average increase of 199%), whereas the number of species with a decrease in distribution decreased slightly to 95 (average decrease of 87%).

To demonstrate the impact of climate change on the distribution of the studied taxa, we selected representative species from each group. Figure 1 showcases habitat suitability maps for four species under varying climatic scenarios: Bombus ussurensis (family Apidae) under current conditions (A), under SSP245 (A1) and under SSP585 (A2); Taraka hamada (family Lycaenidae) under current conditions (B), under SSP245 (B1) and under SSP585 (B2); Eristalinus viridis (family Syrphidae) under current conditions (C), under SSP245 (C1) and under SSP585 (C2); and Mocis ancilla (family Noctuidae) under current conditions (D), under SSP245 (D1) and under SSP585 (D2).

Bombus ussurensis, a bee species, is anticipated to expand its distribution range as a result of rising temperatures, likely shifting towards central and southern regions of the Republic of Korea. Conversely, Taraka hamada, a butterfly species, is expected to experience a decrease in distribution as global warming progresses. Although currently favorable habitats exist in southern and coastal regions of the Republic of Korea, these areas may become less suitable in the future, potentially restricting the species distribution to the southernmost parts of the country. Eristalinus viridis, a hoverfly species, is predicted to expand its distribution across the Republic of Korea in response to climate change. The habitat suitability maps indicate that large portions of the Republic of Korea are likely to become favorable for this species in the future. Lastly, Mocis ancilla, a moth species, is expected to undergo a decrease in distribution, with bidirectional distribution of Northern and Southern parts of South Korea becoming suitable habitats by 2070.

Bees

In this study, we estimated the potential impact of climate change on six species of bees from three families in the Republic of Korea under two climate change scenarios in 2070. Our results indicate that under the SSP245 scenario, a moderate number of species are projected to experience an increase in distribution range, with an average increase of 24.3%, whereas a smaller number of species are expected to experience a decrease in distribution range, with an average decrease of 61.3%. However, under the more severe SSP585 scenario, the magnitude of change increases, with fewer species experiencing an increase (one species, with an average increase of 161%) and with more species experiencing a decrease (six species, with an average decrease of 73.2%). Certain bee species might face substantial declines in specific regions while showing an increase in others. Averaging these fluctuations globally may not provide an accurate representation of the susceptibilities of individual regions (Rahimi & Jung 2024b). These findings are consistent with previous studies, which have also demonstrated the significant impact of temperature changes on the distribution of honeybees in Asia (Rahimi & Jung 2024a).

For example, Rahimi and Jung (2024b) estimated how global warming affects bees on a global scale. Using species distribution models, they assessed the impact of climate change, particularly under the SSP585 scenario by 2070, on the climate suitability of 1365 bee species globally. The findings revealed shifting suitability patterns, with around 65% of bee species showing potential reductions in their distribution ranges associated with climate change. Notably, the effects varied across continents, with Africa and Europe experiencing more significant impacts compared with North America, which was relatively less affected. Kelemen and Rehan (2021) also studied the effects of climate change on Ceratina calcarata (a type of carpenter bee) from 1902 to 2019 and observed that summer temperatures increased during this period and were negatively correlated with the body size of female bees. Similarly, Giannini et al. (2020) found that 95% of the 216 bee species they studied were at risk of decline caused by climate change. Furthermore, Polce et al. (2014) showed that climate change could disrupt the overlap of suitable areas for horticulture and pollinators in Australia. Under RCP8.5, Rader et al. (2013) predicted that the honeybee pollination would decline by 14.5%, whereas native bee pollination would increase by 4.5% by 2099. Our study also found that in the future, the overlap between suitable areas for bees and fruits may decrease because of climate change. As a result, the support that bees can provide for pollinator-dependent products may decrease with increasing temperatures in the future.

Butterflies

For butterflies, under the SSP245 scenario, a substantial number of species are projected to experience an increase in distribution range, with an average increase of 144.6%, whereas a considerable number of species are expected to experience a decrease in distribution range, with an average decrease of 86.1%. Similarly, under the SSP585 scenario, a notable number of species are projected to increase (15 species, with an average increase of 157%), whereas a larger number of species are expected to decrease (32 species, with an average decrease of 89.7%). Crossley et al. (2021), also found that precipitation and temperature changes in the last 26 years have led to a decrease in butterfly abundance in dry and hot regions and an increase in cool and humid regions, and Forister et al. (2010) found a significant reduction in species richness of 169 butterfly species across half of the areas at low elevations in California over a period of 35 years.

Engelhardt et al. (2022) also demonstrated a significant decline in butterfly populations in Germany over the past four decades, indicating a persistent trend. Similarly, our findings for Europe suggest a potential decrease in butterfly distribution in the future, with a shift towards northern regions. Additionally, Dar et al. (2022) have highlighted the adverse impact of increasing altitude on butterfly abundance and diversity in India, aligning with the expectation that species may migrate to higher altitudes in response to rising temperatures. Conversely, Kukkonen et al. (2022) have provided evidence of a decline in a specific butterfly species population in southern Finnish islands, possibly linked to climate change or other factors.

However, conflicting results from other studies do not attribute significant effects to climate change on butterflies (Cormont et al. 2011; Kwon et al. 2021; Peterson et al. 2004; SBARAGLIA 2022). These discrepancies underscore the complexity of the impacts of climate change on butterflies, necessitating further research. Although some studies indicate negligible or even positive effects of climate change on butterflies, others demonstrate significant negative impacts. These variations may stem from differences in study locations, species and methodologies, emphasizing the importance of continuing research to comprehensively understand climate change impacts on butterflies and guide conservation efforts.

Hoverflies

For hoverflies, our findings reveal significant changes in distribution range under both SSP245 and SSP585 scenarios. Under SSP245, an average increase of 75.7% is predicted for six species, whereas under SSP585, there is an average predicted increase of 117.9% for ten species. Conversely, several species are projected to experience a decrease in distribution range, with an average decrease of 68.8% under SSP245 for 13 species, and an average decrease of 90.8% under SSP585 for nine species. Inconsistently with our findings, Miličić et al. (2018a) predicted the effects of climate change on hoverflies in Southeast Europe under RCP8.5 in 2070. Their study revealed that the distribution range of 55% of the 44 studied species is likely to increase by 2070. In another investigation, Miličić et al. (2018b) explored the influence of climate change on hoverflies in Serbia and observed that three out of the ten studied species might encounter a reduction in their distribution range. Conversely, Radenković et al. (2017), in their study on climate change impacts on hoverflies in the Balkan Peninsula, concluded that the distribution range of all examined species might decrease in the future.

Moths

Under both SSP245 and SSP585 scenarios, a considerable number of moth species are projected to experience an increase in distribution range. Under SSP245, there is an average increase of 131% for 37 species, whereas under SSP585, there is an average increase of 199% for 38 species. Conversely, a substantial number of species are expected to decrease in distribution range under both scenarios, with an average decrease of 90.8% under SSP245 for 96 species, and an average decrease of 87% under SSP585 for 95 species. Prior research has revealed the susceptibility of moths to temperature variations, with documented population declines in various regions globally. For instance, Martay et al. (2017) reported a 48% decline in populations among 265 moth species in the UK since 1970. Furthermore, studies by Keret et al. (2020) and Vanhanen et al. (2007) illustrated how moth populations have shifted to higher altitudes and latitudes in response to changes in temperature.

Conversely, certain investigations propose that rising temperatures could enhance the abundance of select moth species (Huang & Hao 2020; Hunter et al. 2014; Jung et al. 2013; Kroschel et al. 2013; Tamburini et al. 2013). However, it is crucial to note that these observations do not necessarily signify a positive outcome for moth populations overall. Temperature fluctuations can introduce complex and sometimes unpredictable effects on ecosystems and species interactions. Consequently, the net implications of these changes on moth populations are anticipated to fluctuate, contingent upon specific species and environmental conditions.