Introduction
Economic globalisation, enhanced connectivity through international trade, and the frequent movement of agricultural commodities have collectively facilitated the rapid global spread of invasive alien species (IASs) (Vilà et al., Reference Vilà, Espinar, Hejda, Hulme, Jarošík, Maron, Pergl, Schaffner, Sun and Pyšek2011). Moreover, the effects of climate change are projected to exacerbate the frequency, severity, and extent of damage caused by IASs, as well as increase their volatility and geographic distribution, thereby profoundly affecting agricultural production (Ramasamy et al., Reference Ramasamy, Das and Ramesh2022). The rapid spread of IASs poses a notable threat to agriculture, causing substantial crop damage (Zhang et al., Reference Zhang, Hu, Wang, Wang, Ji, Fang-Zhou and Lu2024). The resulting crop damage underscores the urgency of understanding invasion pathways, predicting habitat suitability under climate shifts, and deciphering interactions between environmental and human factors, all of which are key to effective monitoring and management.
Rice water weevils (RWWs), Lissorhoptrus oryzophilus Kuschel, and rice weevils (RWs), Echinocnemus squameus Billberg (both Coleoptera: Curculionidae) are two major pests that cause significant economic losses in major rice-producing regions of China. RWW is considered the most significant pest affecting rice worldwide, with its parthenogenesis capability enabling rapid population expansion and invasion (Aghaee and Godfrey, Reference Aghaee and Godfrey2014; Wan and Yang, Reference Wan and Yang2016). Native to North America, it has spread and established in southern China since 1959 (Stout et al., Reference Stout, Riggio, Li and Roberts2002; Saito et al., Reference Saito, Hirai and Way2005). Since its initial detection in Tanghai County, Hebei Province, China, in 1988, the species has expanded rapidly across the country. It is now established in 25 provinces (autonomous regions and municipalities), including coastal regions such as Liaoning, Tianjin, Shandong, Zhejiang, Fujian, and Taiwan, as well as inland areas such as Hunan, Anhui, Shanxi, Shaanxi, Yunnan, Xinjiang, Guizhou, and Sichuan, affecting 492 counties (cities or regions) as of 2025 (Shang et al., Reference Shang, Cheng, Jiang, Tang and Gu2003; Li et al., Reference Li, Du, Duan, Hou, Sheng, Wu, Shi, Lu, Zhou, Sun and Cao2025). In contrast, the RW is native to China and is widely distributed across the country, ranging from Heilongjiang in the north to Guangdong and Hainan in the south and from western regions such as Shaanxi, Gansu, Sichuan, and Yunnan to eastern coastal areas and Taiwan. Both species can be found in the same areas and have overlapping ecological niches, which cause severe damage to rice plants. Adult RWW feed on leaves, and their larvae consume roots, leading to impaired root development, seed dislodgement, and yield reduction (Zou et al., Reference Zou, Stout and Dundand2004; Tindall et al., Reference Tindall, Bernhardt, Stout and Beighley2013). Similarly, adult RWs feed on young stems and leaves, whereas larvae target tender roots, resulting in symptoms such as yellowing leaf tips, stunted growth, incomplete panicle formation, blighted grains, and, in severe cases, patchy plant death.
Species distribution models (SDMs) are valuable tools for assessing the impact of climate change on IASs (Lake et al., Reference Lake, Briscoe Runquist and Moeller2020; Liu et al., Reference Liu, Wolter, Xian and Jeschke2020). Recently, SDMs have been increasingly used to predict the potential geographical distributions of IASs by integrating environmental and species occurrence data, as well as ecophysiological information, under various climate change scenarios (Runquist et al., Reference Runquist, Lake, Tiffin and Moeller2019, Reference Runquist, Lake and Moeller2021). Previous studies have focused primarily on the potential impacts of climate change on the distributions of various species at the regional level, including plants (Garcia et al., Reference Garcia, Lasco, Ines, Lyon and Pulhin2013), mammals (Alamgir et al., Reference Alamgir, Mukul and Turton2002), birds (Tingley et al., Reference Tingley, Monahan, Beissinger and Moritz2009), reptiles (Araújo et al., Reference Araújo, Thuiller and Pearson2006), amphibians (Thomas et al., Reference Thomas, Cameron, Green, Bakkenes, Beaumont, Collingham, Erasmus, De Siqueira, Grainger, Hannah, Hughes, Huntley, Van Jaarsveld, Midgley, Miles, Ortega-Huerta, Peterson, Phillips and Williams2004), and insects (Hongoh et al., Reference Hongoh, Berrang-Ford, Scott and Lindsay2012). Commonly used SDMs for predicting the potential geographical distributions of weeds, biocontrol agents, disease vectors, and pathogens include the maximum entropy model (Maxent), CLIMEX, the generalised linear model (GLM), and the genetic algorithm for rule-set prediction (GARP) (Rathore and Sharma, Reference Rathore and Sharma2023). Numerous SDMs are available, each differing in principles, algorithms, and predictive performance (Luo et al., Reference Luo, Wang and Lyv2017). Integrating results from multiple models into a consolidated output can increase the accuracy of predictions. Accordingly, the R-based modelling platform Biomod was developed in 2003 and has since been widely used for predicting species’ potential geographic distributions (Thuiller, Reference Thuiller2003).The R package Biomod2 supports integrated modelling via multiple algorithms, including the Maxent, GLM, GARP, multiple adaptive regression spline (MARS), surface distance envelope (SRE), flexible discriminant analysis (FDA), categorical regression tree analysis (CTA), gradient boosting model (GBM), random forest (RF), and artificial neural network (ANN) (Lantschner et al., Reference Lantschner, de la Vega and Corley2019). Users can freely customise model combinations and adjust initial conditions, model types, parameters, and boundary conditions to achieve optimal predictive performance. Compared with single-model approaches, ensemble models (EMs) integrate predictions from multiple techniques to improve accuracy, reduce bias, and enhance reliability, making them tools for predicting the potential geographical distributions of species (Araújo and New Reference Araújo and New2007; Dormann et al., Reference Dormann, Calabrese, Guillera-Arroita, Matechou, Bahn, Bartoń and Hartig2018; Lázaro-Lobo et al., Reference Lázaro-Lobo, Ramirez-Reyes, Lucardi and Ervin2021).
The objectives of this study were to (1) predict the potential geographical distribution areas of RWW and RW in China under current and three future 2050s climate change scenarios; (2) assess shifts in their distribution centres under three sustainable paths (SSP126, SSP245, and SSP585); (3) evaluate niche overlap between the two species; and (4) identify the key environmental factors influencing their distributions. These findings provide a critical foundation for predicting the potential distributions of RWW and RW in China, thereby supporting the development of effective prevention and control strategies.
Materials and methods
Species geographical distribution data
Data on the occurrence of RWW and RW in China were collected from the Global Biodiversity Information Facility (GBIF) database (https://www.gbif.org/), relevant literature and papers published in China, and government documents and reports. The distribution data were matched with the Geographic Names Database to obtain the corresponding latitude and longitude for each location. The distribution data were organised by region, longitude, and latitude and saved in .csv format for future use. To ensure compatibility with the resolution of the environmental variables, we refined the occurrence data via ENMTools by retaining only one record per 2.5 × 2.5 km grid, thereby reducing spatial autocorrelation and minimising potential bias from clustered data points (Warren et al., Reference Warren, Glor and Turelli2010; Wang et al., Reference Wang, Li, Feng and Shi2017). To elucidate the spatiotemporal dynamics of the suitable habitats of the two weevil species across China in response to climate change, agricultural landscape transformations, and anthropogenic control measures since their first records, occurrence data have been partitioned into discrete temporal windows for analysis (RW: 1954–2003, 1954–2013, and 1954–2023; RWW: 1988–2003, 1988–2013, and 1988–2023). Both future projections of suitable habitats and the niche equivalence and similarity tests conducted with the ecospat package were based on complete, undivided occurrence datasets (RW: 1954–2023; RWW: 1988–2023), thereby fully encompassing the climatic and environmental space occupied by the two weevil species to date.
Data acquisition and screening of environmental factors
The near-current climate data (1970–2000) and future climate data (2041–2060), including 19 bioclimatic variables with a spatial resolution of 2.5 arcmin (table S1), were downloaded from WorldClim v. 2.1 (www.worldclim.org) (Fick and Hijmans, Reference Fick and Hijmans2017). The BCC-CSM2-MR model has been shown to accurately simulate long-term temperature and precipitation trends in China (Wu et al., Reference Wu, Lu, Fang, Xiaoge, Li, Li, Jie, Zhang, Liu, Zhang, Zhang, Zhang, Wu, Li, Chu, Wang, Shi, Liu, Wei and Liu2019; Xin et al., Reference Xin, Wu, Zhang, Zhang, Li, Zhang, Lu, Fang, Jie and Zhang2019).
Future climate projections for the 2050s were obtained from three Shared Socioeconomic Pathways (SSPs) within the CMIP6 framework: SSP126, SSP245, and SSP585. These scenarios represent varying levels of radiative forcing by the year 2100: 2.6 W/m2 under a sustainability-oriented pathway (SSP126), 4.5 W/m2 under an intermediate development trajectory (SSP245), and 8.5 W/m2 under a fossil-fuelled development scenario with high emissions (SSP585) (Albuquerque et al., Reference Albuquerque, Benito, Macı´as-rodrı´guez and Gray2018; Zhang et al., Reference Zhang, Chen and Xin2019). The projections are based on the BCC-CSM2-MR model, which integrates SSP narratives with Representative Concentration Pathways to simulate future climate conditions (Zhang et al., Reference Zhang, Zhao, W, Li, L and C2022). Climate variable screening is crucial for evaluating species habitat suitability (Sillero, Reference Sillero2011). Pearson correlation analyses were conducted on the 19 environmental variables to reduce collinearity; whenever 2 variables were highly correlated (|r| > 0.8), the more meaningful one was retained. (fig. S1; tables S2 and S3)
Model construction and accuracy evaluation
We used the Biomod2 package in RStudio to model the current and future distributions of RWW and RW based on their distribution points and bioclimatic variable data. Biomod2 integrates ten modelling algorithms (GLM, GBM, GAM, MARS, CTA, ANN, SRE, FDA, RF, Maxent) and is freely available (Cola et al., Reference Cola, Broennimann, Petitpierre, Breiner, DAmen, Randin, Engler, Pottier, Pio, Dubuis, Pellissier, Mateo, Hordijk, Salamin and Guisan2017). For each model, 75% of the data were used for training, and 25% were used for testing (Zhao et al., Reference Zhao, Cui, Sun, Li, Wang, Ye and Fan2021). The dataset was partitioned into training and testing via five repetitions. Additionally, 1000 pseudoabsence points were randomly generated, and this process was repeated three times. Ultimately, 480 models were developed for each insect species. Models with average TSS > 0.75 and AUC > 0.8 were selected for the EM to analyse the distribution in China. Model accuracy was evaluated via the TSS and AUC, with higher values indicating greater accuracy (Allouche et al., Reference Allouche, Tsoar and Kadmon2006; Peterson et al., Reference Peterson, Papeş and Soberón2008).
Based on the EM results, we performed a binary transformation of the potential distributions of each species via the ‘BIOMOD_Ensemble modeling’ function in Biomod2. This generated an ASCII grid with values from 0 to 1, representing the species occurrence probability (p). According to the IPCC-CMIP5 classification, suitability areas were categorised as highly suitable (0.5 ≤ p ≤ 1.0), moderately suitable (0.05 ≤ p < 0.5), lowly suitable (0.01 ≤ p < 0.05), or unsuitable (0 ≤ p < 0.01). We then used ArcGIS spatial tools to map the suitable areas of RWW and RW, including their overlapping areas.
Shift in the potential geographic distribution centre and measurement of the ecological niche
The SDM tool (http://www.sdmtoolbox.org) was used to analyse the migration of RWW and RW distribution centres between current and future climate conditions. This tool was integrated with ArcGIS (version 10.2) for these calculations (Brown, Reference Brown2014). Niche similarity tests were conducted via the ecospat package in R to measure the overlap of ecological niches between the origin and invasion sites of RWW and RW (Broennimann et al., Reference Broennimann, Fitzpatrick, Pearman, Petitpierre, Pellissier, Yoccoz, Thuiller, Fortin, Randin, Zimmermann, Graham and Guisan2012). If the observed D value exceeds the 95% confidence interval of the randomly generated D value, niche similarity is rejected, indicating significant differences (Warren et al., Reference Warren, Glor and Turelli2008). We also generated predicted niche occupancy profiles for key environmental variables using the ‘var.import’ function in the Biomod2.
Results
Occurrence information and selection of environmental variables
We collected data on 638 occurrences of RWW and 388 occurrences of RW. After filtering with ENMTools software, we retained 583 occurrences of RWW and 347 occurrences of RW (fig. S2). Following the screening process, we selected 7 (RWW: Bio02, Bio03, Bio08, Bio10, Bio13, Bio14, and Bio15; RW: Bio03, Bio04, Bio08, Bio10, Bio13, Bio14, and Bio15) environmental variables that were strongly correlated with the occurrence of both RWW and RW. These bioclimatic variables effectively encapsulate key ecological factors, including annual trends, seasonality, and extreme or limiting environmental conditions (table S3).
Model accuracy evaluation
The performance of ten individual model algorithms was assessed, and three models were selected for ensemble modelling for RWW (CTA, GBM, RF) (fig. 1A–F) and RW (GAM, GBM, RF) (fig. 1G–L). These models effectively predict the impacts of climate change on the spread of both rice species. The TSS and AUC values for all 12 EMs exceeded 0.85, indicating high accuracy. This high accuracy reduces uncertainty and enhances the reliability of the predicted geographic distributions of RWW and RW.
Evaluation indices of individual prediction models for Lissorhoptrus oryzophilus and Echinocnemus squameus in China during different occurrence periods and future climate scenarios. (A: 1988–2003, B: 1988–2013, C: 1988–2023, D: SSP126, E: SSP245, F: SSP585, G: 1954–2003, H: 1954–2013, I: 1954–2023, J: SSP126, K: SSP245, l: SSP585).Notes: AUC, area under the receiver operating characteristic curve; TSS, true skill statistic.
Prediction of the potential distributions of RWW and RW under current climatic conditions
Model simulations indicate that suitable areas for RWW in China are located primarily in South, North, Central, Northeast, and Northwest China. High-suitability areas are concentrated in Liaoning, central Jilin, southeastern Hebei, Beijing, Tianjin, Shandong, Jiangsu, Anhui, Zhejiang, northwestern Jiangxi, Guizhou, Sichuan, Fujian, and Chongqing (fig. 2A–C). Over time, the area suitable for RWW has continued to expand from 423.30 × 10⁴ km2 during 1988–2003 to 477.40 × 10⁴ km2 in 1988–2013, and further extended to 538.53 × 10⁴ km2 by 1988–2023. This represents a cumulative gain of 115.22 × 10⁴ km2, accounting for 11.98% of China’s total land area (table S4). This trend reflects spatial diffusion from core suitable areas to peripheral regions, with notable improvements in suitability observed, particularly in parts of Northwest China and Northeast China.
Current geographical distributions of Lissorhoptrus oryzophilus and Echinocnemus squameus in China, as predicted via ensemble models (EMs). (A–C: Lissorhoptrus oryzophilus; D–F: Echinocnemus squameus).Notes: TA, total suitable area; MA, marginal area; SA, suitable area; OA, optimal suitable area.
Suitable areas for RW are located primarily in South, East, Central, Southwest, and Northeast China. High-suitability regions are concentrated in southern Jiangsu, Zhejiang, Anhui, Fujian, Jiangxi, Guangdong, Guangxi Zhuang Autonomous Region, Hainan, southeastern Guizhou, Chongqing, and eastern Hubei and Hunan (fig. 2D–F). Over time, the total suitable area for RW has shown a consistent decline, from 495.55 × 10⁴ km2 during 1954–2003, to 440.10 × 10⁴ km2 in 1954–2013, and further decreased to 376.06 × 10⁴ km2 by 1954–2023. This corresponds to a net reduction of 119.49 × 10⁴ km2, accounting for approximately 12.42% of China’s total land area (table S4). Spatially, the contraction occurred not only within core high-suitability zones but also within peripheral and ecologically marginal areas, particularly in parts of the southwest and northeast.
Prediction of suitable areas for RWW and RW under future climatic conditions
For RWW, by 2041–2060, the total suitable areas for RWW under the three climate change scenarios (SSP126, SSP245, and SSP585) were 526.16 × 104 km2, 481.37 × 104 km2, and 509.37 × 104 km2, respectively. The optimal areas for the RWW under the three climate change scenarios in China were 209.77 × 104 km2, 198.41 × 104 km2 and 215.48 × 104 km2. The areas with low suitability were 183.16 × 104 km2, 158.97 × 104 km2 and 169.16 × 104 km2, respectively (table S5). In contrast, the areas of low suitability and optimal suitability were the highest in the SSP126 scenario and the lowest in the SSP245 scenario. Overall, no significant increase in the total area of suitable areas in China under the three scenarios was observed, which was similar to that in 2023 (fig. 3A–C).
Potential geographical distributions of Lissorhoptrus oryzophilus and Echinocnemus squameus in China in the 2050s, as predicted via ensemble models (EMs). (A–C: Lissorhoptrus oryzophilus; D–F: Echinocnemus squameus).Notes: TA, total suitable area; MA, marginal area; SA, suitable area; OA, optimal suitable area.
For RW, by 2041–2060, the total areas suitable for SBPH under the three climate change scenarios (SSP126, SSP245, and SSP585) were 508.06 × 104 km2, 425.51 × 104 km2, and 349.79 × 104 km2, respectively. The optimal areas were 209.47 km2, 130.01 × 104 km2 and 160.13 × 104 km2, respectively (table S5). The areas with low suitability were 177.82 × 104 km2, 157.08 × 104 km2, and 110.21 × 104 km2, respectively. The areas of low suitability and optimal suitability were the highest in the SSP126 scenario. Overall, the total area of suitable areas for RW decreases gradually, as do carbon emissions, across these three future climate scenarios (fig. 3D–F).
Niche overlap analyses
According to the EMs, the overlapping distribution of RWW and RW is mainly in the central, eastern, southern, northern, and northeastern parts of China. The overlapping area was 376.59 × 102 km2 from 1954–2003, accounting for 88.9% of the total area of RWW and 76.0% of the total area of RW. By 2013, it had increased to 384.14 × 102 km2, with the RWW proportion decreasing to 80.4% and the RW proportion increasing to 87.3%. By 2023, the overlapping area decreased to 364.64 × 102 km2, with the RWW proportion further decreasing to 67.7% and the RW proportion increasing to 97.0%. Future climate projections for the 2050s suggest that the overlapping area of RWW and RW will be 454.25 × 102 km2 under the SSP126 scenario, representing 89.4% of the total area of RWW and 86.3% of the total area of RW. This indicates a relatively stable habitat. However, under the SSP245 scenario, the overlapping area is expected to decrease to 378.28 × 102 km2 and further decline to 333.30 × 102 km2 under the SSP585 scenario. Overall, the overlapping area is projected to shrink gradually due to global warming, with suitable habitats shifting southwards toward Henan Province and eastwards toward Chongqing and southeastern Sichuan (fig. 4 and table 1). In conclusion, climate change notably affects the distributions of both species, with the overlapping area remaining stable only under SSP126 but facing challenges under the other scenarios.
Overlapping areas between Lissorhoptrus oryzophilus (A) and Echinocnemus squameus (B) in China under current and future climates (2050s). (A: 1954–2003; B: 1954–2013; C: 1954–2023; D: SSP126; E: SSP245; F: SSP585).
Overlapping area (OA) of potential geographical distribution of the Lissorhoptrus oryzophilus (A) and Echinocnemus squameus (B) under the current and future (2050s) climate conditions (unit: × 104 km2)
Notes: OA, overlapping are; TA, total suitable area.
We calculated the degree of ecological niche overlap between RWW and RW, with values ranging from 0 (no overlap) to 1 (complete overlap) (Bachiller et al., Reference Bachiller, Gim´enez, Albo-Puigserver, Pennino, Marí-Mena, Esteban, Lloret-Lloret, Bellido and Coll2021). High ecological niche overlap was observed between RWW and RW, with D and I values of 0.658 and 0.816, respectively. The null hypotheses of RWW and RW, which are based on bioclimatic variables of native and invasive areas, were rejected (p = 0.09). Compared with RW, RWW has a slightly expanded niche. PC1 accounted for 54.37% of the environmental variation, whereas PC2 explained 20.01% (fig. 5).
Migration of the centre of the potential geographical distribution of Lissorhoptrus oryzophilus (A) and Echinocnemus squameus (B). Color coding for distribution centres: Green circle: Current climate scenario (1954/1988–2003); Yellow circle: Current climate scenario (1954/1988–2013); Red circle: Current climate scenario (1954/1988–2023); Blue circle: Future climate scenario (2050s SSP126); Purple circle: Future climate scenario (2050s SSP245); Pink circle: Future climate scenario (2050s SSP585).
Shift in potential geographic distribution centres and significant environmental variables
Under the current climate conditions, the distribution centre of the suitable area for RWW shifted from central Shaanxi (1988–2003) to southeastern Shanxi (1988–2013) and then moved slightly southwestward by 2023. Under the future climate scenario, the distribution centre is expected to shift southeastward under SSP126 but migrate toward southwestern Shaanxi under SSP245 and SSP585 (fig. 6A). For RW, the distribution centre moved from southwestern Shaanxi (1954–2003) to northeastern Chongqing (1954–2013) and then northeastward to northwestern Hubei by 2023. Under future scenarios, the distribution centre is projected to continue eastwards to central-eastern Shaanxi under SSP126 but to shift northeastward under SSP245 and SSP585 (fig. 6B).
Ecological niches and niche similarity of Lissorhoptrus oryzophilus and Echinocnemus squameus. (A) Ecological niches under current climatic conditions. (B) Ecological niche similarity between L. Oryzophilus and E. Squameus. Color coding: red: climatic niche of L. Oryzophilus; green: climatic niche of E. Squameus; blue: niche overlap; red arrows: centroids of the realised ecological niches of each species.
For RWW, we used CTA, GBM, and RF to estimate the contribution of each environmental variable and analysed those that significantly influence the potential distribution. Across the current and future climate scenarios, the precipitation of the wettest month (Bio13) consistently had the greatest impact, with contributions of 0.286 (1988–2003), 0.209 (1988–2013), 0.165 (1988–2023), 0.246 (SSP126), 0.251 (SSP245), and 0.310 (SSP585). The mean temperature of the warmest quarter (Bio10) also plays a significant role (fig. 7 A, B). For RW, in the current and SSP126 climate scenarios, Bio10 was the most influential variable. In the SSP245 and SSP585 climate scenarios, Bio14 (precipitation of the driest month) had the greatest impact (fig. 7 C, D).
Contribution of each environmental variable to Lissorhoptrus oryzophilus in the current (A) and future (B) scenarios and to Echinocnemus squameus in the current (C) and future (D) scenarios.
Discussion
This research aimed to predict habitat suitability and ecological niche shifts for both invasive and native species of RWs in China. Using an EM, we explored habitat suitability changes and niche overlaps between both species. Compared with the current climate, the suitable area for RWW showed less fluctuation across the three future climate scenarios, with a trend toward stabilisation and expansion mainly toward the northeast and southwest. In contrast, the suitable area for RW expanded primarily toward the northeast, whereas the total suitable area generally decreased. This study revealed the greatest ecological niche overlap between RWW and RW, although the environmental variables influencing their distributions differed. This study offers valuable insights into how climate change may affect the distribution and ecological interactions of RWs in China and highlights the potential ecological and agricultural implications of niche overlap between native and invasive species.
Model accuracy evaluation
SDMs integrate species occurrence data with biotic and abiotic variables to predict the potential distributions of invasive plant species (Elith and Leathwick, Reference Elith and Leathwick2009). Although widely used, individual SDMs are susceptible to underfitting or overfitting (Peterson et al., Reference Peterson, Papeş and Eaton2007). EMs, which combine multiple modelling algorithms, can improve predictive accuracy and reduce bias (Araújo and New, Reference Araújo and New2007; Naimi and Araújo, Reference Naimi and Araújo2016). Commonly used EM platforms include Biomod (Thuiller et al., Reference Thuiller, Lafourcade, Engler and Miguel2009), ModEco (Guo and Liu, Reference Guo and Liu2010), OpenModel (De Souza Muñoz et al., Reference De Souza Muñoz, De Giovanni, De Siqueira, Sutton, Brewer, Pereira, Canhos and Canhos2011), mopa (Iturbide et al., Reference Iturbide, Bedia, Herrera, Del Hierro, Pinto and Manuel Gutierrez2015), and SDM (Naimi and Araújo, Reference Naimi and Araújo2016), each with specific strengths and limitations that depend on user expertise and objectives (Mugo and Saitoh, Reference Mugo and Saitoh2020). In this study, we developed EMs based on high-performance SDMs for each species. The resulting models showed high predictive performance, with average AUC values above 0.9 and TSS values exceeding 0.8 for both RWW and RW, indicating that the EMs provide more reliable predictions of potential species distributions compared to single-model approaches.
Despite their advantages, SDMs and EMs have several limitations that should be considered. First, model uncertainty remains a fundamental challenge, as different algorithms can yield divergent predictions even when trained on identical data (Araújo and New, Reference Araújo and New2007). Second, the accuracy of predictions heavily depends on the quality and representativeness of species occurrence data. Biased or incomplete distribution records may lead to erroneous projections (Guisan et al., Reference Guisan, Thuiller and Zimmermann2017). Third, the selection of environmental variables is critical; inappropriate variables may cause models to deviate from biological reality (Elith and Leathwick, Reference Elith and Leathwick2009). Fourth, model complexity must be carefully managed to avoid overfitting, particularly when sample sizes are small, as excessively complex models may perform poorly when generalised to new areas (Hastie et al., Reference Hastie, Tibshirani and Friedman2009). Furthermore, key drivers such as terrain heterogeneity and anthropogenic factors are not always sufficiently incorporated; however, they can significantly influence species distributions (González-Moreno et al., Reference González-Moreno, Diez, Ibáñez, Font and Vilà2014; Abdulwahab et al., Reference Abdulwahab, Hammill and Hawkins2022). Finally, EMs often overlook species interactions, such as competition and predation, relying solely on environmental factors and potentially introducing bias into the results (Wisz et al., Reference Wisz, Pottier, Kissling, Pellissier, Lenoir, Damgaard, Dormann, Forchhammer, Grytnes, Guisan, Heikkinen, Høye, Kühn, Luoto, Maiorano, Nilsson, Normand, Öckinger, Schmidt and Svenning2013). In this study, the limited occurrence point data and influencing factors were unable to capture the species’ full fundamental niche. To address these issues in future research, we recommend that future studies incorporate distribution data from the species’ native range in North America to further refine the model. In addition, we integrate multisource data (e.g. remote sensing, soil properties, land use, and dispersal constraints) with ecological mechanism-based modelling. Approaches that account for biotic interactions and dynamic processes such as climate change and invasion pathways could further enhance model realism.
Prediction of the potential distributions of RWW and RW in the present and future
As global temperatures rise, the structure and function of terrestrial ecosystems may undergo substantial alterations, leading to pronounced shifts in the extent and distribution of biological areas (Lawler, Reference Lawler2009). Numerous studies have demonstrated that ecological niche models (ENMs) are effective tools for evaluating the effects of climate change on insect populations (Faleiro et al., Reference Faleiro, Nemésio and Loyola2018; Müller et al., Reference Müller, Dias, Gottschalk, Garcia and Nava2019; Tepa-Yotto et al., Reference Tepa-Yotto, Gouwakinnou, Fagbohoun, Tamò and Sæthre2021; Azrag et al., Reference Azrag, Mohamed, Ndlela and Ekesi2022; Jin et al., Reference Jin, Yu, Zhao, Xian, Jing, Yang, Lu and Liu2022). Therefore, based on the current distributions of RWW and RW, we predicted their potential redistribution in response to climate change under three twenty-first-century scenarios (SSP126, SSP245, and SSP585). The results revealed that the distribution range of suitable areas for RWW was relatively stable across all three future climate scenarios, with the total extent peaking under SSP126 and being lowest under SSP245. For RW, the total adaptive area is projected to decrease with increasing carbon emissions, being largest under SSP585 and smallest under SSP245. Compared with those under the current distribution, the areas suitable for RW are expected to decrease under SSP126 and SSP245 but expand under SSP585.
The present study predicts that RWW and RW have extensive potentially suitable areas in China, covering more than 90% of the rice cultivation regions. The highly suitable areas are mainly concentrated in central and southern China, with limited distributions around the Bohai Bay region. These findings align closely with the research of Qi et al. (Reference Qi, Gao, Huang and Lyu2012), who reported that the distribution of RWW had expanded to 20 provinces, autonomous regions, and municipalities, showing no signs of slowing down. It has spread to Yunnan, Jiangxi, Heilongjiang, Hubei, Guizhou, and Xinjiang, with a noticeably accelerating trend. Although regions such as Guangdong, Guangxi, and Hainan in southern China, Sichuan and Chongqing in southwestern China, Jiangsu and Shanghai in eastern China, and Henan in central China have not yet recorded invasions, their adjacent areas have already been affected, indicating a high risk of introduction (Qi et al., Reference Qi, Gao, Huang and Lyu2012). The suitability distribution projected by our ENM further supports earlier observations regarding the ongoing expansion and high risk of spread of RWW. This study highlights the continued threat of invasion in southern, central, and parts of eastern China, underscoring the urgent need for enhanced monitoring and preventive measures to curb further spread and establishment of this pest.
Shift in the potential geographic distribution centre
RWW has multiple modes of transmission. It can spread to nearby areas through its own activities, such as crawling and flying, as well as through natural dispersal mechanisms such as wind, rain, and water flow (Zhang et al., Reference Zhang, Yu and Xu1994), enabling short-distance spread. However, long-distance transmission primarily occurs through human activities, with transportation vehicles and the movement of straw and rice being the main means of long-distance spread (Zhai et al., Reference Zhai, Zheng, Shang and Cheng1999; Wang et al., Reference Wang, Shang, Li and Cheng2005). In this study, based on migration trend analysis, under current climate conditions, the distribution centre of the suitable area for RWW shifts from central Shanxi to southeastern Shaanxi, whereas the distribution centre for RW moves from western Chongqing to northwestern Hubei. Under future climate scenarios, the distribution centre of the suitable area for RWW within Shanxi is projected to move southeastward and southwestward, whereas the distribution centre for RW is expected to migrate northwards and northeastwards to the central-eastern part of Shanxi and the central-western part of Henan. Overall, these shifts reflect significant changes in the geographic distribution centres of both species under different climate scenarios. Two significant peak periods in the expansion of RWW in China were identified. The first peak occurred between 1988 and 1993, during which RWW rapidly spread from coastal rice areas in North China, migrating southwards into southern rice-growing regions. The second peak occurred between 2001 and 2003, with notable westwards expansion into the southwestern and central-southern regions, thereby further increasing its range of influence. These two peak periods not only highlight the adaptability of RWW to different climatic conditions but also reflect the close relationship between its spread and factors such as climate and changes in agricultural production patterns (Jiang, Reference Jiang, Hao, Guo and Zhang2009). This study revealed that the expansion of RWW has accelerated in recent decades, particularly in North China, East China, and Southwest China. These areas have experienced the expansion of suitable areas and the migration of RWW. Against the backdrop of current climate change, the distribution centre for RWW is undergoing dynamic shifts, showing a trend of moving toward regions with more favourable climatic conditions.
The northwards and upwards shifts in the distribution centres of these pests suggest that they are expanding into regions previously unaffected, where local ecosystems – such as wetland and aquatic communities – may lack evolved resistance mechanisms, posing a substantial threat to biodiversity (Bellard et al., Reference Bellard, Bertelsmeier, Leadley, Thuiller and Courchamp2012; Pecl et al., Reference Pecl, Araújo, Bell, Blanchard, Bonebrake, Chen, Clark, Colwell, Danielsen, Evengard, Falconi, Ferrier, Frusher, Garcia, Griffis, Hobday, Janion-Scheepers, Jarzyna, Jennings and Williams2017). Moreover, the coexistence and niche overlap of the two species, particularly under future climate conditions, indicate that rice fields may experience concurrent infestations by both pests. While larvae of RWW feed on roots, both adults and larvae of RW damage leaves and stems; such combined attacks are likely to produce synergistic effects, leading to significantly greater crop damage than that caused by either species alone. In newly developed rice-growing areas, farmers’ lack of experience in pest identification and control may lead to missed optimal intervention windows, resulting in overreliance on chemical pesticides and consequently increased risks of pollution to local soil and water environments. Therefore, monitoring the spread and biological characteristics of RWW, especially in newly emerging expansion areas, is critical and cannot be overlooked in future pest control efforts.
The role of key environmental factors in driving distributions
ENMs are employed to explore the relationships between species distributions and their corresponding environmental variables. These models are based on the assumption that the species‒environment relationship can elucidate and forecast both current and future spatiotemporal distributions of species (Mulieri and Patitucci, Reference Mulieri and Patitucci2019). By analysing the contribution of each variable, we identified the key factors influencing the distributions of RWW and RW. Under both current and future climate scenarios, the wettest monthly rainfall (Bio13) had the greatest impact on the distribution of RWW, whereas the warmest quarterly average temperature (Bio10) and driest monthly rainfall (Bio14) had the greatest influence on RW. Variable rainfall significantly affects insect populations. For example, increased summer rainfall and drought conditions impact soil-dwelling wireworms (Agriotes lineatus), which are significant pests of crops such as potatoes and corn (Staley et al., Reference Staley, Hodgson, Mortimer, Morecroft, Masters, Brown and Taylor2007). Climate change may exacerbate this issue, potentially intensifying the challenges posed by wireworms (Johnson et al., Reference Johnson, Anderson, Dawson and Griffiths2008). Additionally, climate change markedly alters the geographical distribution of pests, with low temperatures generally playing a more critical role than high temperatures in shaping pest distribution patterns (Hill, Reference Hill1988). The spread of RWW is constrained by cold winter temperatures in Heilongjiang and Jilin Provinces, as well as by high altitudes in Sichuan and Yunnan Provinces, according to CLIMEX software analysis (Chen et al., Reference Chen, Chen and Zhou2005). It is hypothesised that the combination of summer dryness and cold winter temperatures inhibits the spread of the weevil in northwestern regions, including Inner Mongolia and Gansu. As an invasive species, the distribution of RWW is shaped primarily by its sensitivity to water requirements (Bio13). In contrast, the native RW species shows broader adaptability to local heat conditions (Bio10) and drought stress (Bio14). This finding not only explains the current distribution patterns but also provides valuable insights into how future climate change – such as increased precipitation variability and warming winters – may affect the population dynamics of both species. For example, drier summers (decreased Bio14) in certain regions may limit the spread of RW, while simultaneously intensifying expansion or contraction pressures on RWW, which is more dependent on water availability.
Climate change also alters the geographic ranges of numerous pest species, with the global fruit trade, transportation of plant seedlings, and packaging materials acting as key conduits for the inadvertent transmission of IAS to new regions (Yan et al., Reference Yan, Wang, Feng, Wan and Chang2017). This dispersal can significantly impact agricultural production, particularly for crop pests (Meynard et al., Reference Meynard, Migeon and Navajas2013). Invasive insect pest species are likely to experience increased geographic ranges, elevated population densities, and enhanced voltinism due to anticipated climate change scenarios (Hill et al., Reference Hill, Bertelsmeier, Clusella-Trullas, Garnas, Robertson and Terblanche2016), posing significant challenges to sustainable agricultural production (Ziska et al., Reference Ziska, Blumenthal, Runion, Hunt and Diaz-Soltero2011). Compared with native insects, invasive species often possess broader bioclimatic ranges, allowing them to locate a wider array of suitable areas (Walther et al., Reference Walther, Roques, Hulme, Sykes, Pyšek, Kühn, Zobel, Bacher, Botta-Dukát, Bugmann, Czúcz, Dauber, Hickler, Jarošík, Kenis, Klotz, Minchin, Moora, Nentwig, Ott, Panov, Reineking, Robinet, Semenchenko, Solarz, Thuiller, Vilà, Vohland and Settele2009). Consequently, it remains crucial to comprehend and forecast insect responses to climate change across various scenarios.
Pest management implications
RWW and rice water (RW) are highly destructive pests of rice, necessitating the development of effective and sustainable control strategies. In this study, for the first time, an EM was employed to analyse changes in suitable regions for RWW and RW at different stages and to identify potential distribution areas under current and future climate change scenarios. We highlight both the expansion of suitable areas and potential new invasion zones and provide a critical spatial framework for implementing proactive and targeted pest management strategies. Specifically, these findings can be directly integrated into Integrated Pest Management (IPM) programs. For example, in areas identified as newly suitable or high risk, monitoring efforts should be prioritised and enhanced. Biological control, a cornerstone of IPM, is a particularly promising strategy for managing these invasive pests (Nair and Peterson, Reference Nair and Peterson2023). Our model outputs can guide the deliberate introduction or conservation of natural enemies by pinpointing regions where they are most likely to establish and be effective. Further research into the ecological relationships between RWWs and key natural enemies, informed by our distribution predictions, will be invaluable for advancing these methods.
Rising temperatures associated with climate change are expected to accelerate population growth and metabolic rates in insect pests, leading to significant crop losses and increased reliance on chemical insecticides (Porter et al., Reference Porter, Parry and Carter1991; Deutsch et al., Reference Deutsch, Tewksbury, Tigchelaar, Battisti, Merrill, Huey and Naylor2018; Skendžić et al., Reference Skendžić, Zovko, Živković, Lešić and Lemić2021). Therefore, our findings on future distribution shifts underscore the urgency of developing and adopting sustainable alternatives to chemical control. However, research has also shown that agricultural productivity can increase if effective pest management strategies are implemented in response to climate change (Müller, Reference Müller2013). The EM developed here provides a decision-support tool for such strategic implementation. Enhancing resilience and adaptation through improved agricultural practices and a better understanding of the future distribution of agricultural pests are essential for informed decision-making aimed at mitigating the impacts of climate change (Vermeulen et al., Reference Vermeulen, Aggarwal, Ainslie, Angelone, Campbell, Challinor, Hansen, Ingram, Jarvis, Kristjanson, Lau, Nelson, Thornton and Wollenberg2012). Our study contributes to this understanding by offering a predictive outlook that can inform long-term agricultural planning and policy.
In practical terms, for preemptive exclusion, our maps are essential for biosecurity planning. In areas that are potentially suitable but not yet infested, establishing official quarantine zones is vital to prevent further spread. It is also critical to implement stringent quarantine measures for the importation of seedlings, straw, and soil from infested to noninfested regions and to avoid the use of quarantined straw as packaging material (Mulcahy et al., Reference Mulcahy, Wilson and Reagan2022).
Conclusions
This study used an EM to examine changes in habitat suitability, shifts in suitable area centres, and niche overlap for RWW and RW. The results show that the potential distributions of both pests are primarily concentrated in central, eastern, southern, northern, and northeastern China, with high niche overlap in the central, eastern, and southern regions. The precipitation of the wettest month (Bio13) is a key factor influencing the RWW distribution, whereas the mean temperature of the warmest quarter (Bio10) and the precipitation of the driest month (Bio14) significantly influence the RW distribution. Overall, the ensemble model effectively identifies optimal regions for RWW and RW spread, offering a solid theoretical basis for pest control strategies.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0007485325100515.
Acknowledgements
We are grateful to Dr. Bin Lv, at the Chengdu Institute of Biology, Chinese Academy of Sciences, Chengdu, China, for providing valuable comments and suggestions. This research was supported by the Science & Technology Fundamental Resources Investigation Program (Grant No. 2023FY100500) and the Natural Science Foundation of Liaoning Province of China (No. 2023-MS-209).
Author contributions
XPS and XYW conceived the research. XPS and MLY conducted the experiments. XYW contributed material. XPS, JYW and YLF analysed the data and conducted the statistical analyses. XPS and FLQ wrote the manuscript. XYW secured funding. All the authors have read and approved the manuscript, and they participated sufficiently in the work to take public responsibility for appropriate portions of the content.
Competing interests
The authors declare that they have no conflicts of interest.
Data availability statement
The data will be made available on request.
