Two sets of input variables, including environmental variables with and without wildfire and environmental variables, were employed with five machine learning algorithms to predict the spatial distribution of cogon grass. The MaxEnt model with environmental variables and wildfires as input variables demonstrated the best prediction performance among the five algorithms, with an AUC value of 0.92, a TSS value of 0.73, a kappa value of 0.68, a sensitivity value of 0.79, and a specificity value of 0.81 (Table 2). Consequently, the outputs of the MaxEnt model were used to predict the spatial distribution of cogon grass and the habitat suitability of this species worldwide. Considering the values of the chosen evaluation metrics, the results indicated excellent prediction performance and strong agreement between the observations and predictions. The results also suggested that wildfire occurrence is the most significant ecological factor influencing the expansion of cogon grass habitats.

Suitable conditions for cogon grass were analyzed across various global land cover types, revealing a projected significant expansion under future environmental change. Currently, the highest proportion of cogon grass coverage is observed in artificial lands including human settlements, fragmented landscapes, and industrial areas (42.5% of these lands globally), followed by agricultural land (35.3%) and mixed land cover types (24.7%) (Table 6). Under the SSP5-8.5 scenario, coverage is expected to increase to 51.3% in artificial areas, 44.7% in agricultural land, 28.1% in forests, and 19.4% in grasslands, indicating growing ecological threats to native biodiversity and ecosystem stability. Wetlands and water bodies, including marshy and poorly drained soils, also show a steady increase from 12.4% to 17.7%, suggesting cogon grass’s strong adaptability to moist environments. Although non-vegetated areas maintain very low coverage, a slight increase from 0.1% to 0.3% reflects its minor expansion potential in harsh conditions. Overall, these trends highlight the ecological plasticity of cogon grass and its increasing potential to invade diverse landscapes under climate change.

Our analysis of the distribution models and environmental variables, revealed that the MaxEnt model with environmental variables and wildfire provided the most accurate projection of the current and potential future ranges of cogon grass (Table 2), and that wildfire is the greatest contributing ecological driver for determining the habitat suitability of cogon grass (Table 1). We found that 41 countries and 70 countries currently contain extremely suitable for cogon grass (Fig. S2). At least 17 countries, including Angola, Congo, and Suriname, which currently present unsuitable or poorly suitable, are projected to shift to extremely suitable conditions in the future under the SSP5-8.5 scenario (Table 5). Analysis of cogon grass distribution across different land cover categories shows that the species currently is most prevalent in disturbed areas such as urban, semi urban, and agricultural areas (Table 6). The high habitat suitability of cogon grass in artificial and agricultural lands arises because these environments reduce competition and provide open niches. The grass’s fire resilience, extensive rhizome network, efficient resource use, allelopathy, and rapid colonization enable it to dominate and persist. Under the SSP5-8.5 scenario, suitable conditions for cogon grass are projected to expand further within these land cover classes and also into forests and grasslands, underscoring the growing threat it poses to native biodiversity and ecosystem stability. Furthermore, cogon grass is found across a wide range of grassland ecosystems, including alpine meadows, temperate steppes, tropical savannas, and semi-desert scrublands globally. This wide ecological range increases the risk to various grassland types. In particular, regions such as the Asian alpine and subtropical zones, as well as the savannas of Africa and South America, are especially vulnerable to future invasions.

Weed species like cogon grass are rarely at equilibrium, exhibiting rapid expansion across new environments²⁵. This dynamic behavior makes it critical to use predictive models that capture nonequilibrium patterns. Our MaxEnt model identified key climatic and environmental drivers, including temperature, precipitation, wildfires, and soil characteristics, that determine habitat suitability. Results indicate that cogon grass thrives under warm climates with variable rainfall and frequent disturbances, such as wildfires. These conditions create opportunities for rapid colonization and expansion into previously unsuitable regions, highlighting the species’ potential to alter ecosystem composition and challenge management efforts across diverse habitats.

Global climate change is increasing the frequency, intensity and duration of wildfires by altering weather patterns, with increasing temperatures and prolonged droughts impacting native vegetation, ecosystem function, and forest resilience^11,26. Wildfires are major ecological disturbances that shape the structure and composition of many ecosystems, and play a crucial role in shaping plant invasions by serving both as a regulator of natural ecosystems and a driver of pyrogenic succession for alien plant species ²⁷.

Alien and invasive species are particularly opportunistic in utilizing postfire habitats because of the presence of open canopies, a reduced number of competitive native species, and improved soil conditions ⁸. Numerous studies have revealed the rapid invasion of alien plant species following wildfires in both grassland and forest ecosystems ^11,28. This phenomenon can lead to landscape transformation and a reduction in the abundance and diversity of native species.

Cogon grass exhibits both fire resistance and fire-altering traits, allowing quick regeneration in open spaces of postfire landscapes ⁷. The species exhibits high flammability, containing three times the fine-fuel load of many native grasses and creating continuous fuel beds, which lead to the occurrence of frequent fires with greater vertical flame heights^1,7. This species provides a self-reinforcing cycle of fires that can destroy native seed banks, increase the mortality of native plants, disrupt ecosystem functions and allow it to grow and expand ⁵.

Previous studies have indicated that wildfires increase the abundance of cogon grass in burned areas across the United States and other parts of the world through various mechanisms. These mechanisms include increased disturbance, changes in soil chemistry and resource availability, and altered competitive dynamics between invasive and native plant species^9,29. Cogon grass thrives in disturbed environments, particularly after fire events, which creates conditions favorable for its spread.

The United States exhibits a high risk of wildfires¹², which is amplified by the presence of cogon grass. Increasing temperatures and severe drought lead to the presence of dry vegetation and soil, creating highly flammable fuel that, when combined with high winds, exacerbates the potential for wildfires—particularly in the southeastern and western regions of the United States^12,30. In early 2025, southern California experienced a devasting wildfire, followed by outbreaks in Texas and many parts of the southeastern United States, which exhibited abnormal fire activity. These wildfires not only cause damage to native biodiversity but also create conditions for long-term ecological shifts driven by increased plant invasions ³¹.

Cogon grass also contributes to the mortality of juvenile native plants, which has led to a 40–60% decrease in plant species richness and endemic plant populations in the southeastern USA, particularly in longleaf pine woodlands in Mississippi and the Blackwater River State Forest in Florida ^7,8. The rate of fuel accumulation following fires has been measured on a broad scale, and over time, fires in sandhill habitats have caused a shift in the composition of species-rich pine savannas to grasslands dominated by the nonnative I. cylindrica ⁷. Similarly, in Blackwater River State Forest, Florida, USA, areas that were burned or experienced significant biomass removal following a hurricane supported a greater number of cogon grass patches and larger patch sizes ⁸. Similarly, in northern California, repeated high-severity wildfires have increased the invasion risk by several nonnative species, including cogon grass, which competes for resources and alters ecosystem dynamics ³². Our study revealed that wildfire occurrence could increase the area of suitable habitats for cogon grass by 11% in North America, particularly in the southeastern United States, spanning from the coastal regions of Florida to Mississippi, Kentucky, and North Carolina (Fig. 2A and Table S3).

Projected increase in wildfire activity across various regions of Africa significantly elevates the invasion risk of cogon grass. This phenomenon is particularly evident in areas exhibiting climate change, land cover change, and the occurrence of nonnative vegetation. Recent studies have shown that tropical forests in Central Africa and western Africa, which are traditionally resistant to fires, now demonstrate frequent wildfire activities due to climate change and deforestation ³³. Several countries in Africa, e.g., Angola, Mozambique, the Central African Republic, Ghana, Nigeria, and South Sudan, exhibit over 5000 wildfire ignitions annually, and average annual burned areas above 5% of the national land surface area (from 2012 to 2024; Table 5). Global climate change and increased wildfire create more favorable conditions for cogon grass in such countries. Our study revealed that the greatest proportion of change in the habitat suitability of cogon grass occurred in Central and South Africa, and that several countries in at high risk of invasion.

The Congo River Basin in Africa is one of the largest biodiversity hotspots. The riverine forest of the basin holds nearly 10% of the world's forest-based carbon and supports the livelihoods of more than 50 million people ³³. Cogon grass can dominate landscapes in the Congo Basin, outcompeting native flora and altering ecosystem dynamics. In South Africa, the invasion of alien plants has been linked to increased fire activity. For example, the Knysna wildfire of 2017 allowed invasion of the Fynbos ecosystem by several invasive species, including cogon grass, in the southwestern and southern parts of South Africa ^27,34.

South America is one of the continents with the most frequent wildfires, and most fires are anthropogenic in origin, occurring mostly in tropical regions during the dry season ³⁵. The increase in wildfire frequency across South America can increase the invasion risk of cogon grass. Our model revealed the wildfire-induced invasion risk of cogon grass in many countries in the northern, central and southern parts of South America, e.g., Brazil, Venezuela, Colombia, Bolivia and Argentina. Our wildfire model results indicated an increase in the area of suitable habitats in South America of 28% compared with the estimate without wildfires. Studies on fire ecology and its role in determining the invasion success of alien plants in South America have not yet been published ³⁶. Therefore, these results may be useful for estimating the impacts of wildfires on native biodiversity and invasive species on a broad scale in South America.

Global climate change, through fluctuations in the annual temperature and altered precipitation patterns, can weaken metabolic processes, reduce flowering and seed production, and decrease the growth and productivity of native plants, thereby reducing their competitive ability against invasive species ^37,38. These climatic changes tend to favor invasive species, as they generally possess wider physiological tolerance ranges, extended phenologies, and greater competitive abilities, allowing them to establish and spread in new ecosystems ³⁹. Similarly, isothermality influences plant invasion dynamics by regulating temperature stability and the overlap of climatic niches between invasive and native species. Native species typically prefer stable thermal environments, whereas invasive species are often better suited to cope with daily temperature fluctuations ⁴⁰. As a result, lower isothermality creates conditions that align more closely with the climatic requirements of invasive plants, thus promoting their successful establishment and competitive advantage ^25,41.

Our study indicated that the habitat suitability of cogon grass spans a wide range of climatic variables, with mean annual temperatures ranging from 2.6–30.4°C, annual precipitation amounts ranging from 31 to 4,784 mm, and isothermality values between 20.12 and 90.05 under the current climatic conditions. These findings suggest that cogon grass can persist across a broad spectrum of future climatic conditions ⁹. The species exhibits variable cold tolerance depending on its growth stage and environmental conditions. Mature, dormant plants can survive at temperatures as low as -14°C and can tolerate 1–5 days of freezing at -5°C to -14°C. However, prolonged soil freezing exceeding 10–14 days may be lethal, particularly for young or nondormant plants ⁵. This is why large areas in the northern parts of Asia, Europe, and North America (e.g., Russia, Mongolia, Iceland, Sweden, Norway, and the United Kingdom) currently exhibit unsuitabile or very poor habitat suitability.

Under global climate change scenarios based on SSPs, the Earth's annual mean temperature is projected to increase by 1.8–4.4°C relative to 1850–1900 levels ⁴². Future precipitation patterns are predicted to change, with increases in some regions that currently exhibit arid climatic conditions and decreases in regions with high precipitation ⁴³. The rate of precipitation is expected to change by -0.2% to + 4.7% under SSP1-2.6 and by 0.9% to 12.9% under SSP5-8.5⁴². With a warming climate, precipitation is projected to increase at high latitudes (e.g., the average precipitation on the Pothwar Plateau in Pakistan is projected to increase by up to 1,528 mm/year) and in tropical oceans but may decrease in subtropical regions ^42,43.

Global climate change is projected to expand suitable areas for cogon grass across both native and introduced regions. Many areas in Africa (e.g., Cameroon, the Democratic Republic of the Congo), South America (e.g., Brazil and Peru), Europe (e.g., Kosovo and Greece), and the southeastern United States, which were previously unsuitable, may become highly favorable. Moreover, parts of its native range, including India, Nepal, Indonesia, and Australia, are also projected to experience increased suitability, consistent with climax-based modeling of cogon grass in a recent study⁴⁴. These shifts highlight the species’ broad climatic tolerance and signal a growing risk to biodiversity and ecosystem stability worldwide, emphasizing the need for proactive management in both native and invaded landscapes.

Cogon grass has remarkable adaptability allowing it to colonize a broad spectrum of land cover types, including grasslands, open forests, pastures, agricultural fields, wetlands, riparian zones, scrubland, roadsides, and urban areas ¹⁰. Our study on land cover-specific projections shows how cogon grass's ecological adaptability and invasion potential is amplified by climate change. The greatest future habitat area increases are in artificial lands and agricultural lands, showing the species’ strong association with human dominated landscapes ¹⁰. These increases indicate a growing risk to urban infrastructure and agricultural productivity, particularly in regions already vulnerable to invasive species. Also concerning is the projected expansion into forested and grassland ecosystems, which are critical reservoirs of native biodiversity ⁸. These changes imply that cogon grass could alter ecological processes and replace native vegetation, thereby destabilizing ecosystem functions.

We found that cogon grass poses significant ecological threats across a wide range of grassland ecosystems, including alpine meadows, temperate steppes, tropical savannas, and semi-desert scrublands. Globally, the species currently occurs in 64 ecoregions, spanning native ranges in Asia, Africa, Australia, and Southern Europe, and non-native regions in the Americas (Table S6). In its native regions, cogon grass is generally well-integrated, while in non-native regions, particularly in North and South America, it exhibits invasive behavior with substantial ecological impacts. In North America, it has invaded at least 10 ecoregions, such as the California Central Valley grasslands, Nebraska Sand Hills mixed grasslands, and Western Gulf coastal grasslands. In South America, invasions are recorded in at least 11 ecoregions, including the Humid Pampas, Guianan savanna, Uruguayan savanna, and Patagonian steppe. Future projections indicate that suitable conditions may expand to as many as 73 ecoregions globally, highlighting the growing risk of range expansion in non-native regions and the potential threat to biodiversity and ecosystem function.

This expansion is particularly concerning in regions such as the Asian alpine and subtropical grasslands, and the savannas of Africa and South America, where cogon grass is likely to alter fire regimes, outcompete native plant species, and degrade habitat quality for native fauna ⁹. These ecological impacts may trigger shifts in species composition, reduce biodiversity, and disrupt ecosystem processes ^7,31.

The HII variable incorporates a range of anthropogenic activities—such as infrastructure development, land use change, and population pressure—that can disrupt natural biogeographic barriers and facilitate the spread of invasive species (Gallardo et al., 2015). Accurately predicting invasive species distributions without incorporating socioeconomic factors is challenging, since these human-driven processes play critical roles in shaping invasion dynamics ⁴⁵. We incorporated the HII to improve predictions of the global distribution of cogon grass. Notably, under the highest-consumption scenario, i.e., SSP5-8.5, we observed the greatest expansion of highly suitable areas, indicating a strong link between high-intensity human activity and increased invasion risk. However, this risk is closely tied to associated shifts in temperature and precipitation patterns under each SSP scenario. Low-emission scenarios, such as SSP1-2.6, exhibit a reduced overall risk, emphasizing the role of sustainable development pathways in mitigating future invasions.

The global management of cogon grass presents significant challenges due to its aggressive growth, adaptability, and regenerative capacity ⁹; and continuous monitoring, stakeholder involvement, and sustained funding are critical for managing invasions effectively ^10,31.

Current control strategies have integrated mechanical, chemical, biological, and ecological approaches, with varying degrees of success depending on region-specific factors. Mechanical control, such as repeated mowing and tilling, is commonly used but often only provides temporary suppression unless followed by other treatments ¹. Unlike in the United States, herbicidal control, particularly using glyphosate or imazapyr, is widely practiced and has shown effectiveness when applied persistently and in combination with burning or mowing ⁴⁶. However, concerns over herbicide resistance and environmental impacts have heightened interest in integrated management approaches, in which prescribed burning can reduce biomass and enhance herbicide efficacy, although it may also stimulate regrowth if not carefully managed ⁴⁶. Prescribed burning can reduce biomass and enhance herbicide efficacy but may also stimulate regrowth if not carefully managed. Biological control remains limited, though research into host-specific fungal pathogens and competitive native species shows promise ⁴⁷. Restoration-based strategies that re-establish competitive native vegetation following control treatments are gaining traction, especially in conservation areas ⁴⁶. Globally, successful long-term management requires adaptive, site-specific approaches that consider ecological, economic, and social dimensions.

This study provides ecoregional insights that can inform targeted management strategies for cogon grass. For instance, in previously cold-limiting ecosystems of northern Asia and Europe, early eradication efforts may prevent establishment. In agricultural landscapes of Africa and South America, sequential treatments combining mechanical removal and herbicide application could reduce spread and limit economic losses. In fire-prone savannas and grasslands of the southeastern United States, prescribed burning integrated with postfire herbicide treatments may help control post-disturbance expansion. By linking management strategies to specific regional contexts, these approaches can be calibrated to local ecological and climatic conditions, improving their effectiveness.

This study mainly relied on bioclimatic and selected environmental variables (e.g., wildfire, land use, soil properties, HII), while excluding key ecological processes such as biotic interactions, competition with native species, and dispersal constraints. These omissions may reduce the accuracy of invasion risk predictions. In addition, our assessment is based on model projections without field validation, which limits confidence in the results. Future research should integrate long-term field data, mechanistic experiments, and remote sensing to validate and refine models. Incorporating ecological interactions and expanding field-based evidence will improve prediction accuracy and strengthen early detection and management strategies for cogon grass.

A total of 29,159 occurrence records for cogon grass of all varieties were obtained from the Global Biodiversity Information Facility, GBIF)¹⁴, an open-access platform that provides species occurrence data compiled from a wide range of nonsystematic sources, including museum specimens and administrative datasets¹⁶. The GBIF plays a key role in promoting the sharing of biodiversity data for research, policy, and decision-making purposes for biodiversity conservation and sustainable management¹⁴. To increase model accuracy and minimize the effects of sampling bias and duplicate coordinates, we applied spatial rarefaction using SDM toolbox v.2.4 ^48,49 in ArcGIS 10.8 (Esri Korea, Seoul). We excluded records dated prior to 1970, data from unreliable sources, records lacking date information, and points located in ecologically irrelevant regions such as oceans, deserts, and polar zones. Expert consultation was also employed to validate the species occurrence records and finalize the species presence points of cogon grass for species distribution modeling. Rarefaction was performed at multiple spatial resolutions (1, 5, 10, 20, and 50 km) to eliminate duplicate records within each grid cell, which yielded six datasets, including one non-rarefied version. This process resulted in a reduced number of records: 13,780 records at 1 km, 5,934 records at 5 km, 3,637 at 10 km, 2,113 records at 20 km, and 1,006 records at 50 km (Fig. 1). These datasets were subsequently used for MaxEnt-based modeling of the cogon grass distribution.

Bioclimatic variables such as temperature and precipitation are important for modeling the global spatial distribution of cogon grass (Table S1). Current climate data, representing the period from 1979 to 2013, were considered the baseline for current climatic conditions in the analysis process and were obtained from PaleoClim v1.2⁵⁰. Similarly, future climate projections were obtained from the WorldClim 2.1 database⁵¹. All the bioclimatic data exhibit a spatial resolution of 2.5 arc-minutes. We employed four shared socioeconomic pathway (SSP) scenarios, namely, SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5, which represent a range of low to high greenhouse gas concentration trajectories. These scenarios were adopted to model potential climatic conditions for the period from 2061 to 2080. SSP describe potential trajectories of global societal development throughout the 21st century, thereby incorporating factors such as population growth, economic progress, technological advancement, and environmental challenges. Each pathway is based on a distinct socioeconomic narrative—ranging from sustainability-oriented development (SSP1) to fossil fuel–driven growth (SSP5) ^52,53.

SSP1-2.6 represents a sustainability-focused pathway characterized by the rapid adoption of renewable energy and reduced dependence on fossil fuels. This scenario supports environmentally sustainable economic growth and is associated with a projected increase in the global temperature of approximately 2°C above preindustrial levels by 2100 ⁴². SSP2-4.5 reflects an intermediate pathway, extending historical trends of uneven economic development and moderate climate policy implementation. Notably, partial achievement of emission reduction targets is assumed, resulting in projected warming of 2.7°C by 2100⁵².

SSP3-7.0 is characterized by slow economic growth, limited technological advancement, continued heavy reliance on fossil fuels, and moderate population growth in developing countries. These conditions increase societal vulnerability to climate change impacts and are projected to yield a 3.6°C temperature increase by 2100^52,54. SSP5-8.5 entails the assumptions of high fossil fuel use, rapid technological innovation, and globalization, alongside energy-intensive economic growth and unregulated urban-centered development. Under this scenario, a temperature increases of 4.4°C is projected by 2100^42,52. These varied SSP scenarios, which represent a spectrum of moderate to extreme warming trajectories, offer robust frameworks for predicting potential changes in the distributions of invasive species. The bioclimatic data for the future scenarios were derived from the MPI-ESM1-2-HR global climate model ⁵⁵ of the Coupled Model Intercomparison Project Phase 6 (CMIP6), which exhibits a consistent spatial resolution of 2.5 arcminutes (~ 4.5 km at the equator) ⁵⁶.

Wildfires are critical components of terrestrial ecosystems and affect global carbon cycles, vegetation dynamics, and atmospheric composition while posing significant threats to biodiversity, human health, and socioeconomic stability ¹¹. Wildfire can enhance the invasion of alien plants via the alteration in disturbance regimes, thereby increasing resources and competitive success over native plants³³. To account for this factor, we downloaded wildfire data recorded between 2000 and 2024 by the Moderate Resolution Imaging Spectroradiometer (MODIS) from the Fire Information for Resource Management System (FIRMS) of the National Aeronautics and Space Administration ⁵⁷ and Global Wildfire Information System ⁵⁸.

Land use and land cover change are important in invasion ecology because they create suitable condition for promoting the introduction, colonization, and spread of invasive species by affecting niche availability and dispersal potential^23,59. The land cover map used has seven categories including forest, agriculture land, grass land, wetland, non-vegetated land, mixed land, and artificial land (Food and Agriculture Organization database ⁶⁰. The artificial land categories include human settlements, and infrastructure development.

Soil characteristics such as organic matter, carbon content, and pH play crucial roles in determining nutrient availability, microbial dynamics, and habitat suitability for invasive species³⁸. Invasive species exhibit unique properties that allow them to modify soil characteristics⁶¹. These altered soils can provide a more favorable environment for the establishment of invasive species ⁶². 2016). Therefore, we incorporated three soil characteristics, i.e., soil pH, soil carbon, and soil moisture levels, into the model to better understand the distributions of cogon grass-related variables ⁶⁰.

The human influence index (HII) encompasses various human activities, such as global trade and tourism, transportation and infrastructure development, urbanization and industrialization, and agriculture, that contribute to the disruption of natural biogeographical boundaries and promote the invasion of alien species⁴⁵. These anthropogenic variables serve as essential ecological indicators for predicting species distributions under changing environmental conditions. Here, we incorporated HII data sourced from the Socioeconomic Data and Applications Center (SEDAC) of National Aeronautics and Space Administration ⁶³.

We employed Pearson’s correlation analysis to reduce multicollinearity and increase model accuracy. We eliminated highly correlated environmental variables (Pearson’s r > 0.75 and p < 0.05). Finally, six bioclimatic variables, i.e., annual mean temperature (Bio1), mean diurnal range (Bio2), isothermality (Bio3), annual precipitation (Bio12), precipitation in the wettest month (Bio13), and precipitation in the driest month (Bio14), as well as five additional environmental variables, i.e., land cover change, wildfires, soil carbon, soil moisture, soil pH, and the HII, were selected for species distribution modeling of cogon grass (Table 1).

In this study, we employed five machine learning algorithms—boosted regression trees (BRTs), a generalized additive model (GAM), the maximum entropy (MaxEnt) model, a random forest (RF) model, and the extreme gradient boosting (XGBoost) model—to predict the global spatial distribution of cogon grass. Machine learning algorithms (MLAs) contribute significantly to species distribution modeling by facilitating the efficient processing of complex ecological datasets, thus increasing the accuracy of species occurrence predictions and promoting the resolution of complex ecological challenges⁶⁴. These algorithms enable the analysis of the complex relationships between environmental variables and species occurrence, leading to a greater understanding of species distributions and supporting conservation and management efforts⁶⁵.

The BRT model effectively resolves nonlinear relationships, identifies important predictor variables, accounts for interactions among predictors, and captures the relative importance of each predictor—tasks that are often challenging for traditional regression-based models⁶⁶. The BRT model is applicable for modeling a wide range of organisms and environmental variables. GAMs constitute a semiparametric approach for predicting nonlinear relationships between environmental variables (e.g., temperature and chlorophyll) and species abundance, enabling reliable spatial predictions within specific environments⁶⁷. The MaxEnt algorithm can be used to integrate species occurrence data with environmental variables, such as climate, land cover changes, and wildfires, to predict potential distributions in new regions on the basis of the principle of maximum entropy ⁶⁸. It relies exclusively on species presence data, avoiding inaccuracies stemming from the use of unreliable absence data due to the unstable and expanding ranges of invasive species ⁶⁹. The MaxEnt algorithm is widely applied in modeling both native and invasive species and is valuable in conservation and management decision-making processes^19,59. The RF model is an effective ensemble classifier with several advantageous features. It accommodates diverse input variable types and generates numerous decision trees, determining classifications through majority voting ⁷⁰. It effectively manages correlated variables and provides robust predictions. The XGBoost model, which leverages its advanced boosting framework, achieves high accuracy in feature selection and offers excellent scalability, making it particularly well suited for large-scale species distribution modeling applications⁷¹. We evaluated the model performance of each algorithm and selected the best algorithm for simulating cogon grass distributions under current and future climate change scenarios.

The data for alien and invasive species are often unreliable due to their expanding distribution ranges and the likelihood that they have not yet reached equilibrium, which can lead to misinterpretation in modeling and analysis⁶⁹. Therefore, global background points (pseudoabsence) were generated using ArcGIS 10.8 (ESRI, Redlands, CA, USA), following methodologies outlined in previous studies^72,73 (Barbet-Massin et al., 2012; Adhikari et al., 2018). BRT, GAM, MaxEnt, RF, and XGBoost modeling was performed by selecting a single model in the Biomod2 Package v.4.2–6.2 for GNU R (https://cran.r-project.org/web/packages/biomod2/index.html) ⁷⁴. For model evaluation, the occurrence dataset was randomly split at a 3:1 ratio for model training and testing and 100 independent replications were performed to ensure the robustness of each species distribution model.

The binary distribution map, generated using the TSS thresholds defined during model calibration⁷⁴, was employed to quantify suitable areas by country under baseline and projected climate change scenarios. The resulting model outputs indicate the global presence or absence of suitable area for cogon grass. ArcGIS Desktop 10.8 (Esri Korea, Seoul) was employed to map the habitat suitability of cogon grass across six continental regions and 195 countries. Similarly, changes from the historic distribution of suitable areas to the 2061–2080 period were analyzed for the four SSPs. The average habitat suitability of cogon grass was calculated for each country for current and projected future climate scenarios and classified into five categories, i.e., unsuitable (0), poorly suitable (0.01–0.25 of the country’s area), moderately suitable (0.25–0.5), highly suitable (0.5–0.75), and extremely suitable (0.75–1.0), across 195 countries worldwide.

Habitat suitability was also estimated across the seven land cover categories. Finally, we quantified the habitat coverage of cogon grass within global grassland areas and identified changes in the extent of invasible habitat in the African, Australian, and South American savannas. Our definition of grassland areas includes wide variety of grassland formation types, including alpine meadows, temperate steppes, tropical savannas, and semi-desert scrublands and was derived from the World Wildlife Fund ⁶⁰.