Machine Learning Meets Ecology: XGBoost‑Based Prediction of Endangered Species Refugia Using Multi‑Source Environmental Data

Machine Learning Meets Ecology: XGBoost‑Based Prediction of Endangered Species Refugia Using Multi‑Source Environmental Data | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Machine Learning Meets Ecology: XGBoost‑Based Prediction of Endangered Species Refugia Using Multi‑Source Environmental Data Aref Hesabi, Seyed Jalil Alavi, Omid Esmailzadeh This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8858849/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 9 You are reading this latest preprint version Abstract Buxus hyrcana , an endangered and ecologically significant tree species of the Hyrcanian forests, faces severe threats from climate change, land-use pressures, and habitat degradation. Accurate prediction of its potential distribution is therefore critical for conservation and restoration planning. In this study, we applied the eXtreme Gradient Boosting (XGBoost) algorithm to model the distribution of B. hyrcana under three data combinations: (i) WorldClim bioclimatic and topographic variables, (ii) CHELSA bioclimatic and topographic variables, and (iii) WorldClim bioclimatic, topographic, and land-use variables. Model performance was evaluated using AUC, TSS, Kappa, and Accuracy metrics, all of which indicated strong predictive capacity, with the highest performance achieved when land-use data were incorporated. Variable importance analysis revealed a stable set of key predictors—thermal stability (Bio3), annual mean temperature (Bio1), mean temperature of the wettest quarter (Bio8), annual precipitation (Bio12), and slope-related indices (LS Factor)—highlighting the species’ sensitivity to moderate climatic regimes and physiographic constraints. Response curve analysis confirmed that B. hyrcana thrives under moderate temperature and precipitation conditions, while extreme climatic values sharply reduce occurrence probability. Habitat suitability maps consistently identified Mazandaran Province as the most suitable region, with additional restoration potential in Golestan Province. Our findings demonstrate that integrating high-resolution climatic, topographic, and land-use data within advanced machine learning frameworks significantly enhances the accuracy and ecological realism of species distribution models. This study provides a robust methodological framework for predicting the distribution of climate-sensitive, endangered species and offers actionable insights for conservation prioritization and restoration planning in the Hyrcanian forests. Buxus hyrcana Hyrcanian Forests Species Distribution Modelling Habitat Suitability Conservation Planning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Introduction The global population is projected to reach 9 billion by 2026, intensifying demands for housing, food, energy, and natural resources that threaten natural vegetation cover unless sustainable development strategies are implemented. Land-use change has affected 32% of the Earth's terrestrial surface during the period 1960–2019 (Winkler et al. 2021 ). Throughout history, deforestation and forest degradation have profoundly transformed the Earth's land surface (Ellis et al. 2021 ). Of the world's natural forest cover, only 40% maintains high ecosystem integrity after extensive clearing for agriculture and other anthropogenic uses (Grantham et al. 2020 ). Forest ecosystem degradation generates far-reaching environmental and societal consequences (Barlow et al. 2016 ), serving as primary drivers of the global biodiversity crisis (Rosa et al. 2016 ) while contributing significantly to carbon emissions, thereby exacerbating climate change (Baccini et al. 2012 ), undermining water provision and hydrological regulation (Zhang and Wei 2021 ), and adversely affecting the livelihoods of forest-dependent communities (Hou et al. 2023 ; Zhang et al. 2017 ). The Hyrcanian forests, extending along the southern shores of the Caspian Sea across Iran and parts of the Republic of Azerbaijan, constitute unique ancient temperate broadleaf ecosystems inscribed as a UNESCO World Heritage Site in 2019. These Tertiary-period forests preserve an evolutionary legacy spanning millions of years (Sagheb Talebi et al. 2014 ) but face persistent threats including deforestation, land-use change, soil erosion (Ahmadi-sani et al. 2022 ), overgrazing, and illegal logging that jeopardize their ecological integrity (Akbarzadeh et al. 2016 ; Shahnaseri et al. 2023 ). Recent studies within UNESCO-registered sites reveal significant declines in regeneration density, with anthropogenic pressures remaining critical threats: livestock grazing occurs in 64% of areas (23% at high intensity), and 15% are affected by illegal logging that disrupts natural recovery processes and threatens ecosystem sustainability (Sohrabi 2025 ). Species distributions result from complex interactions among climatic, topographic, biotic, and anthropogenic factors (Austin 2002 ; Guisan and Thuiller 2005 ), making understanding these drivers fundamental for biodiversity conservation, ecological restoration, and sustainable forest management (Martínez-Meyer et al. 2006 ; Rushton et al. 2004 ). Species Distribution Models (SDMs) have become indispensable tools, enabling prediction of potential ranges, identification of biodiversity hotspots, and prioritization of conservation areas (Moradi et al. 2019 ; Ramírez-Albores et al. 2021 ). By integrating species occurrence records with environmental predictors, SDMs provide spatially explicit probability maps that guide restoration planning, optimize resource allocation, and reduce management failure risks (Dolos et al. 2015 ; Li and Wang 2013 ). Comprehensive ecological knowledge of potential distribution and suitable habitats is crucial for conservation planning of declining species populations (Akrim et al. 2017 ). Since their emergence in the 1980s, SDMs have evolved from regression-based approaches (Guisan et al. 2002 ) to advanced machine learning algorithms capable of capturing complex, non-linear ecological relationships (Jane Elith et al. 2006 ; Olden et al. 2008 ). These techniques, known as niche-based models, habitat models, habitat suitability models, Environmental Niche Models (ENMs), climate envelopes, and Ecological Niche Models (Sillero et al. 2021 , 2023 ), have been widely applied to identify biodiversity hotspots (Moradi et al., 2019 ), assess habitat suitability (Campbell et al. 2024 ), and evaluate climate change impacts on biodiversity (Hotta et al. 2019 ; Liu et al. 2020 ; Sheykhi Ilanloo et al. 2021 ). The transition toward machine learning approaches stems from their flexible fitting capabilities, automated variable selection, and capacity to handle diverse data types (Jane Elith et al. 2006 ). Among machine learning approaches, tree-based algorithms are particularly prominent due to their flexibility in handling non-linearity and ensemble modeling advantages (J. Elith et al. 2008 ; Muñoz-Mas et al. 2019 ). eXtreme Gradient Boosting (XGBoost), an efficient and scalable gradient boosting implementation based on decision trees (Chen and Guestrin 2016 ), incorporates L1/L2 regularization, random sampling, and second-order optimization, enabling high accuracy, robustness against incomplete data, and reduced overfitting (Chen and Guestrin 2016 ). These features make XGBoost particularly powerful for ecological and environmental modeling, especially for endangered species with complex ecological requirements (Fan et al. 2018 ). Endemic and threatened species are especially vulnerable to habitat loss and climate change, as narrow ecological ranges and elevational distributions make them highly sensitive to disturbance (Hu et al. 2018 ; Kidane et al. 2019 ; Wang et al. 2020 ; Zu et al. 2019 ). Anthropogenic climate change exacerbates these vulnerabilities, driving species range shifts and increasing extinction risk (Asefa et al. 2020 ). Buxus hyrcana Pojark, the Hyrcanian boxwood, is a keystone evergreen broadleaf species of northern Iranian forests. Due to its high-quality timber, the species has long been subjected to illegal harvesting, which combined with widespread habitat degradation has driven dramatic declines in its natural range, resulting in IUCN Red List classification. Despite high seed production and regeneration potential, continued anthropogenic pressures and climate change threaten species persistence and habitat ecological integrity (Biru et al. 2017 ; Ssali et al. 2023 ). Official reports indicate boxwood habitats across Gilan, Mazandaran, and Golestan provinces totaled approximately 72,000 hectares in 2013 (Salehi Shanjani et al. 2018 ), with escalating anthropogenic pressures elevating the species to conservation priority status. Although not yet extinct, ecosystems have been severely degraded, though this semi-endemic species (also occurring in the Republic of Azerbaijan) retains high seed production capacity and considerable viability that may support generational continuity if suitable habitats are preserved. Given the ecological significance of Hyrcanian forests and conservation priority of B. hyrcana , precise habitat identification is essential for guiding restoration and management strategies. Advanced SDMs like XGBoost, capable of detecting complex non-linear relationships between species occurrence and environmental variables, provide valuable opportunities to generate accurate habitat suitability maps and identify conservation refugia. This study applies XGBoost to model B. hyrcana potential distribution using multi-source environmental data, including bioclimatic, topographic, and land-use variables. Specifically, the study aims to (i) evaluate model performance across different environmental data combinations, (ii) identify most influential variables shaping species distribution, and (iii) generate high-resolution habitat suitability maps to inform conservation prioritization and restoration planning. By integrating advanced machine learning with ecological knowledge, this research provides a robust framework for long-term B. hyrcana conservation and contributes to broader biodiversity safeguarding efforts within ancient temperate forest ecosystems. Materials and Methods Study Area The Hyrcanian forests constitute a narrow, longitudinal belt along the northern slopes of the Alborz Mountain range, stretching from Astara in the northwest to Golidaghi in Bojnord Province in the northeast. This unique forest ecosystem spans approximately 800 km in length and 110 km in width, encompassing a total area of 1.85 million hectares—equivalent to 15% of Iran’s forested lands and 1.1% of its national territory (Sagheb Talebi et al. 2014) (Fig 1). Altitudinal variation within the Hyrcanian Forests ranges from sea level to elevations exceeding 2,800 meters. Climatic conditions exhibit pronounced spatial gradients: mean annual precipitation varies from 530 mm in the eastern regions to 1,530 mm in the west, with localized maxima reaching up to 2,000 mm near Asalem. Over the past decade, mean annual temperatures have averaged 15°C in the western zones and 17.5°C in the eastern sectors, reflecting the region’s diverse microclimatic regimes. Occurrence Data Collection and Processing A total of 570 georeferenced presence records for Buxus hyrcana were compiled from field surveys across the Hyrcanian forest range in northern Iran (Fig 1). Coordinate uncertainty was addressed by buffering records with a 100-m radius to account for potential geolocation errors, and no records fell within this buffer of others to avoid spatial overlap. Due to the absence of verified absence data, pseudo-absence points (n = 1,140, twice the number of presence records) were generated using random sampling within the study area extent, excluding a 1-km buffer around presence points to minimize bias toward known habitats. This approach followed established practices for balanced datasets in species distribution modeling. Pseudo-absence generation was implemented via the sample_pseudoabs() function from the flexsdm R package (Velazco et al. 2022). To mitigate spatial autocorrelation and sampling bias, spatial thinning was applied using the thinData() function from the SDMtune R package (Vignali et al. 2020). Thinning retained one occurrence per 1-km² raster cell (matching environmental data resolution), resulting in a final dataset of about 200 presence points. All spatial analyses were performed in the WGS 84 coordinate reference system (CRS: EPSG:4326) using the sf (Pebesma and Bivand 2023) and terra (Hijmans R 2025) packages in R programming language (R Core Team, 2025). Environmental Variables Environmental predictors were selected to capture bioclimatic, topographic, and anthropogenic drivers of B. hyrcana distribution, following established frameworks for species distribution models (SDMs; Guisan and Thuiller, 2005). A standardized set of 19 bioclimatic variables (Bio1–Bio19) was derived from two global datasets to evaluate data source impacts: WorldClim version 2.1 (1970–2000 period; Fick and Hijmans, 2017) and CHELSA version 2.1 (1980–2010 period; Karger et al., 2017). Both datasets were downloaded from their respective portals (WorldClim: https://www.worldclim.org; CHELSA: https://chelsa-climate.org) at 1-km resolution (30 arc-seconds) (Table 1). Integrating these variables—either partially or in full—has been shown to significantly improve the predictive performance of species distribution models (SDMs) and yield more nuanced insights into the ecological drivers of species ranges (Amiri et al., 2020; Hesabi et al., 2025). Variables were cropped to the study area extent using a polygon buffer of 50 km around the Hyrcanian forests and masked to land areas, with no gap-filling or interpolation applied. Table 1 Description of variables used for modeling in the study area Category Variable Description Unite Source Use for modeling Climate Bio1 Annual Mean Temperature ◦C Worldclim & Chelsa ü Bio2 Mean Diurnal Range ◦C û Bio3 Isothermality - ü Bio4 Temperature Seasonality (Standard deviation) - ü Bio5 Max Temperature of Warmest Month ◦C û Bio6 Min Temperature of Coldest Month ◦C û Bio7 Temperature Annual Range ◦C û Bio8 Mean Temperature of Wettest Quarter ◦C ü Bio9 Mean Temperature of Driest Quarter ◦C û Bio10 Mean Temperature of Warmest Quarter ◦C û Bio11 Mean Temperature of Coldest Quarter ◦C û Bio12 Annual Precipitation mm ü Bio13 Precipitation of Wettest Month mm û Bio14 Precipitation of Driest Month mm û Bio15 Precipitation Seasonality (Coefficient of variation) - ü Bio16 Precipitation of Wettest Quarter mm û Bio17 Precipitation of Driest Quarter mm û Bio18 Precipitation of Warmest Quarter mm û Bio19 Precipitation of Coldest Quarter mm û Topography Elevation Elevation m SRTM û TRASP Topographic solar radiation aspect index - ü Slope Slope - û TWI Topographic Wetness Index - ü TPI Topography Position Index - ü VD Valley Depth - ü ChND Channel Network Distance - ü CI Convergence Index - ü LS Length and Steepness Factor - ü Topographic Variables For this study, a DEM with a spatial resolution of 1 kilometer was acquired from the Shuttle Radar Topography Mission (SRTM). All primary and secondary topographic variables were computed using SAGA GIS software (version 9.2), ensuring standardized derivation and reproducibility across the study area (Table 1). Among topographic attributes, elevation, slope, and aspect are considered primary variables and are widely employed in ecological modeling due to their strong influence on abiotic conditions and their ease of extraction from remote sensing data. In addition to these primary parameters, a suite of secondary topographic indices—including the Topographic Wetness Index (TWI), Convergence Index (CI), Valley Depth (VD), Channel Network Distance (ChND), Slope Length and Steepness Factor (LS Factor), and Topographic Position Index (TPI)—provides further insights into terrain-driven ecological gradients (Gama et al. 2016). These variables collectively modulate key environmental drivers such as solar radiation, temperature regimes, precipitation patterns, soil moisture availability, and microclimatic conditions, thereby influencing species distributions and community composition (Baker and Barnes 1998). Land Use Layer Preparation The Land Cover Explorer layer from ArcGIS Living Atlas of the World was used as the source of land use/land cover data; this dataset provides annual global maps at ~10 m resolution derived from Sentinel-2 imagery and machine-learning classification (ESRI 2024). The 2024 product was selected to represent the most recent land-use status and to better reflect contemporary potential habitats for Buxus hyrcana. To harmonize spatial resolution with bioclimatic and topographic predictors, the 2024 land-use raster was resampled to 1 km (30 arc-seconds) using modal (majority) resampling, thereby preserving categorical integrity. Resampling and reclassification were performed in R (R Core Team, 2025) using spatial raster tools. The final categorical layer was encoded for model use by converting reclassified classes into binary dummy variables (one-hot encoding) prior to fitting the XGBoost algorithm, ensuring proper representation of categorical effects in the model. Multicollinearity Assessment To ensure model stability and interpretability, multicollinearity among bioclimatic and topographic predictors was evaluated using the Variance Inflation Factor (VIF), a widely adopted diagnostic metric in ecological modeling, econometrics, and machine learning. VIF quantifies the extent to which the variance of a regression coefficient is inflated due to linear dependence with other predictors, thereby serving as a proxy for redundancy and collinearity (Kutner et al. 2005). Each predictor was assessed individually, and VIF values were computed to identify overlapping information among variables. High VIF scores indicate strong collinearity, which can lead to unstable parameter estimates, reduced model generalizability, and misleading ecological interpretations (Hastie et al. 2021). By eliminating or consolidating collinear variables, VIF screening contributes to the development of parsimonious and robust models. The analysis was conducted using the vifstep function from the usdm package in R (Naimi et al., 2014; R Core Team, 2025), which iteratively removes the variable with the highest VIF until all remaining predictors fall below a predefined threshold. In this study, a conservative cutoff of VIF < 10 was applied, consistent with best practices in species distribution modeling. The final set of retained variables is presented in Table 1. Modeling Process Initial modeling efforts employed the Boosted Regression Trees (BRT) algorithm, a well-established ensemble method based on decision tree learning and iterative error reduction. BRT has been widely applied in ecological studies, particularly for species distribution modeling (J. Elith et al. 2008). However, recent advancements in machine learning have led to the development of more efficient and scalable algorithms, notably XGBoost (Extreme Gradient Boosting), which offers enhanced performance in terms of predictive accuracy, computational speed, handling of missing data, and mitigation of overfitting (Chen and Guestrin 2016; Wang et al. 2020). Following a comparative review of current literature and expert recommendations, XGBoost was selected as the preferred modeling approach for this study. Its adoption not only improved model precision but also contributed to greater stability and interpretability of the final predictions. Modeling was conducted using the tidymodels framework in R (Kuhn and Wickham 2020). The XGBoost algorithm requires both presence and absence data. Since verified absence records were unavailable, pseudo-absence points were generated using the the sample_pseudoabs() function from the flexsdm package (Velazco et al. 2022). Environmental predictor values—including bioclimatic and physiographic variables—were extracted at each presence and pseudo-absence location using the sdm_extract() function (Barbet-Massin et al. 2012). The final modeling matrix included the following variables: Bio1, Bio3, Bio4, Bio8, Bio12, Bio15, slope, Topographic Wetness Index (TWI), Valley Depth (VD), Solar Radiation Index (SRI), Convergence Index (CI), LS Factor, and Channel Network Distance (ChND). Data were partitioned into training (70%) and testing (30%) subsets. To prevent overfitting and ensure model generalizability, 10-fold cross-validation was applied. Hyperparameter tuning was performed using a grid search across combinations of number of trees (500–2500), tree depth (3, 5, 7), learning rate (0.3, 0.1, 0.05, 0.01, 0.005), number of predictors sampled per split (mtry = 2–13), and minimum samples per node (10–50). The optimal configuration—mtry = 12, trees = 1500, min_n = 10, tree_depth = 7, learn_rate = 0.005—yielded an AUC of 0.96, indicating excellent model performance. Model evaluation was conducted using multiple metrics: Area Under the Curve (AUC) from the Receiver Operating Characteristic (ROC), True Skill Statistic (TSS), sensitivity, specificity (1-specificity), overall accuracy, and Cohen’s Kappa coefficient. To assess the influence of different bioclimatic datasets, three modeling combinations were tested: Combination 1 : WorldClim bioclimatic variables + topographic predictors + presence/pseudo-absence data Combination 2 : Chelsa bioclimatic variables + topographic predictors + presence/pseudo-absence data Combination 3 : WorldClim bioclimatic variables + topographic predictors + land use layer + presence/pseudo-absence data These combinations enabled comparative evaluation of dataset contributions to model performance and ecological inference. Results Model Performance and Discriminatory Capacity The predictive performance of the XGBoost model for Buxus hyrcana distribution was assessed using widely accepted evaluation metrics, including the Area Under the Receiver Operating Characteristic Curve (AUC), True Skill Statistic (TSS), overall Accuracy, and the Kappa coefficient. The XGBoost model demonstrated strong predictive performance across all three data combinations, with performance metrics revealing distinct trade-offs between environmental data sources and model complexity. Detailed evaluation metrics are presented in Table 2. Table 2 Model performance evaluation metrics for Buxus hyrcana distribution modeling XGboost model (Combination 3) XGboost model (Combination 2) XGboost model (Combination 1) Evaluation criteria 0.98 0.97 0.98 Area Under the Curve 0.86 0.45 0.67 True Skill Statistic 0.98 0.93 0.95 Accuracy 0.91 0.55 0.73 Kappa Combination 1 exhibited strong discriminatory power between presence and absence records (AUC = 0.98, TSS = 0.67, Accuracy = 0.95), indicating reliable capacity to distinguish suitable from unsuitable habitats (Fielding and Bell 1997). The Kappa coefficient of 0.73 demonstrates substantial agreement between predicted and observed distributions, substantially exceeding the minimum threshold of 0.60 typically required in species distribution modeling (Fielding and Bell 1997; Franklin 2013). Combination 2 showed a marked decline in TSS (0.45) relative to Combination 1, representing a 0.22-point reduction in this threshold-sensitive metric, despite maintaining comparable AUC values (0.97). This pattern indicates that Chelsa-derived bioclimatic variables, while capturing broad-scale climatic variation, may be less discriminative for fine-scale presence-absence distinctions in Buxus hyrcana . The Kappa coefficient of 0.55 falls within the "moderate agreement" range, suggesting that this data combination, while acceptable for broad biogeographic patterns, introduces classification uncertainty not evident in Combination 1. Combination 3 achieved optimal performance across all metrics (TSS = 0.86, Accuracy = 0.98, Kappa = 0.91), substantially exceeding both previous combinations. The 0.19-point improvement in TSS relative to Combination 1 and 0.41-point improvement relative to Combination 2 demonstrates that the integration of land-use data captured critical environmental constraints on Buxus hyrcana distribution not fully represented by climate and topography alone. This improvement is particularly significant given that TSS is sensitive to both commission and omission errors and provides robust evaluation independent of prevalence. Environmental Variable Importance and Hierarchical Drivers Variable importance analysis, based on the Gain criterion (representing average improvement in log-loss at split nodes), identified distinct hierarchies of environmental drivers across data combinations. This approach quantifies the relative contribution of each predictor to reducing model uncertainty, revealing which environmental gradients most strongly constrain Buxus hyrcana distribution. Combination 1: WorldClim Bioclimatic and Topographic Variables The importance ranking in Combination 1 identified isothermality (Bio3), Length-Slope Factor (LS Factor), precipitation seasonality (Bio15), and annual precipitation (Bio12) as the four dominant drivers (Fig 2). These variables collectively explain the primary environmental niche dimensions for Buxus hyrcana . Combination 2: Chelsa Bioclimatic and Topographic Variables In Combination 2, the importance hierarchy shifted, with mean temperature of warmest quarter (Bio8), temperature seasonality (Bio4), LS Factor, and Channel Network Distance ranked as top predictors (Fig 3). Notably, the dominance of Bio8 and Bio4 in this combination versus Bio3 in Combination 1 reflects database-specific differences in temperature representation between WorldClim and Chelsa sources. The persistence of LS Factor across both combinations underscores the consistent importance of topographic heterogeneity. The emergence of Channel Network Distance suggests that Chelsa enhanced spatial resolution may better capture fine-scale hydrological connectivity. Combination 3: Integrated Environmental Layers Combination 3 revealed a fundamentally reorganized importance hierarchy, with land use, isothermality (Bio3), annual mean temperature (Bio1), and annual precipitation (Bio12) as the dominant predictors (Fig 4). This reorganization carries substantial ecological significance. Land Use emerged as the single most influential variable, accounting for the substantial performance improvement observed in evaluation metrics, indicating that current or recent human land-use practices represent the primary contemporary constraint on Buxus hyrcana habitat suitability. Bio3, Bio1, and Bio12 retained importance similar to Combination 1, but with shifted relative rankings, suggesting that land-use patterns correlate with and partially subsume climate-topography effects. This pattern indicates that Buxus hyrcana distribution in the Hyrcanian forests is jointly determined by underlying bioclimatic-topographic niches and anthropogenic landscape modification. Species-Environment Response Functions Response curves derived from XGBoost predictions illustrate the functional relationships between Buxus hyrcana occurrence probability and leading environmental predictors across gradient space. These non-parametric response curves reveal the realized ecological niche dimensions and identify critical environmental thresholds. The response curves for Combination 1 (Fig 5) reveal Bio3 (Isothermality) showed a marked threshold response, with maximum occurrence probability concentrated between 27-32% isothermality values. Probabilities decline steeply above 32%, indicating that environments with highly variable seasonal temperature cycles exceed physiological tolerance limits of Buxus hyrcana . This threshold response suggests that stabilized thermal regimes, characteristic of humid continental and subtropical climates, represent fundamental niche requirements. LS Factor, with a unimodal response curve and optimal suitability at intermediate LS values (approximately 2.5-4.5), indicates that boxwood establishes on moderately sloping terrain with moderate water flow dynamics. Both very low LS values (flat terrain with poor drainage) and very high values (steep erosion-prone slopes) show reduced suitability, reflecting trade-offs between waterlogging and erosion stress. Bio15 (Precipitation Seasonality) showed an inverse relationship between precipitation seasonality and occurrence probability, with maximum suitability at low seasonality values (<60 coefficient of variation). This pattern indicates that Buxus hyrcana is restricted to regions with relatively equitable year-round precipitation distribution, consistent with its distribution in humid evergreen forests. Bio12 (Annual Precipitation) exhibits a distinct threshold response, with Buxus hyrcana occurrence probability rising sharply once annual precipitation exceeds approximately 1000 mm. Suitability continues to increase until reaching a plateau around 2500 mm, beyond which additional rainfall does not confer further ecological advantage. This pattern underscores the species’ reliance on consistently moist environments, where water availability supports physiological processes and forest canopy stability. In Combination 2, which explores the Chelsa climate–topography niche space, the response curves (Fig 6) reaffirm the central role of temperature and seasonality in shaping Buxus hyrcana’s ecological preferences, while also highlighting patterns unique to the Chelsa dataset. The mean temperature of the warmest quarter reveals a steadily declining occurrence probability across the temperature gradient, with the highest suitability observed at temperatures below 22°C. This suggests that although the species is present in subtropical regions, it tends to occupy cooler microhabitats or higher elevations, aligning with its known distribution in the montane Hyrcanian forests. Temperature seasonality exhibits a unimodal response, with optimal suitability occurring within a standard deviation range of 700 to 900°C. This indicates a preference for climates that maintain moderate seasonal temperature fluctuations, avoiding both highly stable and highly variable thermal regimes. The LS factor and channel network distance continue to demonstrate monotonic or complex relationships similar to those observed in Combination 1, reinforcing the consistent influence of hydrological and topographic variables across different environmental data sources. In Combination 3, which integrates land-use, climate, and topographic variables (Fig 7), the inclusion of land-use data reveals critical interactions that further constrain the realized niche of Buxus hyrcana . The highest occurrence probabilities are associated with land-use categories corresponding to intact or minimally disturbed forests. In contrast, areas characterized by agricultural activity, urban development, or significant anthropogenic disturbance exhibit markedly reduced suitability. This pattern underscores the dominant role of historical and current land-use practices in shaping the contemporary distribution of the species, effectively filtering the broader climate–topography niche. The responses of annual precipitation, isothermality, and mean annual temperature in this combination mirror those observed in Combination 1, but with subtle shifts that reflect their interaction with land-use patterns. These modifications suggest a nested constraint structure, where only landscapes that simultaneously meet the climatic and topographic requirements and align with favorable land-use conditions can support viable populations of Buxus hyrcana . This highlights the compounded impact of environmental suitability and human land-use history in determining the species’ distribution. Spatial Distribution Patterns and Habitat Suitability Mapping Habitat suitability maps generated by XGBoost provide spatially explicit predictions of potential distribution zones for Buxus hyrcana across the Hyrcanian forest region (Fig 8-10). These maps translate multivariate species-environment relationships into actionable spatial predictions for conservation planning and biodiversity assessment. The map derived from Combination 1 (Fig 8) delineates suitable habitats primarily in regions characterized by moderate temperature and precipitation regimes, with topographic heterogeneity further refining habitat suitability. In Combination 2 (Fig 9), the predicted distribution pattern is broadly consistent with Combination 1, though with slightly reduced spatial precision, reflecting the lower overall model performance observed in this combination. The most accurate and ecologically realistic predictions were obtained from Combination 3 (Fig 10), where the integration of land-use data enhanced the model’s ability to delineate suitable habitats. Collectively, these maps emphasize the critical role of climatic and topographic variables in shaping the distribution of Buxus hyrcana , while also demonstrating that the careful integration of additional environmental layers can refine habitat suitability predictions. The coordinated analysis across three data combinations demonstrates that Buxus hyrcana distribution in the Hyrcanian forests results from hierarchically nested environmental constraints. At the broadest scale, bioclimatic variables—particularly temperature consistency (isothermality) and moisture availability (annual precipitation)—define a fundamental niche envelope. Topographic variables refine this envelope by capturing local water availability and erosion dynamics. Most restrictively, contemporary land use defines the realized distribution by determining which climatically and topographically suitable areas remain available within intact or semi-intact forest ecosystems. This nested constraint structure has direct implications for species conservation and habitat restoration: expanding Buxus hyrcana distribution will require not only climatic suitability and topographic compatibility, but also land-use policy changes that permit forest regeneration in currently degraded areas. Discussion Model Performance Analysis The XGBoost model demonstrated consistently strong predictive performance across all tested combinations. Combination 1 achieved robust metrics (Accuracy = 0.95, Kappa = 0.73, AUC = 0.98, TSS = 0.67), while Combination 2 showed moderate declines (Accuracy = 0.93, Kappa = 0.55, AUC = 0.97, TSS = 0.45) likely due to Chelsa's interpolation challenges in mountainous terrains. The most notable improvement occurred in Combination 3, where land-use integration substantially enhanced performance (Accuracy = 0.98, Kappa = 0.91, AUC = 0.98, TSS = 0.86), highlighting the importance of incorporating anthropogenic variables alongside climatic and topographic predictors. When compared with previous studies, our XGBoost models consistently outperformed established benchmarks. (Jane Elith et al. 2006 ) reported mean AUC values ranging from 0.75 to 0.95 for MaxEnt across diverse ecosystems, while our models consistently achieved AUC values above 0.97. Similarly, although TSS values in Combinations 1 and 2 were lower than those reported by (Mohan et al. 2024 ) for Quercus oblongata (TSS = 0.86), the integration of physiographic and land-use variables in Combination 3 elevated TSS to 0.86, underscoring the synergistic effect of multi-dimensional predictors. (Asadi et al. 2025 ) compared multiple algorithms for Quercus castaneifolia habitats, with Random Forest achieving the highest AUC (0.77), followed by MaxEnt (0.75), SVM (0.72), KNN (0.71), and GLM (0.70). In contrast, our XGBoost models consistently outperformed these approaches across varying data combinations, confirming the algorithm's superior capacity for handling complex, non-linear ecological relationships. Analysis of Relative Variable Importance In Combination 1 (WorldClim + topography), variables such as Bio3 (isothermality), LS Factor (slope length and steepness), Bio12 (annual precipitation), and Bio15 (precipitation seasonality) ranked highest in importance, emphasizing thermal stability and topographic constraints as foundational drivers of B. hyrcana distribution. Combination 2 (Chelsa + topography) introduced Bio8 (mean temperature of the wettest quarter) and ChND (channel network distance) among top predictors, illustrating the algorithm's sensitivity to nuanced climatic data structures and their interactions in topographically complex regions. This shift highlights Chelsa's superior representation of local microclimates in mountainous areas, as validated by (Hesabi et al. 2025 ), leading to greater emphasis on temperature-humidity dynamics over pure isothermality. A striking pattern across combinations was the stability of core variables—LS Factor, Bio12, and Bio3—consistently ranking in the top five, underscoring their fundamental role in defining boxwood habitats influenced by slope stability, annual moisture, and thermal uniformity. Bio8 and ChND gained prominence in Combinations 2 and 3, reinforcing the species' reliance on moderate wet-season temperatures and proximity to water sources. However, in Combination 3, the land-use variable exerted the greatest influence, ranking first and comprising over 20% of total importance via the Gain index. This dominance stems from the near-exclusive association of presence records with forested land-use classes in the Sentinel-2 dataset (ESRI 2024 ), coupled with the species' avoidance of urban, agricultural, or degraded areas. The integration of land-use data in Combination 3 markedly elevated predictive performance, underscoring the pivotal role of anthropogenic factors in shaping B. hyrcana distribution. Variable importance analysis positioned land use as the top predictor, surpassing even core climatic variables like Bio3 and Bio12, likely because the majority of occurrence records were associated with forested land-use categories. This dominance reflects the species' strict dependence on intact, undisturbed forest habitats, where shade-tolerant, evergreen boxwood thrives in understory conditions but is highly vulnerable to deforestation. In the Hyrcanian context, where historical land-use changes have reduced forest cover by up to 40% since the mid-20th century, the land-use layer effectively captured human-induced constraints, refining model outputs to exclude non-forested areas and highlighting how anthropogenic pressures amplify climate sensitivities. The prominence of land use has profound implications for conservation and restoration strategies in the Hyrcanian forests. The refined habitat suitability maps from Combination 3 precisely delineated high-potential zones in Mazandaran Province while identifying restoration opportunities in Golestan, where suitable climatic and topographic conditions exist but are undermined by current agricultural dominance. This suggests that without addressing land-use barriers, projected suitable habitats may remain inaccessible, exacerbating B. hyrcana's decline, which has already seen population losses of over 50% in some areas due to illegal harvesting and fragmentation (Sagheb Talebi et al. 2014 ). Comparing results with (Habibi Kilak et al. 2025 ) on Taxus baccata shows both studies emphasize the essential role of bioclimatic and topographic variables in determining shade-loving tree species distribution. Habibi Kilak et al. identified Bio2, Bio3, Bio7, and Bio8 as key climatic variables, with altitude and slope as important topographic factors. These findings share significant similarities with our XGBoost model results in Combinations 1 and 2, especially emphasizing Bio12, Bio3, and LS Factor. However, differences exist; for example, in Habibi Kilak's study, Bio2 and Bio7 played greater roles, while in the present study, these were not primary. This may arise from physiological differences: yew is more sensitive to daily temperature fluctuations, whereas Hyrcanian boxwood responds more to annual temperature uniformity. (Z. A. Wani et al. 2023 ) modeled Buxus wallichiana habitats in the Himalayas using MaxEnt, identifying only 0.4% suitable area, with annual mean temperature, driest-month precipitation, and elevation as top variables. Similarities with our Combination 1 include isothermality (Bio3) as critical for Buxus viability, but differences emerge: our model prioritized annual precipitation (Bio12) over driest-month metrics, and LS Factor over elevation. These reflect ecological distinctions— B. hyrcana in humid Caspian forests versus B. wallichiana in drier Himalayas—highlighting the need for species- and region-specific modeling. Response Curve Analysis In Combination 1, variables Bio3, LS Factor, Bio15, and Bio12 showed the greatest impact. The Bio3 response curve trended upward to ~ 45 units, stabilizing at high presence probability, indicating positive thermal adaptation (Becklin et al. 2016 ; Z. Wani et al. 2024 ). LS Factor decreased sharply with steeper slopes, peaking again at ~ 30, confirming sensitivity to erosion-prone terrains. Bio15 favored 40–60 mm seasonal rainfall, remaining stable above 60 mm but dropping below, showing aversion to precipitation deficits. Bio12 peaked below 500 mm annual rainfall, declining with excess, aligning with west-east moisture gradients in Hyrcanian forests (Moghbel Esfahani et al. 2023 ). Overall, B. hyrcana prefers high isothermality, gentle slopes, moderate temperatures, and balanced precipitation. In Combination 2, Bio4, Bio8, ChND, and LS Factor dominated. Bio4 rose rapidly from 700–800 units, favoring seasonal temperature range stability. Bio8 peaked at 0–8°C in wet seasons, dropping above 10°C, matching evergreen optima (Moore et al. 2021 ). ChND showed complex patterns, increasing 200–300 m from streams for moisture without flooding. LS Factor again favored low values. Results indicate suitability in high seasonal range, moderate wet-season temperatures, gentle slopes, and water access. In Combination 3, Bio12, Bio1, Bio3, and land use were key. Bio12 declined above 450–500 mm, preferring drier conditions. Bio3 rose from ~ 29 units. Bio1 stabilized at 0–14°C, dropping above 15°C, signaling thermal limits and warming vulnerability. Across combinations, patterns converged on moderate climates and topography, with data variations altering thresholds, emphasizing careful variable selection for conservation. Habitat Suitability Map Analysis The habitat suitability maps across the three data combinations revealed consistent spatial patterns, with Mazandaran Province emerging as the primary stronghold for Buxus hyrcana . Combination 1 (WorldClim + topography) concentrated high-density suitable areas in mid-mountain forests of Mazandaran, particularly from Kelardasht to Neka, while Gilan exhibited fragmented suitability restricted to specific elevations. Combination 2 produced broadly similar results, again highlighting central Mazandaran as the core distribution zone. In Combination 3, the model delineated highly suitable areas with greater precision, closely matching the actual distribution of Hyrcanian forests. Key hotspots included forests south of Chalus, Nur, Sari, and Behshahr. Importantly, the inclusion of land-use data enhanced spatial realism by excluding unsuitable areas outside forested zones. Mazandaran Province, particularly its central and eastern sections, consistently exhibited the highest habitat suitability, aligning with current stand density and natural distribution patterns (Alipour and Walas 2023 ; Habibi kilak et al. 2019 ). The ecological conditions of this region—high humidity, maritime climate, moderate temperatures, and thermal stability—are fully compatible with the ecological requirements of B. hyrcana as a shade-tolerant, moisture-dependent species (Sagheb Talebi et al. 2014 ). In Gilan Province, suitable habitats were also identified, particularly in western low-elevation humid areas, confirming the model's ability to capture species-environment relationships. A particularly noteworthy finding was the prediction of moderate to high suitability areas in Golestan Province, where no natural stands of boxwood have been reported. This suggests significant restoration potential in eastern Hyrcanian forests, provided that biological, ecological, and management conditions are favorable. Such results are consistent with previous studies advocating the identification of potential habitats for restoration in regions currently lacking populations (Becklin et al. 2016 ; Esmailzadeh and Soleymanipour 2015 ). Taken together, the comparative evaluation indicates that conservation priorities should focus on Mazandaran and western Gilan, while Golestan should be considered a priority region for future restoration initiatives. Conclusions This study demonstrated that the XGBoost algorithm is a highly effective and accurate tool for predicting the distribution of Buxus hyrcana in the Hyrcanian forests of northern Iran. Across three different data combinations, the model consistently achieved strong performance, with high AUC (> 0.97), Kappa, and TSS values confirming its ability to discriminate between suitable and unsuitable habitats. Notably, Combination 3, which integrated WorldClim bioclimatic variables, topographic predictors, and land-use data, provided the highest accuracy and ecological realism (TSS = 0.86), underscoring the value of combining multiple environmental datasets—especially anthropogenic factors—in species distribution modeling to capture real-world threats. The analysis of variable importance revealed a stable set of key predictors—Bio3 (isothermality), Bio15 (precipitation seasonality), Bio8 (mean temperature of the wettest quarter), Bio12 (annual precipitation), and LS Factor—highlighting the central role of thermal stability, moderate temperatures, precipitation balance, and topographic constraints in shaping boxwood distribution, with land use emerging as the overriding influence in human-impacted scenarios. The persistence of these variables across combinations indicates that B. hyrcana is highly sensitive to deviations from moderate climatic conditions and physiographic stability, amplified by land-use pressures. Response curve analysis clarified the species' ecological niche, showing that B. hyrcana thrives under moderate temperature and precipitation regimes, while extreme conditions sharply reduce occurrence probability. The dual role of LS Factor in Combination 3 suggested that the species can persist in both valley bottoms and certain stable steep slopes, provided other environmental conditions and intact land use are favorable. Habitat suitability maps confirmed Mazandaran Province as the most suitable and continuous habitat, followed by more fragmented areas in Gilan. The identification of potential habitats in Golestan Province, despite the absence of current populations, highlights opportunities for restoration and reintroduction programs in eastern Hyrcanian forests, contingent on land-use restoration. Overall, this research demonstrates that advanced machine learning approaches such as XGBoost, when combined with comprehensive climatic, topographic, and land-use data, provide powerful tools for identifying and prioritizing suitable habitats for endangered species. The findings not only enhance ecological understanding of B. hyrcana but also provide a robust scientific foundation for conservation planning, sustainable management, and restoration strategies in the Hyrcanian forests. Integrating accurate environmental data with cutting-edge modeling techniques is therefore essential for safeguarding endemic species and promoting biodiversity conservation in these globally significant ecosystems. Declarations Conflicts of interest/Competing interests the authors have no competing interests to declare that are relevant to the content of this article. Open Access Not applicable. Ethics Approval All authors have given their full consent regarding ethics approval. Consent to Participate All authors have given their full consent regarding participation. Consent for Publication All authors have given their full consent for publication. Funding No funding was received to assist with the preparation of this manuscript. Data availability Not accessible. Code availability the current study code is available upon reasonable request from the corresponding author. Author contributions All authors contributed to the study conception and design. O.E conducted field sampling. A.H performed preliminary analysis and wrote the first draft of the manuscript in Persian and O.E assisted in editing the draft. S.J.A Provided the initial idea for the study, conducted the data analysis, translated the text into English, and polished the manuscript. All authors read and approved the final manuscript. References Ahmadi‐sani N, Razaghnia L, Pukkala T (2022) Effect of Land‐Use Change on Runoff in Hyrcania. Land (Basel) 11:. https://doi.org/10.3390/land11020220 Akbarzadeh A, Ghorbani-Dashtaki S, Naderi-Khorasgani M, et al (2016) Monitoring and assessment of soil erosion at micro-scale and macro-scale in forests affected by fire damage in northern Iran. Environ Monit Assess 188:. https://doi.org/10.1007/s10661-016-5712-6 Akrim F, Mahmood T, Hussain R, et al (2017) Distribution pattern, population estimation and threats to the Indian Pangolin Manis crassicaudata (Mammalia: Pholidota: Manidae) in and around Pir Lasura National Park, Azad Jammu & Kashmir, Pakistan. J Threat Taxa 9:. https://doi.org/10.11609/jott.2914.9.3.9920-9927 Alipour S, Walas Ł (2023) The influence of climate and population density on Buxus hyrcana potential distribution and habitat connectivity. J Plant Res 136:. https://doi.org/10.1007/s10265-023-01457-5 Amiri M, Tarkesh M, Jafari R, Jetschke G (2020) Bioclimatic variables from precipitation and temperature records vs. remote sensing-based bioclimatic variables: Which side can perform better in species distribution modeling? Ecol Inform 57:. https://doi.org/10.1016/j.ecoinf.2020.101060 Asadi H, Jalilvand H, Tafazoli M, Hosseini SF (2025) Modeling habitat suitability of Quercus castaneifolia in the Hyrcanian forest: a comprehensive integration of environmental factors for conservation insights. Biodivers Conserv 34:315–334. https://doi.org/10.1007/s10531-024-02973-z Asefa M, Cao M, He Y, et al (2020) Ethiopian vegetation types, climate and topography. Plant Divers 42:. https://doi.org/10.1016/j.pld.2020.04.004 Austin MP (2002) Spatial prediction of species distribution: An interface between ecological theory and statistical modelling. Ecol Modell 157:. https://doi.org/10.1016/S0304-3800(02)00205-3 Baccini A, Goetz SJ, Walker WS, et al (2012) Estimated carbon dioxide emissions from tropical deforestation improved by carbon-density maps. Nat Clim Chang 2:. https://doi.org/10.1038/nclimate1354 Baker ME, Barnes B V. (1998) Landscape ecosystem diversity of river floodplains in northwestern Lower Michigan, U.S.A. Canadian Journal of Forest Research 28:. https://doi.org/10.1139/x98-107 Barbet-Massin M, Jiguet F, Albert CH, Thuiller W (2012) Selecting pseudo-absences for species distribution models: How, where and how many? Methods Ecol Evol 3:. https://doi.org/10.1111/j.2041-210X.2011.00172.x Barlow J, Lennox GD, Ferreira J, et al (2016) Anthropogenic disturbance in tropical forests can double biodiversity loss from deforestation. Nature 535:. https://doi.org/10.1038/nature18326 Becklin KM, Anderson JT, Gerhart LM, et al (2016) Examining plant physiological responses to climate change through an evolutionary lens. Plant Physiol 172:. https://doi.org/10.1104/pp.16.00793 Biru Y, Tessema ZK, Urge M (2017) Perception and attitude of pastoralists on livestock-wildlife interactions around Awash National Park, Ethiopia: implication for biodiversity conservation. Ecol Process 6:. https://doi.org/10.1186/s13717-017-0081-9 Campbell CE, Jones DN, Awasthy M, et al (2024) Which birds have the most to lose? an analysis of bird species’ feeding habitat in changing Australian landscapes. Biodivers Conserv 33:2867–2883. https://doi.org/10.1007/s10531-024-02890-1 Chen T, Guestrin C (2016) XGBoost: A scalable tree boosting system. In: Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining Dolos K, Bauer A, Albrecht S (2015) Site suitability for tree species: Is there a positive relation between a tree species’ occurrence and its growth? Eur J For Res 134:. https://doi.org/10.1007/s10342-015-0876-0 Elith J, H. Graham C, P. Anderson R, et al (2006) Novel methods improve prediction of species’ distributions from occurrence data. Ecography 29:. https://doi.org/10.1111/j.2006.0906-7590.04596.x Elith J, Leathwick JR, Hastie T (2008) A working guide to boosted regression trees. Journal of Animal Ecology 77 Ellis EC, Gauthier N, Goldewijk KK, et al (2021) People have shaped most of terrestrial nature for at least 12,000 years. Proc Natl Acad Sci U S A 118:. https://doi.org/10.1073/pnas.2023483118 Esmailzadeh O, Soleymanipour S (2015) Habitat suitability and ecological requirements of Buxus hyrcana in the Hyrcanian forests. Iranian Journal of Forest and Poplar Research 23:45–58 ESRI (2024) Sentinel-2 10 m Land Use/Land Cover Time Series (2017–2024) [Imagery layer]. In: ArcGIS Living Atlas of the World Fan J, Yue W, Wu L, et al (2018) Evaluation of SVM, ELM and four tree-based ensemble models for predicting daily reference evapotranspiration using limited meteorological data in different climates of China. Agric For Meteorol 263:. https://doi.org/10.1016/j.agrformet.2018.08.019 Fick SE, Hijmans RJ (2017) WorldClim 2: new 1-km spatial resolution climate surfaces for global land areas. International Journal of Climatology 37:. https://doi.org/10.1002/joc.5086 Fielding AH, Bell JF (1997) A review of methods for the assessment of prediction errors in conservation presence/absence models. Environ Conserv 24:38–49. https://doi.org/10.1017/S0376892997000088 Franklin J (2013) Species distribution models in conservation biogeography: Developments and challenges. Divers Distrib 19 Gama M, Crespo D, Dolbeth M, Anastácio P (2016) Predicting global habitat suitability for Corbicula fluminea using species distribution models: The importance of different environmental datasets. Ecol Modell 319:. https://doi.org/10.1016/j.ecolmodel.2015.06.001 Grantham HS, Duncan A, Evans TD, et al (2020) Anthropogenic modification of forests means only 40% of remaining forests have high ecosystem integrity. Nat Commun 11:. https://doi.org/10.1038/s41467-020-19493-3 Guisan A, Edwards TC, Hastie T (2002) Generalized linear and generalized additive models in studies of species distributions: Setting the scene. Ecol Modell 157:. https://doi.org/10.1016/S0304-3800(02)00204-1 Guisan A, Thuiller W (2005) Predicting species distribution: Offering more than simple habitat models. Ecol Lett 8 Habibi kilak S, Alavi SJ, Esmailzadeh O (2019) Analyzing the ecological niche of Buxus hyrcana Pojark in the northern forests of Iran. Forest and Wood Products 72:21–31 Habibi Kilak S, Alavi SJ, Esmailzadeh O (2025) Spatial resolution matters: unveiling the role of environmental predictors in English yew (Taxus bacata L.) distribution using MaxEnt modeling. Earth Sci Inform 18:214. https://doi.org/10.1007/s12145-024-01646-3 Hastie T, Tibshirani R, James G, Witten D (2021) An introduction to Statistical Learning with Applications in R (2nd Edition). Springer texts 102: Hesabi A, Alavi SJ, Esmailzadeh O (2025) Evaluation of the Accuracy of Climatic Data from the Worldclim and Chelsa Databases in Three Northern Provinces of Iran. Forest Research and Development. https://doi.org/10.30466/jfrd.2025.55648.1743 Hijmans R (2025) terra: Spatial Data Analysis Hotta M, Tsuyama I, Nakao K, et al (2019) Modeling future wildlife habitat suitability: Serious climate change impacts on the potential distribution of the Rock Ptarmigan Lagopus muta japonica in Japan’s northern Alps. BMC Ecol 19:. https://doi.org/10.1186/s12898-019-0238-8 Hou Y, Wei X, Zhang M, et al (2023) A global synthesis of hydrological sensitivities to deforestation and forestation. For Ecol Manage 529 Hu Y, Ding Z, Jiang Z, et al (2018) Birds in the Himalayas: What drives beta diversity patterns along an elevational gradient? Ecol Evol 8:. https://doi.org/10.1002/ece3.4622 Karger DN, Conrad O, Böhner J, et al (2017) Climatologies at high resolution for the earth’s land surface areas. Sci Data 4:. https://doi.org/10.1038/sdata.2017.122 Kidane YO, Steinbauer MJ, Beierkuhnlein C (2019) Dead end for endemic plant species? A biodiversity hotspot under pressure. Glob Ecol Conserv 19:. https://doi.org/10.1016/j.gecco.2019.e00670 Kuhn M, Wickham H (2020) Tidymodels: a collection of packages for modeling and machine learning using tidyverse principles Kutner MH, Nachtsheim CJ, Neter J, Li W (2005) Chapter 11: Building the Regression Model III: Remedial Measures. In: Applied Linear Statistical Models Li X, Wang Y (2013) Applying various algorithms for species distribution modelling. Integr Zool 8 Liu L, Liao J, Wu Y, Zhang Y (2020) Breeding range shift of the red-crowned crane (Grus japonensis) under climate change. PLoS One 15:. https://doi.org/10.1371/journal.pone.0229984 Martínez-Meyer E, Peterson AT, Servín JI, Kiff LF (2006) Ecological niche modelling and prioritizing areas for species reintroductions. ORYX 40:. https://doi.org/10.1017/S0030605306001360 Moghbel Esfahani F, Alavi SJ, Hosseini SM, Tabari Kochaksarai M (2023) Determining the habitat suitability of Quercus castaneifolia C. A. Mey In order to plan restoration using species distribution modeling. Forest Research and Development 9:419–436. https://doi.org/10.30466/jfrd.2023.54577.1654 Mohan R, Maurya A, Nair NK, et al (2024) Multi-Model Analysis to Predict the Potential Suitable Habitat for Quercus oblongata D. Don (Fagaceae) in the Western Himalayan Region. Environment and Ecology 42:687–696. https://doi.org/10.60151/envec/zoyl8560 Moore CE, Meacham-Hensold K, Lemonnier P, et al (2021) The effect of increasing temperature on crop photosynthesis: From enzymes to ecosystems. J Exp Bot 72:. https://doi.org/10.1093/jxb/erab090 Moradi S, Sheykhi Ilanloo S, Kafash A, Yousefi M (2019) Identifying high-priority conservation areas for avian biodiversity using species distribution modeling. Ecol Indic 97:. https://doi.org/10.1016/j.ecolind.2018.10.003 Muñoz-Mas R, Gil-Martínez E, Oliva-Paterna FJ, et al (2019) Tree-based ensembles unveil the microhabitat suitability for the invasive bleak (Alburnus alburnus L.) and pumpkinseed (Lepomis gibbosus L.): Introducing XGBoost to eco-informatics. Ecol Inform 53:. https://doi.org/10.1016/j.ecoinf.2019.100974 Naimi B, Hamm NAS, Groen TA, et al (2014) Where is positional uncertainty a problem for species distribution modelling? Ecography 37:. https://doi.org/10.1111/j.1600-0587.2013.00205.x Olden JD, Lawler JJ, Poff NL (2008) Machine learning methods without tears: A primer for ecologists. Quarterly Review of Biology 83 Pebesma E, Bivand R (2023) Spatial Data Science: With Applications in R R Core Team (2024) A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. In: URL https://www.R-project.org/. Ramírez-Albores JE, Prieto-Torres DA, Gordillo-Martínez A, et al (2021) Insights for protection of high species richness areas for the conservation of Mesoamerican endemic birds. Divers Distrib 27:. https://doi.org/10.1111/ddi.13153 Rosa IMD, Smith MJ, Wearn OR, et al (2016) The Environmental Legacy of Modern Tropical Deforestation. Current Biology 26:. https://doi.org/10.1016/j.cub.2016.06.013 Rushton SP, Ormerod SJ, Kerby G (2004) New paradigms for modelling species distributions? Journal of Applied Ecology 41 Sagheb Talebi K, Sajedi T, Pourhashemi M (2014) Forests of Iran: A treasure from the past, a hope for the future. Springer, Dordrecht Salehi Shanjani P, Javadi H, Rasoulzadeh L, Amirkhani M (2018) Evaluation of genetic differentiation among healthy and infected Buxus hyrcana with boxwood blight using RAPD and ISSR markers. N Z J For Sci 48:. https://doi.org/10.1186/s40490-018-0120-z Shahnaseri G, Malekian M, Pourmoghadam K (2023) Habitat loss of the chestnut-leaved oak (Quercus castaneifolia) in the Hyrcanian forests of Iran: impacts of anthropogenic factors on forest thinning and degradation. Glob Ecol Conserv 46:. https://doi.org/10.1016/j.gecco.2023.e02600 Sheykhi Ilanloo S, Khani A, Kafash A, et al (2021) Applying opportunistic observations to model current and future suitability of the Kopet Dagh Mountains for a Near Threatened avian scavenger. Avian Biol Res 14:. https://doi.org/10.1177/1758155920962750 Sillero N, Arenas-Castro S, Enriquez‐Urzelai U, et al (2021) Want to model a species niche? A step-by-step guideline on correlative ecological niche modelling. Ecol Modell 456:. https://doi.org/10.1016/j.ecolmodel.2021.109671 Sillero N, Campos JC, Arenas-Castro S, Barbosa AM (2023) A curated list of R packages for ecological niche modelling. Ecol Modell 476 Sohrabi H (2025) Does UNESCO designation enhance forest protection? Evidence from the Hyrcanian national forest inventory. Trees, Forests and People 21:100956. https://doi.org/https://doi.org/10.1016/j.tfp.2025.100956 Ssali F, Mugerwa B, van Heist M, et al (2023) Plant diversity and composition vary with elevation on two equatorial high mountains in Uganda: baselines for assessing the influence of climate change. Alp Bot 133:. https://doi.org/10.1007/s00035-023-00301-9 Velazco SJE, Rose MB, de Andrade AFA, et al (2022) flexsdm: An r package for supporting a comprehensive and flexible species distribution modelling workflow. Methods Ecol Evol 13:. https://doi.org/10.1111/2041-210X.13874 Vignali S, Barras AG, Arlettaz R, Braunisch V (2020) SDMtune: An R package to tune and evaluate species distribution models. Ecol Evol 10:. https://doi.org/10.1002/ece3.6786 Wang SW, Boru BH, Njogu AW, et al (2020) Floristic composition and endemism pattern of vascular plants in Ethiopia and Eritrea. J Syst Evol 58:. https://doi.org/10.1111/jse.12527 Wani Z, Khan S, Satish K, et al (2024) Ensemble modelling reveals shrinkage of suitable habitat for Himalayan Boxwood (Buxus wallichiana Bail.) under climate change – implications for conservation. Phytocoenologia 52:. https://doi.org/10.1127/phyto/2024/0427 Wani ZA, Satish K V., Islam T, et al (2023) Habitat suitability modelling of Buxus wallichiana Bail.: an endemic tree species of Himalaya. Vegetos 36:. https://doi.org/10.1007/s42535-022-00428-w Winkler K, Fuchs R, Rounsevell M, Herold M (2021) Global land use changes are four times greater than previously estimated. Nat Commun 12:. https://doi.org/10.1038/s41467-021-22702-2 Zhang M, Liu N, Harper R, et al (2017) A global review on hydrological responses to forest change across multiple spatial scales: Importance of scale, climate, forest type and hydrological regime. J Hydrol (Amst) 546 Zhang M, Wei X (2021) Deforestation, forestation, and water supply. Science (1979) 371 Zu K, Luo A, Shrestha N, et al (2019) Altitudinal biodiversity patterns of seed plants along Gongga Mountain in the southeastern Qinghai–Tibetan Plateau. Ecol Evol 9:. https://doi.org/10.1002/ece3.5483 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers agreed at journal 14 May, 2026 Reviewers agreed at journal 14 May, 2026 Reviewers agreed at journal 12 May, 2026 Reviewers agreed at journal 15 Apr, 2026 Reviewers agreed at journal 25 Mar, 2026 Reviewers invited by journal 25 Mar, 2026 Editor assigned by journal 23 Feb, 2026 Submission checks completed at journal 23 Feb, 2026 First submitted to journal 12 Feb, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8858849","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":612123847,"identity":"5b48bf9b-4f1c-4420-b592-d194e4c4620d","order_by":0,"name":"Aref Hesabi","email":"","orcid":"","institution":"Tarbiat Modares University","correspondingAuthor":false,"prefix":"","firstName":"Aref","middleName":"","lastName":"Hesabi","suffix":""},{"id":612123848,"identity":"229cb735-abc2-4b74-b966-3ec5fa83dd40","order_by":1,"name":"Seyed Jalil Alavi","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA5UlEQVRIie3RMQrCMBSA4VcKugRcU5D2CimForfRqYuCIEgGKSmFdvIoDiI4pwTskgPUTXcHxUkQMUUUcUgdHfJNIXk/JATAMP6QxcDiANhVa/Le5Q1JPYCD35PXwJB9Jlo2s/f8mvWiZZ6uLpTG0Mm5Jaj2Yi1SLDI83sjt1JFSAJYDKKQ2AcKRVEk1Ik6SqUtWAAXTJu1TcZM4CqtRcE3uMXjNCSICUTxQSegkzAbSmKRoIroU++otsz7bCuTLIdMmfp6vDkcSe2GZrndsHrtuKcRZm6RfG+j5vRqe9tQwDMOoPQA/V1J4E+eCkAAAAABJRU5ErkJggg==","orcid":"","institution":"Tarbiat Modares University","correspondingAuthor":true,"prefix":"","firstName":"Seyed","middleName":"Jalil","lastName":"Alavi","suffix":""},{"id":612123849,"identity":"da7f6c1e-951f-484a-9add-a6ec07aee4d0","order_by":2,"name":"Omid Esmailzadeh","email":"","orcid":"","institution":"Tarbiat Modares University","correspondingAuthor":false,"prefix":"","firstName":"Omid","middleName":"","lastName":"Esmailzadeh","suffix":""}],"badges":[],"createdAt":"2026-02-12 07:24:41","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8858849/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8858849/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":105586552,"identity":"34abe1e5-7cee-4d54-a402-fdce27fe4e91","added_by":"auto","created_at":"2026-03-27 15:35:03","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":761627,"visible":true,"origin":"","legend":"\u003cp\u003eResearch area and sampled presence points\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8858849/v1/83fd5029479f265a7b349e17.jpeg"},{"id":105727999,"identity":"61e91e55-be48-4401-a83d-41440e73b0bc","added_by":"auto","created_at":"2026-03-30 11:07:47","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":129406,"visible":true,"origin":"","legend":"\u003cp\u003eRelative importance of variables using the Gain method (Combination 1)\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8858849/v1/826113bfec84eed95fd3f0a1.png"},{"id":105586559,"identity":"d6bc0736-cc0b-4407-8e11-90807adcea59","added_by":"auto","created_at":"2026-03-27 15:35:03","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":127555,"visible":true,"origin":"","legend":"\u003cp\u003eRelative importance of variables using the Gain method (Combination 2)\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8858849/v1/9a478312940b02f5d34e282c.png"},{"id":105903855,"identity":"c9c0a3cd-69da-4fd1-a534-e6ac88f3e0d3","added_by":"auto","created_at":"2026-04-01 09:55:28","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":134039,"visible":true,"origin":"","legend":"\u003cp\u003eRelative importance of variables using the Gain method (Combination 3)\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8858849/v1/8d4d2b4b0604b19c0e6f16f9.png"},{"id":105728139,"identity":"439098f0-3c02-449f-92b9-826be1ca26b6","added_by":"auto","created_at":"2026-03-30 11:10:06","extension":"jpeg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":254491,"visible":true,"origin":"","legend":"\u003cp\u003eBoxwood response curve to 4 effective environmental variables based on Gain method (Combination 1) (upper left bio3, upper right LS, lower left bio 15, lower right bio12)\u003c/p\u003e","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8858849/v1/7278c4dae916607cd508ccad.jpeg"},{"id":105586554,"identity":"4911a7b7-e37b-47c7-beb1-7b777a49e691","added_by":"auto","created_at":"2026-03-27 15:35:03","extension":"jpeg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":269364,"visible":true,"origin":"","legend":"\u003cp\u003eBoxwood response curve to 4 effective environmental variables based on Gain method (Combination 2) (upper left bio4, upper right bio8, lower left ChND, lower right LS)\u003c/p\u003e","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8858849/v1/5c27f7c29df57962fe3e3a49.jpeg"},{"id":105728135,"identity":"f0caef28-6dfb-4443-bf89-84f98a75e4c9","added_by":"auto","created_at":"2026-03-30 11:10:04","extension":"jpeg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":202349,"visible":true,"origin":"","legend":"\u003cp\u003eBoxwood response curve to 4 effective environmental variables based on Gain method (Combination 3)(upper Land Use, lower left bio3, lower right Bio1)\u003c/p\u003e","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8858849/v1/349e10faaee1a6faff8ac438.jpeg"},{"id":105727998,"identity":"9cef68ac-2c03-4db2-8094-ce136032ce62","added_by":"auto","created_at":"2026-03-30 11:07:47","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":698697,"visible":true,"origin":"","legend":"\u003cp\u003eHabitat suitability map for Buxus hyrcana in the Hyrcanian forests (Combination 1)\u003c/p\u003e","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-8858849/v1/f55fa2180eb4681412f10ee7.png"},{"id":105586557,"identity":"1b6bcacb-5a8b-4ead-b7c5-791529f3d268","added_by":"auto","created_at":"2026-03-27 15:35:03","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":699964,"visible":true,"origin":"","legend":"\u003cp\u003eHabitat suitability map for Buxus hyrcana in the Hyrcanian forests (Combination 2)\u003c/p\u003e","description":"","filename":"floatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-8858849/v1/5d647afbcffd1b6515351f48.png"},{"id":105586560,"identity":"64e5e671-badc-4347-984f-f588b2910656","added_by":"auto","created_at":"2026-03-27 15:35:03","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":710791,"visible":true,"origin":"","legend":"\u003cp\u003eHabitat suitability map for Buxus hyrcana in the Hyrcanian forests (Combination 3)\u003c/p\u003e","description":"","filename":"floatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-8858849/v1/eb391d5276b564d1c589948b.png"},{"id":106414884,"identity":"8a8e2df1-0166-4641-a519-4c0bf2b34512","added_by":"auto","created_at":"2026-04-08 10:29:50","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":4957075,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8858849/v1/cf281b56-84fd-43f1-b230-002fbb0422da.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Machine Learning Meets Ecology: XGBoost‑Based Prediction of Endangered Species Refugia Using Multi‑Source Environmental Data","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThe global population is projected to reach 9\u0026nbsp;billion by 2026, intensifying demands for housing, food, energy, and natural resources that threaten natural vegetation cover unless sustainable development strategies are implemented. Land-use change has affected 32% of the Earth's terrestrial surface during the period 1960\u0026ndash;2019 (Winkler et al. \u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Throughout history, deforestation and forest degradation have profoundly transformed the Earth's land surface (Ellis et al. \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Of the world's natural forest cover, only 40% maintains high ecosystem integrity after extensive clearing for agriculture and other anthropogenic uses (Grantham et al. \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Forest ecosystem degradation generates far-reaching environmental and societal consequences (Barlow et al. \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2016\u003c/span\u003e), serving as primary drivers of the global biodiversity crisis (Rosa et al. \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e2016\u003c/span\u003e) while contributing significantly to carbon emissions, thereby exacerbating climate change (Baccini et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2012\u003c/span\u003e), undermining water provision and hydrological regulation (Zhang and Wei \u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), and adversely affecting the livelihoods of forest-dependent communities (Hou et al. \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Zhang et al. \u003cspan citationid=\"CR72\" class=\"CitationRef\"\u003e2017\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe Hyrcanian forests, extending along the southern shores of the Caspian Sea across Iran and parts of the Republic of Azerbaijan, constitute unique ancient temperate broadleaf ecosystems inscribed as a UNESCO World Heritage Site in 2019. These Tertiary-period forests preserve an evolutionary legacy spanning millions of years (Sagheb Talebi et al. \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2014\u003c/span\u003e) but face persistent threats including deforestation, land-use change, soil erosion (Ahmadi-sani et al. \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), overgrazing, and illegal logging that jeopardize their ecological integrity (Akbarzadeh et al. \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Shahnaseri et al. \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Recent studies within UNESCO-registered sites reveal significant declines in regeneration density, with anthropogenic pressures remaining critical threats: livestock grazing occurs in 64% of areas (23% at high intensity), and 15% are affected by illegal logging that disrupts natural recovery processes and threatens ecosystem sustainability (Sohrabi \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSpecies distributions result from complex interactions among climatic, topographic, biotic, and anthropogenic factors (Austin \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2002\u003c/span\u003e; Guisan and Thuiller \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2005\u003c/span\u003e), making understanding these drivers fundamental for biodiversity conservation, ecological restoration, and sustainable forest management (Mart\u0026iacute;nez-Meyer et al. \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Rushton et al. \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2004\u003c/span\u003e). Species Distribution Models (SDMs) have become indispensable tools, enabling prediction of potential ranges, identification of biodiversity hotspots, and prioritization of conservation areas (Moradi et al. \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Ram\u0026iacute;rez-Albores et al. \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). By integrating species occurrence records with environmental predictors, SDMs provide spatially explicit probability maps that guide restoration planning, optimize resource allocation, and reduce management failure risks (Dolos et al. \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Li and Wang \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2013\u003c/span\u003e). Comprehensive ecological knowledge of potential distribution and suitable habitats is crucial for conservation planning of declining species populations (Akrim et al. \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Since their emergence in the 1980s, SDMs have evolved from regression-based approaches (Guisan et al. \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2002\u003c/span\u003e) to advanced machine learning algorithms capable of capturing complex, non-linear ecological relationships (Jane Elith et al. \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Olden et al. \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2008\u003c/span\u003e). These techniques, known as niche-based models, habitat models, habitat suitability models, Environmental Niche Models (ENMs), climate envelopes, and Ecological Niche Models (Sillero et al. \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2021\u003c/span\u003e, \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), have been widely applied to identify biodiversity hotspots (Moradi et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), assess habitat suitability (Campbell et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), and evaluate climate change impacts on biodiversity (Hotta et al. \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Liu et al. \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Sheykhi Ilanloo et al. \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). The transition toward machine learning approaches stems from their flexible fitting capabilities, automated variable selection, and capacity to handle diverse data types (Jane Elith et al. \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2006\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAmong machine learning approaches, tree-based algorithms are particularly prominent due to their flexibility in handling non-linearity and ensemble modeling advantages (J. Elith et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2008\u003c/span\u003e; Mu\u0026ntilde;oz-Mas et al. \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). eXtreme Gradient Boosting (XGBoost), an efficient and scalable gradient boosting implementation based on decision trees (Chen and Guestrin \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2016\u003c/span\u003e), incorporates L1/L2 regularization, random sampling, and second-order optimization, enabling high accuracy, robustness against incomplete data, and reduced overfitting (Chen and Guestrin \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). These features make XGBoost particularly powerful for ecological and environmental modeling, especially for endangered species with complex ecological requirements (Fan et al. \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2018\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eEndemic and threatened species are especially vulnerable to habitat loss and climate change, as narrow ecological ranges and elevational distributions make them highly sensitive to disturbance (Hu et al. \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Kidane et al. \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Wang et al. \u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Zu et al. \u003cspan citationid=\"CR74\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Anthropogenic climate change exacerbates these vulnerabilities, driving species range shifts and increasing extinction risk (Asefa et al. \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). \u003cem\u003eBuxus hyrcana\u003c/em\u003e Pojark, the Hyrcanian boxwood, is a keystone evergreen broadleaf species of northern Iranian forests. Due to its high-quality timber, the species has long been subjected to illegal harvesting, which combined with widespread habitat degradation has driven dramatic declines in its natural range, resulting in IUCN Red List classification. Despite high seed production and regeneration potential, continued anthropogenic pressures and climate change threaten species persistence and habitat ecological integrity (Biru et al. \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Ssali et al. \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Official reports indicate boxwood habitats across Gilan, Mazandaran, and Golestan provinces totaled approximately 72,000 hectares in 2013 (Salehi Shanjani et al. \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2018\u003c/span\u003e), with escalating anthropogenic pressures elevating the species to conservation priority status. Although not yet extinct, ecosystems have been severely degraded, though this semi-endemic species (also occurring in the Republic of Azerbaijan) retains high seed production capacity and considerable viability that may support generational continuity if suitable habitats are preserved.\u003c/p\u003e \u003cp\u003eGiven the ecological significance of Hyrcanian forests and conservation priority of \u003cem\u003eB. hyrcana\u003c/em\u003e, precise habitat identification is essential for guiding restoration and management strategies. Advanced SDMs like XGBoost, capable of detecting complex non-linear relationships between species occurrence and environmental variables, provide valuable opportunities to generate accurate habitat suitability maps and identify conservation refugia. This study applies XGBoost to model \u003cem\u003eB. hyrcana\u003c/em\u003e potential distribution using multi-source environmental data, including bioclimatic, topographic, and land-use variables. Specifically, the study aims to (i) evaluate model performance across different environmental data combinations, (ii) identify most influential variables shaping species distribution, and (iii) generate high-resolution habitat suitability maps to inform conservation prioritization and restoration planning. By integrating advanced machine learning with ecological knowledge, this research provides a robust framework for long-term \u003cem\u003eB. hyrcana\u003c/em\u003e conservation and contributes to broader biodiversity safeguarding efforts within ancient temperate forest ecosystems.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cp\u003e\u003cstrong\u003eStudy Area\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe Hyrcanian forests constitute a narrow, longitudinal belt along the northern slopes of the Alborz Mountain range, stretching from Astara in the northwest to Golidaghi in Bojnord Province in the northeast. This unique forest ecosystem spans approximately 800 km in length and 110 km in width, encompassing a total area of 1.85 million hectares\u0026mdash;equivalent to 15% of Iran\u0026rsquo;s forested lands and 1.1% of its national territory (Sagheb Talebi et al. 2014) (Fig 1).\u003c/p\u003e\n\u003cp\u003eAltitudinal variation within the Hyrcanian Forests ranges from sea level to elevations exceeding 2,800 meters. Climatic conditions exhibit pronounced spatial gradients: mean annual precipitation varies from 530 mm in the eastern regions to 1,530 mm in the west, with localized maxima reaching up to 2,000 mm near Asalem. Over the past decade, mean annual temperatures have averaged 15\u0026deg;C in the western zones and 17.5\u0026deg;C in the eastern sectors, reflecting the region\u0026rsquo;s diverse microclimatic regimes.\u003c/p\u003e\n\u003cp dir=\"\"\u003e\u003cstrong\u003eOccurrence Data Collection and Processing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA total of 570 georeferenced presence records for \u003cem\u003eBuxus hyrcana\u003c/em\u003e were compiled from field surveys across the Hyrcanian forest range in northern Iran (Fig 1). Coordinate uncertainty was addressed by buffering records with a 100-m radius to account for potential geolocation errors, and no records fell within this buffer of others to avoid spatial overlap. Due to the absence of verified absence data, pseudo-absence points (n = 1,140, twice the number of presence records) were generated using random sampling within the study area extent, excluding a 1-km buffer around presence points to minimize bias toward known habitats. This approach followed established practices for balanced datasets in species distribution modeling. Pseudo-absence generation was implemented via the sample_pseudoabs() function from the \u003cem\u003eflexsdm\u003c/em\u003e R package (Velazco et al. 2022).\u003c/p\u003e\n\u003cp\u003eTo mitigate spatial autocorrelation and sampling bias, spatial thinning was applied using the thinData() function from the \u003cem\u003eSDMtune\u003c/em\u003e R package (Vignali et al. 2020). Thinning retained one occurrence per 1-km\u0026sup2; raster cell (matching environmental data resolution), resulting in a final dataset of about 200 presence points. All spatial analyses were performed in the WGS 84 coordinate reference system (CRS: EPSG:4326) using the \u003cem\u003esf\u003c/em\u003e (Pebesma and Bivand 2023) and \u003cem\u003eterra\u003c/em\u003e (Hijmans R 2025) packages in R programming language (R Core Team, 2025).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEnvironmental Variables\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eEnvironmental predictors were selected to capture bioclimatic, topographic, and anthropogenic drivers of \u003cem\u003eB. hyrcana\u003c/em\u003e distribution, following established frameworks for species distribution models (SDMs; Guisan and Thuiller, 2005). A standardized set of 19 bioclimatic variables (Bio1\u0026ndash;Bio19) was derived from two global datasets to evaluate data source impacts: WorldClim version 2.1 (1970\u0026ndash;2000 period; Fick and Hijmans, 2017) and CHELSA version 2.1 (1980\u0026ndash;2010 period; Karger et al., 2017). Both datasets were downloaded from their respective portals (WorldClim: https://www.worldclim.org; CHELSA: https://chelsa-climate.org) at 1-km resolution (30 arc-seconds) (Table 1). Integrating these variables\u0026mdash;either partially or in full\u0026mdash;has been shown to significantly improve the predictive performance of species distribution models (SDMs) and yield more nuanced insights into the ecological drivers of species ranges (Amiri et al., 2020; Hesabi et al., 2025). Variables were cropped to the study area extent using a polygon buffer of 50 km around the Hyrcanian forests and masked to land areas, with no gap-filling or interpolation applied.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1\u0026nbsp;\u003c/strong\u003eDescription of variables used for modeling in the study area\u003c/p\u003e\n\u003ctable style=\"width: 4.8e+2pt;border: none;\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eCategory\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eVariable\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eDescription\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eUnite\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eSource\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eUse for modeling\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"19\"\u003e\n \u003cp\u003e\u003cstrong\u003eClimate\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;Bio1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAnnual Mean Temperature\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e◦C\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"19\"\u003e\n \u003cp\u003eWorldclim \u0026amp; Chelsa\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026uuml;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;Bio2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMean Diurnal Range\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e◦C\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ucirc;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;Bio3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eIsothermality\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026uuml;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;Bio4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eTemperature Seasonality (Standard deviation)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026uuml;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;Bio5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMax Temperature of Warmest Month\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e◦C\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ucirc;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;Bio6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMin Temperature of Coldest Month\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e◦C\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ucirc;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;Bio7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eTemperature Annual Range\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e◦C\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ucirc;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;Bio8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMean Temperature of Wettest Quarter\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e◦C\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026uuml;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;Bio9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMean Temperature of Driest Quarter\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e◦C\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ucirc;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; Bio10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMean Temperature of Warmest Quarter\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e◦C\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ucirc;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;Bio11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMean Temperature of Coldest Quarter\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e◦C\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ucirc;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; Bio12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAnnual Precipitation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003emm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026uuml;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; Bio13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePrecipitation of Wettest Month\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003emm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ucirc;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; Bio14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePrecipitation of Driest Month\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003emm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ucirc;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; Bio15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePrecipitation Seasonality (Coefficient of variation)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026uuml;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; Bio16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePrecipitation of Wettest Quarter\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003emm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ucirc;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; Bio17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePrecipitation of Driest Quarter\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003emm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ucirc;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; Bio18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePrecipitation of Warmest Quarter\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003emm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ucirc;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; Bio19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePrecipitation of Coldest Quarter\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003emm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ucirc;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"9\"\u003e\n \u003cp\u003e\u003cstrong\u003eTopography\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp;Elevation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eElevation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003em\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"9\"\u003e\n \u003cp\u003eSRTM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ucirc;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp;TRASP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eTopographic solar radiation aspect index\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026uuml;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;Slope\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSlope\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ucirc;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;TWI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eTopographic Wetness Index\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026uuml;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;TPI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eTopography Position Index\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026uuml;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;VD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eValley Depth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026uuml;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;ChND\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eChannel Network Distance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026uuml;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;CI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eConvergence Index\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026uuml;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;LS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eLength and Steepness Factor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026uuml;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp dir=\"RTL\"\u003e\u003cspan dir=\"LTR\"\u003e\u0026nbsp;\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTopographic Variables\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFor this study, a DEM with a spatial resolution of 1 kilometer was acquired from the Shuttle Radar Topography Mission (SRTM). All primary and secondary topographic variables were computed using SAGA GIS software (version 9.2), ensuring standardized derivation and reproducibility across the study area (Table 1). Among topographic attributes, elevation, slope, and aspect are considered primary variables and are widely employed in ecological modeling due to their strong influence on abiotic conditions and their ease of extraction from remote sensing data.\u003c/p\u003e\n\u003cp\u003eIn addition to these primary parameters, a suite of secondary topographic indices\u0026mdash;including the Topographic Wetness Index (TWI), Convergence Index (CI), Valley Depth (VD), Channel Network Distance (ChND), Slope Length and Steepness Factor (LS Factor), and Topographic Position Index (TPI)\u0026mdash;provides further insights into terrain-driven ecological gradients (Gama et al. 2016). These variables collectively modulate key environmental drivers such as solar radiation, temperature regimes, precipitation patterns, soil moisture availability, and microclimatic conditions, thereby influencing species distributions and community composition (Baker and Barnes 1998).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLand Use Layer Preparation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe Land Cover Explorer layer from ArcGIS Living Atlas of the World was used as the source of land use/land cover data; this dataset provides annual global maps at ~10 m resolution derived from Sentinel-2 imagery and machine-learning classification (ESRI 2024). The 2024 product was selected to represent the most recent land-use status and to better reflect contemporary potential habitats for \u003cem\u003eBuxus hyrcana.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eTo harmonize spatial resolution with bioclimatic and topographic predictors, the 2024 land-use raster was resampled to 1 km (30 arc-seconds) using modal (majority) resampling, thereby preserving categorical integrity. Resampling and reclassification were performed in R (R Core Team, 2025) using spatial raster tools. The final categorical layer was encoded for model use by converting reclassified classes into binary dummy variables (one-hot encoding) prior to fitting the XGBoost algorithm, ensuring proper representation of categorical effects in the model.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMulticollinearity Assessment\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo ensure model stability and interpretability, multicollinearity among bioclimatic and topographic predictors was evaluated using the Variance Inflation Factor (VIF), a widely adopted diagnostic metric in ecological modeling, econometrics, and machine learning. VIF quantifies the extent to which the variance of a regression coefficient is inflated due to linear dependence with other predictors, thereby serving as a proxy for redundancy and collinearity (Kutner et al. 2005).\u003c/p\u003e\n\u003cp\u003eEach predictor was assessed individually, and VIF values were computed to identify overlapping information among variables. High VIF scores indicate strong collinearity, which can lead to unstable parameter estimates, reduced model generalizability, and misleading ecological interpretations (Hastie et al. 2021). By eliminating or consolidating collinear variables, VIF screening contributes to the development of parsimonious and robust models.\u003c/p\u003e\n\u003cp\u003eThe analysis was conducted using the vifstep function from the usdm package in R (Naimi et al., 2014; R Core Team, 2025), which iteratively removes the variable with the highest VIF until all remaining predictors fall below a predefined threshold. In this study, a conservative cutoff of VIF \u0026lt; 10 was applied, consistent with best practices in species distribution modeling. The final set of retained variables is presented in Table 1.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModeling Process\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eInitial modeling efforts employed the Boosted Regression Trees (BRT) algorithm, a well-established ensemble method based on decision tree learning and iterative error reduction. BRT has been widely applied in ecological studies, particularly for species distribution modeling\u0026nbsp;(J. Elith et al. 2008). However, recent advancements in machine learning have led to the development of more efficient and scalable algorithms, notably XGBoost (Extreme Gradient Boosting), which offers enhanced performance in terms of predictive accuracy, computational speed, handling of missing data, and mitigation of overfitting (Chen and Guestrin 2016; Wang et al. 2020).\u003c/p\u003e\n\u003cp\u003eFollowing a comparative review of current literature and expert recommendations, XGBoost was selected as the preferred modeling approach for this study. Its adoption not only improved model precision but also contributed to greater stability and interpretability of the final predictions.\u003c/p\u003e\n\u003cp\u003eModeling was conducted using the tidymodels framework in R (Kuhn and Wickham 2020). The XGBoost algorithm requires both presence and absence data. Since verified absence records were unavailable, pseudo-absence points were generated using the the sample_pseudoabs() function from the flexsdm package (Velazco et al. 2022). Environmental predictor values\u0026mdash;including bioclimatic and physiographic variables\u0026mdash;were extracted at each presence and pseudo-absence location using the sdm_extract() function (Barbet-Massin et al. 2012). The final modeling matrix included the following variables: Bio1, Bio3, Bio4, Bio8, Bio12, Bio15, slope, Topographic Wetness Index (TWI), Valley Depth (VD), Solar Radiation Index (SRI), Convergence Index (CI), LS Factor, and Channel Network Distance (ChND).\u003c/p\u003e\n\u003cp\u003eData were partitioned into training (70%) and testing (30%) subsets. To prevent overfitting and ensure model generalizability, 10-fold cross-validation was applied. Hyperparameter tuning was performed using a grid search across combinations of number of trees (500\u0026ndash;2500), tree depth (3, 5, 7), learning rate (0.3, 0.1, 0.05, 0.01, 0.005), number of predictors sampled per split (mtry = 2\u0026ndash;13), and minimum samples per node (10\u0026ndash;50). The optimal configuration\u0026mdash;mtry = 12, trees = 1500, min_n = 10, tree_depth = 7, learn_rate = 0.005\u0026mdash;yielded an AUC of 0.96, indicating excellent model performance.\u003c/p\u003e\n\u003cp\u003eModel evaluation was conducted using multiple metrics: Area Under the Curve (AUC) from the Receiver Operating Characteristic (ROC), True Skill Statistic (TSS), sensitivity, specificity (1-specificity), overall accuracy, and Cohen\u0026rsquo;s Kappa coefficient.\u003c/p\u003e\n\u003cp\u003eTo assess the influence of different bioclimatic datasets, three modeling combinations were tested:\u003c/p\u003e\n\u003cul type=\"disc\"\u003e\n \u003cli\u003e\u003cstrong\u003eCombination 1\u003c/strong\u003e: WorldClim bioclimatic variables + topographic predictors + presence/pseudo-absence data\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eCombination 2\u003c/strong\u003e: Chelsa bioclimatic variables + topographic predictors + presence/pseudo-absence data\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eCombination 3\u003c/strong\u003e: WorldClim bioclimatic variables + topographic predictors + land use layer + presence/pseudo-absence data\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThese combinations enabled comparative evaluation of dataset contributions to model performance and ecological inference.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eModel Performance and Discriminatory Capacity\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe predictive performance of the XGBoost model for \u003cem\u003eBuxus hyrcana\u003c/em\u003e distribution was assessed using widely accepted evaluation metrics, including the Area Under the Receiver Operating Characteristic Curve (AUC), True Skill Statistic (TSS), overall Accuracy, and the Kappa coefficient. The XGBoost model demonstrated strong predictive performance across all three data combinations, with performance metrics revealing distinct trade-offs between environmental data sources and model complexity. Detailed evaluation metrics are presented in Table 2.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2\u0026nbsp;\u003c/strong\u003eModel performance evaluation metrics for \u003cem\u003eBuxus hyrcana\u003c/em\u003e distribution modeling\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable style=\"width: 3.5e+2pt;\" dir=\"rtl\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cstrong\u003e\u003cspan dir=\"LTR\"\u003eXGboost model (Combination 3)\u003c/span\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cstrong\u003e\u003cspan dir=\"LTR\"\u003eXGboost model (Combination 2)\u003c/span\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cstrong\u003e\u003cspan dir=\"LTR\"\u003eXGboost model (Combination 1)\u003c/span\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cstrong\u003e\u003cspan dir=\"LTR\"\u003eEvaluation criteria\u003c/span\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cspan dir=\"LTR\"\u003e0.98\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cspan dir=\"LTR\"\u003e0.97\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cspan dir=\"LTR\"\u003e0.98\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cspan dir=\"LTR\"\u003eArea Under the Curve\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cspan dir=\"LTR\"\u003e0.86\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cspan dir=\"LTR\"\u003e0.45\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cspan dir=\"LTR\"\u003e0.67\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cspan dir=\"LTR\"\u003eTrue Skill Statistic\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cspan dir=\"LTR\"\u003e0.98\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cspan dir=\"LTR\"\u003e0.93\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cspan dir=\"LTR\"\u003e0.95\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cspan dir=\"LTR\"\u003eAccuracy\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cspan dir=\"LTR\"\u003e0.91\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cspan dir=\"LTR\"\u003e0.55\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cspan dir=\"LTR\"\u003e0.73\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp dir=\"RTL\"\u003e\u003cspan dir=\"LTR\"\u003eKappa\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eCombination 1 exhibited strong discriminatory power between presence and absence records (AUC = 0.98, TSS = 0.67, Accuracy = 0.95), indicating reliable capacity to distinguish suitable from unsuitable habitats (Fielding and Bell 1997). The Kappa coefficient of 0.73 demonstrates substantial agreement between predicted and observed distributions, substantially exceeding the minimum threshold of 0.60 typically required in species distribution modeling (Fielding and Bell 1997; Franklin 2013).\u003c/p\u003e\n\u003cp\u003eCombination 2 showed a marked decline in TSS (0.45) relative to Combination 1, representing a 0.22-point reduction in this threshold-sensitive metric, despite maintaining comparable AUC values (0.97). This pattern indicates that Chelsa-derived bioclimatic variables, while capturing broad-scale climatic variation, may be less discriminative for fine-scale presence-absence distinctions in \u003cem\u003eBuxus hyrcana\u003c/em\u003e. The Kappa coefficient of 0.55 falls within the \u0026quot;moderate agreement\u0026quot; range, suggesting that this data combination, while acceptable for broad biogeographic patterns, introduces classification uncertainty not evident in Combination 1.\u003c/p\u003e\n\u003cp\u003eCombination 3 achieved optimal performance across all metrics (TSS = 0.86, Accuracy = 0.98, Kappa = 0.91), substantially exceeding both previous combinations. The 0.19-point improvement in TSS relative to Combination 1 and 0.41-point improvement relative to Combination 2 demonstrates that the integration of land-use data captured critical environmental constraints on \u003cem\u003eBuxus hyrcana\u003c/em\u003e distribution not fully represented by climate and topography alone. This improvement is particularly significant given that TSS is sensitive to both commission and omission errors and provides robust evaluation independent of prevalence.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEnvironmental Variable Importance and Hierarchical Drivers\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eVariable importance analysis, based on the Gain criterion (representing average improvement in log-loss at split nodes), identified distinct hierarchies of environmental drivers across data combinations. This approach quantifies the relative contribution of each predictor to reducing model uncertainty, revealing which environmental gradients most strongly constrain \u003cem\u003eBuxus hyrcana\u003c/em\u003e distribution.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCombination 1: WorldClim Bioclimatic and Topographic Variables\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe importance ranking in Combination 1 identified isothermality (Bio3), Length-Slope Factor (LS Factor), precipitation seasonality (Bio15), and annual precipitation (Bio12) as the four dominant drivers (Fig 2). These variables collectively explain the primary environmental niche dimensions for \u003cem\u003eBuxus hyrcana\u003c/em\u003e. \u003cv:shapetype id=\"_x0000_t75\" coordsize=\"21600,21600\" o:spt=\"75\" o:preferrelative=\"t\" path=\"m@4@5l@4@11@9@11@9@5xe\" filled=\"f\" stroked=\"f\"\u003e\u0026nbsp;\u003cv:stroke joinstyle=\"miter\"\u003e\u0026nbsp;\u003cv:formulas\u003e\u0026nbsp;\u003cv:f eqn=\"if lineDrawn pixelLineWidth 0\"\u003e\u0026nbsp;\u003cv:f eqn=\"sum @0 1 0\"\u003e\u0026nbsp;\u003cv:f eqn=\"sum 0 0 @1\"\u003e\u0026nbsp;\u003cv:f eqn=\"prod @2 1 2\"\u003e\u0026nbsp;\u003cv:f eqn=\"prod @3 21600 pixelWidth\"\u003e\u0026nbsp;\u003cv:f eqn=\"prod @3 21600 pixelHeight\"\u003e\u0026nbsp;\u003cv:f eqn=\"sum @0 0 1\"\u003e\u0026nbsp;\u003cv:f eqn=\"prod @6 1 2\"\u003e\u0026nbsp;\u003cv:f eqn=\"prod @7 21600 pixelWidth\"\u003e\u0026nbsp;\u003cv:f eqn=\"sum @8 21600 0\"\u003e\u0026nbsp;\u003cv:f eqn=\"prod @7 21600 pixelHeight\"\u003e\u0026nbsp;\u003cv:f eqn=\"sum @10 21600 0\"\u003e\u0026nbsp;\u003c/v:f\u003e\u0026nbsp;\u003c/v:f\u003e\u0026nbsp;\u003c/v:f\u003e\u0026nbsp;\u003c/v:f\u003e\u0026nbsp;\u003c/v:f\u003e\u0026nbsp;\u003c/v:f\u003e\u0026nbsp;\u003c/v:f\u003e\u0026nbsp;\u003c/v:f\u003e\u0026nbsp;\u003c/v:f\u003e\u0026nbsp;\u003c/v:f\u003e\u0026nbsp;\u003c/v:f\u003e\u0026nbsp;\u003c/v:f\u003e\u0026nbsp;\u003c/v:formulas\u003e\n \u003cv:path o:extrusionok=\"f\" gradientshapeok=\"t\" o:connecttype=\"rect\"\u003e\u0026nbsp;\u003c/v:path\u003e\u0026nbsp; \u0026nbsp;\n \u003c/v:stroke\u003e\u0026nbsp;\u003c/v:shapetype\u003e\n \u003cv:shape id=\"_x0000_i1041\" type=\"#_x0000_t75\"\u003e\u0026nbsp;\u003cv:imagedata src=\"file:///C%3A/Users/btr8097/AppData/Local/Packages/oice_16_974fa576_32c1d314_330/AC/Temp/msohtmlclip1/01/clip_image001.png\" o:title=\"\"\u003e\u0026nbsp;\u003c/v:imagedata\u003e\u0026nbsp;\u003c/v:shape\u003e\n\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCombination 2: Chelsa Bioclimatic and Topographic Variables\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn Combination 2, the importance hierarchy shifted, with mean temperature of warmest quarter (Bio8), temperature seasonality (Bio4), LS Factor, and Channel Network Distance ranked as top predictors (Fig 3). Notably, the dominance of Bio8 and Bio4 in this combination versus Bio3 in Combination 1 reflects database-specific differences in temperature representation between WorldClim and Chelsa sources. The persistence of LS Factor across both combinations underscores the consistent importance of topographic heterogeneity. The emergence of Channel Network Distance suggests that Chelsa enhanced spatial resolution may better capture fine-scale hydrological connectivity.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCombination 3: Integrated Environmental Layers\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCombination 3 revealed a fundamentally reorganized importance hierarchy, with land use, isothermality (Bio3), annual mean temperature (Bio1), and annual precipitation (Bio12) as the dominant predictors (Fig 4). This reorganization carries substantial ecological significance. Land Use emerged as the single most influential variable, accounting for the substantial performance improvement observed in evaluation metrics, indicating that current or recent human land-use practices represent the primary contemporary constraint on \u003cem\u003eBuxus hyrcana\u003c/em\u003e habitat suitability. Bio3, Bio1, and Bio12 retained importance similar to Combination 1, but with shifted relative rankings, suggesting that land-use patterns correlate with and partially subsume climate-topography effects. This pattern indicates that \u003cem\u003eBuxus hyrcana\u003c/em\u003e distribution in the Hyrcanian forests is jointly determined by underlying bioclimatic-topographic niches and anthropogenic landscape modification.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSpecies-Environment Response Functions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eResponse curves derived from XGBoost predictions illustrate the functional relationships between \u003cem\u003eBuxus hyrcana\u003c/em\u003e occurrence probability and leading environmental predictors across gradient space. These non-parametric response curves reveal the realized ecological niche dimensions and identify critical environmental thresholds. The response curves for Combination 1 (Fig 5) reveal Bio3 (Isothermality) showed a marked threshold response, with maximum occurrence probability concentrated between 27-32% isothermality values. Probabilities decline steeply above 32%, indicating that environments with highly variable seasonal temperature cycles exceed physiological tolerance limits of \u003cem\u003eBuxus hyrcana\u003c/em\u003e. This threshold response suggests that stabilized thermal regimes, characteristic of humid continental and subtropical climates, represent fundamental niche requirements. LS Factor, with a unimodal response curve and optimal suitability at intermediate LS values (approximately 2.5-4.5), indicates that boxwood establishes on moderately sloping terrain with moderate water flow dynamics. Both very low LS values (flat terrain with poor drainage) and very high values (steep erosion-prone slopes) show reduced suitability, reflecting trade-offs between waterlogging and erosion stress.\u003c/p\u003e\n\u003cp\u003eBio15 (Precipitation Seasonality) showed an inverse relationship between precipitation seasonality and occurrence probability, with maximum suitability at low seasonality values (\u0026lt;60 coefficient of variation). This pattern indicates that \u003cem\u003eBuxus hyrcana\u003c/em\u003e is restricted to regions with relatively equitable year-round precipitation distribution, consistent with its distribution in humid evergreen forests. Bio12 (Annual Precipitation) exhibits a distinct threshold response, with \u003cem\u003eBuxus hyrcana\u003c/em\u003e occurrence probability rising sharply once annual precipitation exceeds approximately 1000 mm. Suitability continues to increase until reaching a plateau around 2500 mm, beyond which additional rainfall does not confer further ecological advantage. This pattern underscores the species\u0026rsquo; reliance on consistently moist environments, where water availability supports physiological processes and forest canopy stability.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn Combination 2, which explores the Chelsa climate\u0026ndash;topography niche space, the response curves (Fig 6) reaffirm the central role of temperature and seasonality in shaping \u003cem\u003eBuxus hyrcana\u0026rsquo;s\u003c/em\u003e ecological preferences, while also highlighting patterns unique to the Chelsa dataset. The mean temperature of the warmest quarter reveals a steadily declining occurrence probability across the temperature gradient, with the highest suitability observed at temperatures below 22\u0026deg;C. This suggests that although the species is present in subtropical regions, it tends to occupy cooler microhabitats or higher elevations, aligning with its known distribution in the montane Hyrcanian forests. Temperature seasonality exhibits a unimodal response, with optimal suitability occurring within a standard deviation range of 700 to 900\u0026deg;C. This indicates a preference for climates that maintain moderate seasonal temperature fluctuations, avoiding both highly stable and highly variable thermal regimes. The LS factor and channel network distance continue to demonstrate monotonic or complex relationships similar to those observed in Combination 1, reinforcing the consistent influence of hydrological and topographic variables across different environmental data sources.\u003c/p\u003e\n\u003cp\u003eIn Combination 3, which integrates land-use, climate, and topographic variables (Fig 7), the inclusion of land-use data reveals critical interactions that further constrain the realized niche of \u003cem\u003eBuxus hyrcana\u003c/em\u003e. The highest occurrence probabilities are associated with land-use categories corresponding to intact or minimally disturbed forests. In contrast, areas characterized by agricultural activity, urban development, or significant anthropogenic disturbance exhibit markedly reduced suitability. This pattern underscores the dominant role of historical and current land-use practices in shaping the contemporary distribution of the species, effectively filtering the broader climate\u0026ndash;topography niche. The responses of annual precipitation, isothermality, and mean annual temperature in this combination mirror those observed in Combination 1, but with subtle shifts that reflect their interaction with land-use patterns. These modifications suggest a nested constraint structure, where only landscapes that simultaneously meet the climatic and topographic requirements and align with favorable land-use conditions can support viable populations of \u003cem\u003eBuxus hyrcana\u003c/em\u003e. This highlights the compounded impact of environmental suitability and human land-use history in determining the species\u0026rsquo; distribution.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSpatial Distribution Patterns and Habitat Suitability Mapping\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHabitat suitability maps generated by XGBoost provide spatially explicit predictions of potential distribution zones for \u003cem\u003eBuxus hyrcana\u003c/em\u003e across the Hyrcanian forest region (Fig 8-10). These maps translate multivariate species-environment relationships into actionable spatial predictions for conservation planning and biodiversity assessment.\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003eThe map derived from Combination 1 (Fig 8) delineates suitable habitats primarily in regions characterized by moderate temperature and precipitation regimes, with topographic heterogeneity further refining habitat suitability.\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003eIn Combination 2 (Fig 9), the predicted distribution pattern is broadly consistent with Combination 1, though with slightly reduced spatial precision, reflecting the lower overall model performance observed in this combination.\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003eThe most accurate and ecologically realistic predictions were obtained from Combination 3 (Fig 10), where the integration of land-use data enhanced the model\u0026rsquo;s ability to delineate suitable habitats.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eCollectively, these maps emphasize the critical role of climatic and topographic variables in shaping the distribution of \u003cem\u003eBuxus hyrcana\u003c/em\u003e, while also demonstrating that the careful integration of additional environmental layers can refine habitat suitability predictions. The coordinated analysis across three data combinations demonstrates that \u003cem\u003eBuxus hyrcana\u003c/em\u003e distribution in the Hyrcanian forests results from hierarchically nested environmental constraints. At the broadest scale, bioclimatic variables\u0026mdash;particularly temperature consistency (isothermality) and moisture availability (annual precipitation)\u0026mdash;define a fundamental niche envelope. Topographic variables refine this envelope by capturing local water availability and erosion dynamics. Most restrictively, contemporary land use defines the realized distribution by determining which climatically and topographically suitable areas remain available within intact or semi-intact forest ecosystems. This nested constraint structure has direct implications for species conservation and habitat restoration: expanding \u003cem\u003eBuxus hyrcana\u003c/em\u003e distribution will require not only climatic suitability and topographic compatibility, but also land-use policy changes that permit forest regeneration in currently degraded areas.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eModel Performance Analysis\u003c/h2\u003e \u003cp\u003eThe XGBoost model demonstrated consistently strong predictive performance across all tested combinations. Combination 1 achieved robust metrics (Accuracy\u0026thinsp;=\u0026thinsp;0.95, Kappa\u0026thinsp;=\u0026thinsp;0.73, AUC\u0026thinsp;=\u0026thinsp;0.98, TSS\u0026thinsp;=\u0026thinsp;0.67), while Combination 2 showed moderate declines (Accuracy\u0026thinsp;=\u0026thinsp;0.93, Kappa\u0026thinsp;=\u0026thinsp;0.55, AUC\u0026thinsp;=\u0026thinsp;0.97, TSS\u0026thinsp;=\u0026thinsp;0.45) likely due to Chelsa's interpolation challenges in mountainous terrains. The most notable improvement occurred in Combination 3, where land-use integration substantially enhanced performance (Accuracy\u0026thinsp;=\u0026thinsp;0.98, Kappa\u0026thinsp;=\u0026thinsp;0.91, AUC\u0026thinsp;=\u0026thinsp;0.98, TSS\u0026thinsp;=\u0026thinsp;0.86), highlighting the importance of incorporating anthropogenic variables alongside climatic and topographic predictors.\u003c/p\u003e \u003cp\u003eWhen compared with previous studies, our XGBoost models consistently outperformed established benchmarks. (Jane Elith et al. \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2006\u003c/span\u003e) reported mean AUC values ranging from 0.75 to 0.95 for MaxEnt across diverse ecosystems, while our models consistently achieved AUC values above 0.97. Similarly, although TSS values in Combinations 1 and 2 were lower than those reported by (Mohan et al. \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) for Quercus oblongata (TSS\u0026thinsp;=\u0026thinsp;0.86), the integration of physiographic and land-use variables in Combination 3 elevated TSS to 0.86, underscoring the synergistic effect of multi-dimensional predictors. (Asadi et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) compared multiple algorithms for Quercus castaneifolia habitats, with Random Forest achieving the highest AUC (0.77), followed by MaxEnt (0.75), SVM (0.72), KNN (0.71), and GLM (0.70). In contrast, our XGBoost models consistently outperformed these approaches across varying data combinations, confirming the algorithm's superior capacity for handling complex, non-linear ecological relationships.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003eAnalysis of Relative Variable Importance\u003c/h2\u003e \u003cp\u003eIn Combination 1 (WorldClim\u0026thinsp;+\u0026thinsp;topography), variables such as Bio3 (isothermality), LS Factor (slope length and steepness), Bio12 (annual precipitation), and Bio15 (precipitation seasonality) ranked highest in importance, emphasizing thermal stability and topographic constraints as foundational drivers of \u003cem\u003eB. hyrcana\u003c/em\u003e distribution. Combination 2 (Chelsa\u0026thinsp;+\u0026thinsp;topography) introduced Bio8 (mean temperature of the wettest quarter) and ChND (channel network distance) among top predictors, illustrating the algorithm's sensitivity to nuanced climatic data structures and their interactions in topographically complex regions. This shift highlights Chelsa's superior representation of local microclimates in mountainous areas, as validated by (Hesabi et al. \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), leading to greater emphasis on temperature-humidity dynamics over pure isothermality.\u003c/p\u003e \u003cp\u003eA striking pattern across combinations was the stability of core variables\u0026mdash;LS Factor, Bio12, and Bio3\u0026mdash;consistently ranking in the top five, underscoring their fundamental role in defining boxwood habitats influenced by slope stability, annual moisture, and thermal uniformity. Bio8 and ChND gained prominence in Combinations 2 and 3, reinforcing the species' reliance on moderate wet-season temperatures and proximity to water sources. However, in Combination 3, the land-use variable exerted the greatest influence, ranking first and comprising over 20% of total importance via the Gain index. This dominance stems from the near-exclusive association of presence records with forested land-use classes in the Sentinel-2 dataset (ESRI \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), coupled with the species' avoidance of urban, agricultural, or degraded areas.\u003c/p\u003e \u003cp\u003eThe integration of land-use data in Combination 3 markedly elevated predictive performance, underscoring the pivotal role of anthropogenic factors in shaping \u003cem\u003eB. hyrcana\u003c/em\u003e distribution. Variable importance analysis positioned land use as the top predictor, surpassing even core climatic variables like Bio3 and Bio12, likely because the majority of occurrence records were associated with forested land-use categories. This dominance reflects the species' strict dependence on intact, undisturbed forest habitats, where shade-tolerant, evergreen boxwood thrives in understory conditions but is highly vulnerable to deforestation. In the Hyrcanian context, where historical land-use changes have reduced forest cover by up to 40% since the mid-20th century, the land-use layer effectively captured human-induced constraints, refining model outputs to exclude non-forested areas and highlighting how anthropogenic pressures amplify climate sensitivities.\u003c/p\u003e \u003cp\u003eThe prominence of land use has profound implications for conservation and restoration strategies in the Hyrcanian forests. The refined habitat suitability maps from Combination 3 precisely delineated high-potential zones in Mazandaran Province while identifying restoration opportunities in Golestan, where suitable climatic and topographic conditions exist but are undermined by current agricultural dominance. This suggests that without addressing land-use barriers, projected suitable habitats may remain inaccessible, exacerbating \u003cem\u003eB. hyrcana's\u003c/em\u003e decline, which has already seen population losses of over 50% in some areas due to illegal harvesting and fragmentation (Sagheb Talebi et al. \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2014\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eComparing results with (Habibi Kilak et al. \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) on Taxus baccata shows both studies emphasize the essential role of bioclimatic and topographic variables in determining shade-loving tree species distribution. Habibi Kilak et al. identified Bio2, Bio3, Bio7, and Bio8 as key climatic variables, with altitude and slope as important topographic factors. These findings share significant similarities with our XGBoost model results in Combinations 1 and 2, especially emphasizing Bio12, Bio3, and LS Factor. However, differences exist; for example, in Habibi Kilak's study, Bio2 and Bio7 played greater roles, while in the present study, these were not primary. This may arise from physiological differences: yew is more sensitive to daily temperature fluctuations, whereas Hyrcanian boxwood responds more to annual temperature uniformity.\u003c/p\u003e \u003cp\u003e(Z. A. Wani et al. \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) modeled Buxus wallichiana habitats in the Himalayas using MaxEnt, identifying only 0.4% suitable area, with annual mean temperature, driest-month precipitation, and elevation as top variables. Similarities with our Combination 1 include isothermality (Bio3) as critical for Buxus viability, but differences emerge: our model prioritized annual precipitation (Bio12) over driest-month metrics, and LS Factor over elevation. These reflect ecological distinctions\u0026mdash;\u003cem\u003eB. hyrcana\u003c/em\u003e in humid Caspian forests versus B. wallichiana in drier Himalayas\u0026mdash;highlighting the need for species- and region-specific modeling.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003eResponse Curve Analysis\u003c/h2\u003e \u003cp\u003eIn Combination 1, variables Bio3, LS Factor, Bio15, and Bio12 showed the greatest impact. The Bio3 response curve trended upward to ~\u0026thinsp;45 units, stabilizing at high presence probability, indicating positive thermal adaptation (Becklin et al. \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Z. Wani et al. \u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). LS Factor decreased sharply with steeper slopes, peaking again at ~\u0026thinsp;30, confirming sensitivity to erosion-prone terrains. Bio15 favored 40\u0026ndash;60 mm seasonal rainfall, remaining stable above 60 mm but dropping below, showing aversion to precipitation deficits. Bio12 peaked below 500 mm annual rainfall, declining with excess, aligning with west-east moisture gradients in Hyrcanian forests (Moghbel Esfahani et al. \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Overall, \u003cem\u003eB. hyrcana\u003c/em\u003e prefers high isothermality, gentle slopes, moderate temperatures, and balanced precipitation.\u003c/p\u003e \u003cp\u003eIn Combination 2, Bio4, Bio8, ChND, and LS Factor dominated. Bio4 rose rapidly from 700\u0026ndash;800 units, favoring seasonal temperature range stability. Bio8 peaked at 0\u0026ndash;8\u0026deg;C in wet seasons, dropping above 10\u0026deg;C, matching evergreen optima (Moore et al. \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). ChND showed complex patterns, increasing 200\u0026ndash;300 m from streams for moisture without flooding. LS Factor again favored low values. Results indicate suitability in high seasonal range, moderate wet-season temperatures, gentle slopes, and water access. In Combination 3, Bio12, Bio1, Bio3, and land use were key. Bio12 declined above 450\u0026ndash;500 mm, preferring drier conditions. Bio3 rose from ~\u0026thinsp;29 units. Bio1 stabilized at 0\u0026ndash;14\u0026deg;C, dropping above 15\u0026deg;C, signaling thermal limits and warming vulnerability. Across combinations, patterns converged on moderate climates and topography, with data variations altering thresholds, emphasizing careful variable selection for conservation.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003eHabitat Suitability Map Analysis\u003c/h2\u003e \u003cp\u003eThe habitat suitability maps across the three data combinations revealed consistent spatial patterns, with Mazandaran Province emerging as the primary stronghold for \u003cem\u003eBuxus hyrcana\u003c/em\u003e. Combination 1 (WorldClim\u0026thinsp;+\u0026thinsp;topography) concentrated high-density suitable areas in mid-mountain forests of Mazandaran, particularly from Kelardasht to Neka, while Gilan exhibited fragmented suitability restricted to specific elevations. Combination 2 produced broadly similar results, again highlighting central Mazandaran as the core distribution zone. In Combination 3, the model delineated highly suitable areas with greater precision, closely matching the actual distribution of Hyrcanian forests. Key hotspots included forests south of Chalus, Nur, Sari, and Behshahr. Importantly, the inclusion of land-use data enhanced spatial realism by excluding unsuitable areas outside forested zones.\u003c/p\u003e \u003cp\u003eMazandaran Province, particularly its central and eastern sections, consistently exhibited the highest habitat suitability, aligning with current stand density and natural distribution patterns (Alipour and Walas \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Habibi kilak et al. \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). The ecological conditions of this region\u0026mdash;high humidity, maritime climate, moderate temperatures, and thermal stability\u0026mdash;are fully compatible with the ecological requirements of \u003cem\u003eB. hyrcana\u003c/em\u003e as a shade-tolerant, moisture-dependent species (Sagheb Talebi et al. \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). In Gilan Province, suitable habitats were also identified, particularly in western low-elevation humid areas, confirming the model's ability to capture species-environment relationships.\u003c/p\u003e \u003cp\u003eA particularly noteworthy finding was the prediction of moderate to high suitability areas in Golestan Province, where no natural stands of boxwood have been reported. This suggests significant restoration potential in eastern Hyrcanian forests, provided that biological, ecological, and management conditions are favorable. Such results are consistent with previous studies advocating the identification of potential habitats for restoration in regions currently lacking populations (Becklin et al. \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Esmailzadeh and Soleymanipour \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). Taken together, the comparative evaluation indicates that conservation priorities should focus on Mazandaran and western Gilan, while Golestan should be considered a priority region for future restoration initiatives.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusions","content":"\u003cp\u003eThis study demonstrated that the XGBoost algorithm is a highly effective and accurate tool for predicting the distribution of \u003cem\u003eBuxus hyrcana\u003c/em\u003e in the Hyrcanian forests of northern Iran. Across three different data combinations, the model consistently achieved strong performance, with high AUC (\u0026gt;\u0026thinsp;0.97), Kappa, and TSS values confirming its ability to discriminate between suitable and unsuitable habitats. Notably, Combination 3, which integrated WorldClim bioclimatic variables, topographic predictors, and land-use data, provided the highest accuracy and ecological realism (TSS\u0026thinsp;=\u0026thinsp;0.86), underscoring the value of combining multiple environmental datasets\u0026mdash;especially anthropogenic factors\u0026mdash;in species distribution modeling to capture real-world threats.\u003c/p\u003e \u003cp\u003eThe analysis of variable importance revealed a stable set of key predictors\u0026mdash;Bio3 (isothermality), Bio15 (precipitation seasonality), Bio8 (mean temperature of the wettest quarter), Bio12 (annual precipitation), and LS Factor\u0026mdash;highlighting the central role of thermal stability, moderate temperatures, precipitation balance, and topographic constraints in shaping boxwood distribution, with land use emerging as the overriding influence in human-impacted scenarios. The persistence of these variables across combinations indicates that \u003cem\u003eB. hyrcana\u003c/em\u003e is highly sensitive to deviations from moderate climatic conditions and physiographic stability, amplified by land-use pressures.\u003c/p\u003e \u003cp\u003eResponse curve analysis clarified the species' ecological niche, showing that \u003cem\u003eB. hyrcana\u003c/em\u003e thrives under moderate temperature and precipitation regimes, while extreme conditions sharply reduce occurrence probability. The dual role of LS Factor in Combination 3 suggested that the species can persist in both valley bottoms and certain stable steep slopes, provided other environmental conditions and intact land use are favorable.\u003c/p\u003e \u003cp\u003eHabitat suitability maps confirmed Mazandaran Province as the most suitable and continuous habitat, followed by more fragmented areas in Gilan. The identification of potential habitats in Golestan Province, despite the absence of current populations, highlights opportunities for restoration and reintroduction programs in eastern Hyrcanian forests, contingent on land-use restoration.\u003c/p\u003e \u003cp\u003eOverall, this research demonstrates that advanced machine learning approaches such as XGBoost, when combined with comprehensive climatic, topographic, and land-use data, provide powerful tools for identifying and prioritizing suitable habitats for endangered species. The findings not only enhance ecological understanding of \u003cem\u003eB. hyrcana\u003c/em\u003e but also provide a robust scientific foundation for conservation planning, sustainable management, and restoration strategies in the Hyrcanian forests. Integrating accurate environmental data with cutting-edge modeling techniques is therefore essential for safeguarding endemic species and promoting biodiversity conservation in these globally significant ecosystems.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eConflicts of interest/Competing interests\u0026nbsp;\u003c/strong\u003ethe authors have no competing interests to declare that are relevant to the content of this article.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eOpen Access\u0026nbsp;\u003c/strong\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics Approval\u003cbr\u003e\u003c/strong\u003eAll authors have given their full consent regarding ethics approval.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent to Participate\u003cbr\u003e\u003c/strong\u003eAll authors have given their full consent regarding participation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for Publication\u003cbr\u003e\u003c/strong\u003eAll authors have given their full consent for publication.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e No funding was received to assist with the preparation of this manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e Not accessible.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCode availability\u003c/strong\u003e the current study code is available upon reasonable request from the corresponding author.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u0026nbsp;\u003c/strong\u003eAll authors contributed to the study conception and design. O.E conducted field sampling. A.H performed preliminary analysis and wrote the first draft of the manuscript in Persian and O.E assisted in editing the draft. S.J.A Provided the initial idea for the study, conducted the data analysis, translated the text into English, and polished the manuscript. All authors read and approved the final manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eAhmadi‐sani N, Razaghnia L, Pukkala T (2022) Effect of Land‐Use Change on Runoff in Hyrcania. Land (Basel) 11:. https://doi.org/10.3390/land11020220\u003c/li\u003e\n \u003cli\u003eAkbarzadeh A, Ghorbani-Dashtaki S, Naderi-Khorasgani M, et al (2016) Monitoring and assessment of soil erosion at micro-scale and macro-scale in forests affected by fire damage in northern Iran. Environ Monit Assess 188:. https://doi.org/10.1007/s10661-016-5712-6\u003c/li\u003e\n \u003cli\u003eAkrim F, Mahmood T, Hussain R, et al (2017) Distribution pattern, population estimation and threats to the Indian Pangolin Manis crassicaudata (Mammalia: Pholidota: Manidae) in and around Pir Lasura National Park, Azad Jammu \u0026amp; Kashmir, Pakistan. J Threat Taxa 9:. https://doi.org/10.11609/jott.2914.9.3.9920-9927\u003c/li\u003e\n \u003cli\u003eAlipour S, Walas Ł (2023) The influence of climate and population density on Buxus hyrcana potential distribution and habitat connectivity. J Plant Res 136:. https://doi.org/10.1007/s10265-023-01457-5\u003c/li\u003e\n \u003cli\u003eAmiri M, Tarkesh M, Jafari R, Jetschke G (2020) Bioclimatic variables from precipitation and temperature records vs. remote sensing-based bioclimatic variables: Which side can perform better in species distribution modeling? Ecol Inform 57:. https://doi.org/10.1016/j.ecoinf.2020.101060\u003c/li\u003e\n \u003cli\u003eAsadi H, Jalilvand H, Tafazoli M, Hosseini SF (2025) Modeling habitat suitability of Quercus castaneifolia in the Hyrcanian forest: a comprehensive integration of environmental factors for conservation insights. Biodivers Conserv 34:315\u0026ndash;334. https://doi.org/10.1007/s10531-024-02973-z\u003c/li\u003e\n \u003cli\u003eAsefa M, Cao M, He Y, et al (2020) Ethiopian vegetation types, climate and topography. Plant Divers 42:. https://doi.org/10.1016/j.pld.2020.04.004\u003c/li\u003e\n \u003cli\u003eAustin MP (2002) Spatial prediction of species distribution: An interface between ecological theory and statistical modelling. Ecol Modell 157:. https://doi.org/10.1016/S0304-3800(02)00205-3\u003c/li\u003e\n \u003cli\u003eBaccini A, Goetz SJ, Walker WS, et al (2012) Estimated carbon dioxide emissions from tropical deforestation improved by carbon-density maps. Nat Clim Chang 2:. https://doi.org/10.1038/nclimate1354\u003c/li\u003e\n \u003cli\u003eBaker ME, Barnes B V. (1998) Landscape ecosystem diversity of river floodplains in northwestern Lower Michigan, U.S.A. Canadian Journal of Forest Research 28:. https://doi.org/10.1139/x98-107\u003c/li\u003e\n \u003cli\u003eBarbet-Massin M, Jiguet F, Albert CH, Thuiller W (2012) Selecting pseudo-absences for species distribution models: How, where and how many? Methods Ecol Evol 3:. https://doi.org/10.1111/j.2041-210X.2011.00172.x\u003c/li\u003e\n \u003cli\u003eBarlow J, Lennox GD, Ferreira J, et al (2016) Anthropogenic disturbance in tropical forests can double biodiversity loss from deforestation. Nature 535:. https://doi.org/10.1038/nature18326\u003c/li\u003e\n \u003cli\u003eBecklin KM, Anderson JT, Gerhart LM, et al (2016) Examining plant physiological responses to climate change through an evolutionary lens. Plant Physiol 172:. https://doi.org/10.1104/pp.16.00793\u003c/li\u003e\n \u003cli\u003eBiru Y, Tessema ZK, Urge M (2017) Perception and attitude of pastoralists on livestock-wildlife interactions around Awash National Park, Ethiopia: implication for biodiversity conservation. Ecol Process 6:. https://doi.org/10.1186/s13717-017-0081-9\u003c/li\u003e\n \u003cli\u003eCampbell CE, Jones DN, Awasthy M, et al (2024) Which birds have the most to lose? an analysis of bird species\u0026rsquo; feeding habitat in changing Australian landscapes. Biodivers Conserv 33:2867\u0026ndash;2883. https://doi.org/10.1007/s10531-024-02890-1\u003c/li\u003e\n \u003cli\u003eChen T, Guestrin C (2016) XGBoost: A scalable tree boosting system. In: Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining\u003c/li\u003e\n \u003cli\u003eDolos K, Bauer A, Albrecht S (2015) Site suitability for tree species: Is there a positive relation between a tree species\u0026rsquo; occurrence and its growth? Eur J For Res 134:. https://doi.org/10.1007/s10342-015-0876-0\u003c/li\u003e\n \u003cli\u003eElith J, H. Graham C, P. Anderson R, et al (2006) Novel methods improve prediction of species\u0026rsquo; distributions from occurrence data. Ecography 29:. https://doi.org/10.1111/j.2006.0906-7590.04596.x\u003c/li\u003e\n \u003cli\u003eElith J, Leathwick JR, Hastie T (2008) A working guide to boosted regression trees. Journal of Animal Ecology 77\u003c/li\u003e\n \u003cli\u003eEllis EC, Gauthier N, Goldewijk KK, et al (2021) People have shaped most of terrestrial nature for at least 12,000 years. Proc Natl Acad Sci U S A 118:. https://doi.org/10.1073/pnas.2023483118\u003c/li\u003e\n \u003cli\u003eEsmailzadeh O, Soleymanipour S (2015) Habitat suitability and ecological requirements of Buxus hyrcana in the Hyrcanian forests. Iranian Journal of Forest and Poplar Research 23:45\u0026ndash;58\u003c/li\u003e\n \u003cli\u003eESRI (2024) Sentinel-2 10 m Land Use/Land Cover Time Series (2017\u0026ndash;2024) [Imagery layer]. In: ArcGIS Living Atlas of the World\u003c/li\u003e\n \u003cli\u003eFan J, Yue W, Wu L, et al (2018) Evaluation of SVM, ELM and four tree-based ensemble models for predicting daily reference evapotranspiration using limited meteorological data in different climates of China. Agric For Meteorol 263:. https://doi.org/10.1016/j.agrformet.2018.08.019\u003c/li\u003e\n \u003cli\u003eFick SE, Hijmans RJ (2017) WorldClim 2: new 1-km spatial resolution climate surfaces for global land areas. International Journal of Climatology 37:. https://doi.org/10.1002/joc.5086\u003c/li\u003e\n \u003cli\u003eFielding AH, Bell JF (1997) A review of methods for the assessment of prediction errors in conservation presence/absence models. Environ Conserv 24:38\u0026ndash;49. https://doi.org/10.1017/S0376892997000088\u003c/li\u003e\n \u003cli\u003eFranklin J (2013) Species distribution models in conservation biogeography: Developments and challenges. Divers Distrib 19\u003c/li\u003e\n \u003cli\u003eGama M, Crespo D, Dolbeth M, Anast\u0026aacute;cio P (2016) Predicting global habitat suitability for Corbicula fluminea using species distribution models: The importance of different environmental datasets. Ecol Modell 319:. https://doi.org/10.1016/j.ecolmodel.2015.06.001\u003c/li\u003e\n \u003cli\u003eGrantham HS, Duncan A, Evans TD, et al (2020) Anthropogenic modification of forests means only 40% of remaining forests have high ecosystem integrity. Nat Commun 11:. https://doi.org/10.1038/s41467-020-19493-3\u003c/li\u003e\n \u003cli\u003eGuisan A, Edwards TC, Hastie T (2002) Generalized linear and generalized additive models in studies of species distributions: Setting the scene. Ecol Modell 157:. https://doi.org/10.1016/S0304-3800(02)00204-1\u003c/li\u003e\n \u003cli\u003eGuisan A, Thuiller W (2005) Predicting species distribution: Offering more than simple habitat models. Ecol Lett 8\u003c/li\u003e\n \u003cli\u003eHabibi kilak S, Alavi SJ, Esmailzadeh O (2019) Analyzing the ecological niche of Buxus hyrcana Pojark in the northern forests of Iran. Forest and Wood Products 72:21\u0026ndash;31\u003c/li\u003e\n \u003cli\u003eHabibi Kilak S, Alavi SJ, Esmailzadeh O (2025) Spatial resolution matters: unveiling the role of environmental predictors in English yew (Taxus bacata L.) distribution using MaxEnt modeling. Earth Sci Inform 18:214. https://doi.org/10.1007/s12145-024-01646-3\u003c/li\u003e\n \u003cli\u003eHastie T, Tibshirani R, James G, Witten D (2021) An introduction to Statistical Learning with Applications in R (2nd Edition). Springer texts 102:\u003c/li\u003e\n \u003cli\u003eHesabi A, Alavi SJ, Esmailzadeh O (2025) Evaluation of the Accuracy of Climatic Data from the Worldclim and Chelsa Databases in Three Northern Provinces of Iran. Forest Research and Development. https://doi.org/10.30466/jfrd.2025.55648.1743\u003c/li\u003e\n \u003cli\u003eHijmans R (2025) terra: Spatial Data Analysis\u003c/li\u003e\n \u003cli\u003eHotta M, Tsuyama I, Nakao K, et al (2019) Modeling future wildlife habitat suitability: Serious climate change impacts on the potential distribution of the Rock Ptarmigan Lagopus muta japonica in Japan\u0026rsquo;s northern Alps. BMC Ecol 19:. https://doi.org/10.1186/s12898-019-0238-8\u003c/li\u003e\n \u003cli\u003eHou Y, Wei X, Zhang M, et al (2023) A global synthesis of hydrological sensitivities to deforestation and forestation. For Ecol Manage 529\u003c/li\u003e\n \u003cli\u003eHu Y, Ding Z, Jiang Z, et al (2018) Birds in the Himalayas: What drives beta diversity patterns along an elevational gradient? Ecol Evol 8:. https://doi.org/10.1002/ece3.4622\u003c/li\u003e\n \u003cli\u003eKarger DN, Conrad O, B\u0026ouml;hner J, et al (2017) Climatologies at high resolution for the earth\u0026rsquo;s land surface areas. Sci Data 4:. https://doi.org/10.1038/sdata.2017.122\u003c/li\u003e\n \u003cli\u003eKidane YO, Steinbauer MJ, Beierkuhnlein C (2019) Dead end for endemic plant species? A biodiversity hotspot under pressure. Glob Ecol Conserv 19:. https://doi.org/10.1016/j.gecco.2019.e00670\u003c/li\u003e\n \u003cli\u003eKuhn M, Wickham H (2020) Tidymodels: a collection of packages for modeling and machine learning using tidyverse principles\u003c/li\u003e\n \u003cli\u003eKutner MH, Nachtsheim CJ, Neter J, Li W (2005) Chapter 11: Building the Regression Model III: Remedial Measures. In: Applied Linear Statistical Models\u003c/li\u003e\n \u003cli\u003eLi X, Wang Y (2013) Applying various algorithms for species distribution modelling. Integr Zool 8\u003c/li\u003e\n \u003cli\u003eLiu L, Liao J, Wu Y, Zhang Y (2020) Breeding range shift of the red-crowned crane (Grus japonensis) under climate change. PLoS One 15:. https://doi.org/10.1371/journal.pone.0229984\u003c/li\u003e\n \u003cli\u003eMart\u0026iacute;nez-Meyer E, Peterson AT, Serv\u0026iacute;n JI, Kiff LF (2006) Ecological niche modelling and prioritizing areas for species reintroductions. ORYX 40:. https://doi.org/10.1017/S0030605306001360\u003c/li\u003e\n \u003cli\u003eMoghbel Esfahani F, Alavi SJ, Hosseini SM, Tabari Kochaksarai M (2023) Determining the habitat suitability of Quercus castaneifolia C. A. Mey In order to plan restoration using species distribution modeling. Forest Research and Development 9:419\u0026ndash;436. https://doi.org/10.30466/jfrd.2023.54577.1654\u003c/li\u003e\n \u003cli\u003eMohan R, Maurya A, Nair NK, et al (2024) Multi-Model Analysis to Predict the Potential Suitable Habitat for Quercus oblongata D. Don (Fagaceae) in the Western Himalayan Region. Environment and Ecology 42:687\u0026ndash;696. https://doi.org/10.60151/envec/zoyl8560\u003c/li\u003e\n \u003cli\u003eMoore CE, Meacham-Hensold K, Lemonnier P, et al (2021) The effect of increasing temperature on crop photosynthesis: From enzymes to ecosystems. J Exp Bot 72:. https://doi.org/10.1093/jxb/erab090\u003c/li\u003e\n \u003cli\u003eMoradi S, Sheykhi Ilanloo S, Kafash A, Yousefi M (2019) Identifying high-priority conservation areas for avian biodiversity using species distribution modeling. Ecol Indic 97:. https://doi.org/10.1016/j.ecolind.2018.10.003\u003c/li\u003e\n \u003cli\u003eMu\u0026ntilde;oz-Mas R, Gil-Mart\u0026iacute;nez E, Oliva-Paterna FJ, et al (2019) Tree-based ensembles unveil the microhabitat suitability for the invasive bleak (Alburnus alburnus L.) and pumpkinseed (Lepomis gibbosus L.): Introducing XGBoost to eco-informatics. Ecol Inform 53:. https://doi.org/10.1016/j.ecoinf.2019.100974\u003c/li\u003e\n \u003cli\u003eNaimi B, Hamm NAS, Groen TA, et al (2014) Where is positional uncertainty a problem for species distribution modelling? Ecography 37:. https://doi.org/10.1111/j.1600-0587.2013.00205.x\u003c/li\u003e\n \u003cli\u003eOlden JD, Lawler JJ, Poff NL (2008) Machine learning methods without tears: A primer for ecologists. Quarterly Review of Biology 83\u003c/li\u003e\n \u003cli\u003ePebesma E, Bivand R (2023) Spatial Data Science: With Applications in R\u003c/li\u003e\n \u003cli\u003eR Core Team (2024) A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. In: URL https://www.R-project.org/.\u003c/li\u003e\n \u003cli\u003eRam\u0026iacute;rez-Albores JE, Prieto-Torres DA, Gordillo-Mart\u0026iacute;nez A, et al (2021) Insights for protection of high species richness areas for the conservation of Mesoamerican endemic birds. Divers Distrib 27:. https://doi.org/10.1111/ddi.13153\u003c/li\u003e\n \u003cli\u003eRosa IMD, Smith MJ, Wearn OR, et al (2016) The Environmental Legacy of Modern Tropical Deforestation. Current Biology 26:. https://doi.org/10.1016/j.cub.2016.06.013\u003c/li\u003e\n \u003cli\u003eRushton SP, Ormerod SJ, Kerby G (2004) New paradigms for modelling species distributions? Journal of Applied Ecology 41\u003c/li\u003e\n \u003cli\u003eSagheb Talebi K, Sajedi T, Pourhashemi M (2014) Forests of Iran: A treasure from the past, a hope for the future. Springer, Dordrecht\u003c/li\u003e\n \u003cli\u003eSalehi Shanjani P, Javadi H, Rasoulzadeh L, Amirkhani M (2018) Evaluation of genetic differentiation among healthy and infected Buxus hyrcana with boxwood blight using RAPD and ISSR markers. N Z J For Sci 48:. https://doi.org/10.1186/s40490-018-0120-z\u003c/li\u003e\n \u003cli\u003eShahnaseri G, Malekian M, Pourmoghadam K (2023) Habitat loss of the chestnut-leaved oak (Quercus castaneifolia) in the Hyrcanian forests of Iran: impacts of anthropogenic factors on forest thinning and degradation. Glob Ecol Conserv 46:. https://doi.org/10.1016/j.gecco.2023.e02600\u003c/li\u003e\n \u003cli\u003eSheykhi Ilanloo S, Khani A, Kafash A, et al (2021) Applying opportunistic observations to model current and future suitability of the Kopet Dagh Mountains for a Near Threatened avian scavenger. Avian Biol Res 14:. https://doi.org/10.1177/1758155920962750\u003c/li\u003e\n \u003cli\u003eSillero N, Arenas-Castro S, Enriquez‐Urzelai U, et al (2021) Want to model a species niche? A step-by-step guideline on correlative ecological niche modelling. Ecol Modell 456:. https://doi.org/10.1016/j.ecolmodel.2021.109671\u003c/li\u003e\n \u003cli\u003eSillero N, Campos JC, Arenas-Castro S, Barbosa AM (2023) A curated list of R packages for ecological niche modelling. Ecol Modell 476\u003c/li\u003e\n \u003cli\u003eSohrabi H (2025) Does UNESCO designation enhance forest protection? Evidence from the Hyrcanian national forest inventory. Trees, Forests and People 21:100956. https://doi.org/https://doi.org/10.1016/j.tfp.2025.100956\u003c/li\u003e\n \u003cli\u003eSsali F, Mugerwa B, van Heist M, et al (2023) Plant diversity and composition vary with elevation on two equatorial high mountains in Uganda: baselines for assessing the influence of climate change. Alp Bot 133:. https://doi.org/10.1007/s00035-023-00301-9\u003c/li\u003e\n \u003cli\u003eVelazco SJE, Rose MB, de Andrade AFA, et al (2022) flexsdm: An r package for supporting a comprehensive and flexible species distribution modelling workflow. Methods Ecol Evol 13:. https://doi.org/10.1111/2041-210X.13874\u003c/li\u003e\n \u003cli\u003eVignali S, Barras AG, Arlettaz R, Braunisch V (2020) SDMtune: An R package to tune and evaluate species distribution models. Ecol Evol 10:. https://doi.org/10.1002/ece3.6786\u003c/li\u003e\n \u003cli\u003eWang SW, Boru BH, Njogu AW, et al (2020) Floristic composition and endemism pattern of vascular plants in Ethiopia and Eritrea. J Syst Evol 58:. https://doi.org/10.1111/jse.12527\u003c/li\u003e\n \u003cli\u003eWani Z, Khan S, Satish K, et al (2024) Ensemble modelling reveals shrinkage of suitable habitat for Himalayan Boxwood (Buxus wallichiana Bail.) under climate change \u0026ndash; implications for conservation. Phytocoenologia 52:. https://doi.org/10.1127/phyto/2024/0427\u003c/li\u003e\n \u003cli\u003eWani ZA, Satish K V., Islam T, et al (2023) Habitat suitability modelling of Buxus wallichiana Bail.: an endemic tree species of Himalaya. Vegetos 36:. https://doi.org/10.1007/s42535-022-00428-w\u003c/li\u003e\n \u003cli\u003eWinkler K, Fuchs R, Rounsevell M, Herold M (2021) Global land use changes are four times greater than previously estimated. Nat Commun 12:. https://doi.org/10.1038/s41467-021-22702-2\u003c/li\u003e\n \u003cli\u003eZhang M, Liu N, Harper R, et al (2017) A global review on hydrological responses to forest change across multiple spatial scales: Importance of scale, climate, forest type and hydrological regime. J Hydrol (Amst) 546\u003c/li\u003e\n \u003cli\u003eZhang M, Wei X (2021) Deforestation, forestation, and water supply. Science (1979) 371\u003c/li\u003e\n \u003cli\u003eZu K, Luo A, Shrestha N, et al (2019) Altitudinal biodiversity patterns of seed plants along Gongga Mountain in the southeastern Qinghai\u0026ndash;Tibetan Plateau. Ecol Evol 9:. https://doi.org/10.1002/ece3.5483\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"environmental-monitoring-and-assessment","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"emas","sideBox":"Learn more about [Environmental Monitoring and Assessment](http://link.springer.com/journal/10661)","snPcode":"10661","submissionUrl":"https://submission.nature.com/new-submission/10661/3","title":"Environmental Monitoring and Assessment","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Buxus hyrcana, Hyrcanian Forests, Species Distribution Modelling, Habitat Suitability, Conservation Planning","lastPublishedDoi":"10.21203/rs.3.rs-8858849/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8858849/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e \u003cem\u003eBuxus hyrcana\u003c/em\u003e, an endangered and ecologically significant tree species of the Hyrcanian forests, faces severe threats from climate change, land-use pressures, and habitat degradation. Accurate prediction of its potential distribution is therefore critical for conservation and restoration planning. In this study, we applied the eXtreme Gradient Boosting (XGBoost) algorithm to model the distribution of \u003cem\u003eB. hyrcana\u003c/em\u003e under three data combinations: (i) WorldClim bioclimatic and topographic variables, (ii) CHELSA bioclimatic and topographic variables, and (iii) WorldClim bioclimatic, topographic, and land-use variables. Model performance was evaluated using AUC, TSS, Kappa, and Accuracy metrics, all of which indicated strong predictive capacity, with the highest performance achieved when land-use data were incorporated. Variable importance analysis revealed a stable set of key predictors\u0026mdash;thermal stability (Bio3), annual mean temperature (Bio1), mean temperature of the wettest quarter (Bio8), annual precipitation (Bio12), and slope-related indices (LS Factor)\u0026mdash;highlighting the species\u0026rsquo; sensitivity to moderate climatic regimes and physiographic constraints. Response curve analysis confirmed that \u003cem\u003eB. hyrcana\u003c/em\u003e thrives under moderate temperature and precipitation conditions, while extreme climatic values sharply reduce occurrence probability. Habitat suitability maps consistently identified Mazandaran Province as the most suitable region, with additional restoration potential in Golestan Province. Our findings demonstrate that integrating high-resolution climatic, topographic, and land-use data within advanced machine learning frameworks significantly enhances the accuracy and ecological realism of species distribution models. This study provides a robust methodological framework for predicting the distribution of climate-sensitive, endangered species and offers actionable insights for conservation prioritization and restoration planning in the Hyrcanian forests.\u003c/p\u003e","manuscriptTitle":"Machine Learning Meets Ecology: XGBoost‑Based Prediction of Endangered Species Refugia Using Multi‑Source Environmental Data","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-27 15:34:58","doi":"10.21203/rs.3.rs-8858849/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewerAgreed","content":"248057578355035295161481316032287796250","date":"2026-05-14T14:31:33+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"85289903414110334656178542236294394229","date":"2026-05-14T06:55:09+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"339935587020669963049825998811663157857","date":"2026-05-12T20:22:14+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"242263944176069942581061152376619205744","date":"2026-04-15T12:29:49+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"123660069543191493394511823780806247590","date":"2026-03-25T11:56:25+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-03-25T09:29:07+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-02-24T01:55:45+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-02-24T01:54:02+00:00","index":"","fulltext":""},{"type":"submitted","content":"Environmental Monitoring and Assessment","date":"2026-02-12T07:08:31+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"environmental-monitoring-and-assessment","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"emas","sideBox":"Learn more about [Environmental Monitoring and Assessment](http://link.springer.com/journal/10661)","snPcode":"10661","submissionUrl":"https://submission.nature.com/new-submission/10661/3","title":"Environmental Monitoring and Assessment","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"6e2049fc-9a83-4f0a-af9e-f932514031c4","owner":[],"postedDate":"March 27th, 2026","published":true,"recentEditorialEvents":[{"type":"reviewerAgreed","content":"248057578355035295161481316032287796250","date":"2026-05-14T14:31:33+00:00","index":37,"fulltext":""},{"type":"reviewerAgreed","content":"85289903414110334656178542236294394229","date":"2026-05-14T06:55:09+00:00","index":35,"fulltext":""},{"type":"reviewerAgreed","content":"339935587020669963049825998811663157857","date":"2026-05-12T20:22:14+00:00","index":34,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-03-27T15:34:58+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-27 15:34:58","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8858849","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8858849","identity":"rs-8858849","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}