Table S1. Geographic distribution of occurrence records of Paspalum guenoarum derived from herbarium specimens and germplasm collections.
| Scientific name | Longitude | Latitude |
|---|---|---|
| Paspalum guenoarum | -38.31667 | -6.83333 |
| Paspalum guenoarum | -49.25944 | -22.79111 |
| Paspalum guenoarum | -52.07583 | -25.93083 |
| Paspalum guenoarum | -47.13 | -22.25 |
| Paspalum guenoarum | -47.97 | -23.58 |
| Paspalum guenoarum | -49.91361 | -25.235 |
| Paspalum guenoarum | -47.16472 | -22.2425 |
| Paspalum guenoarum | -49.33 | -24.13 |
| Paspalum guenoarum | -41.06167 | -11.05 |
| Paspalum guenoarum | -47.55361 | -15.43528 |
| Paspalum guenoarum | -48 | -17 |
| Paspalum guenoarum | -48.65 | -16.6 |
| Paspalum guenoarum | -49.20528 | -24.26833 |
| Paspalum guenoarum | -41.36667 | -13 |
| Paspalum guenoarum | -47.84167 | -21.96194 |
| Paspalum guenoarum | -61.9379 | -16.4708 |
| Paspalum guenoarum | -51.90806 | -26.36917 |
| Paspalum guenoarum | -49.24215 | -22.81042 |
| Paspalum guenoarum | -47.11667 | -22.18333 |
| Paspalum guenoarum | -47.13333 | -22.25 |
| Paspalum guenoarum | -58.3566 | -18.4199 |
| Paspalum guenoarum | -47.17736 | -22.24882 |
| Paspalum guenoarum | -56.71667 | -20.03333 |
| Paspalum guenoarum | -49.8071 | -25.5527 |
| Paspalum guenoarum | -54.41972 | -21.56778 |
| Paspalum guenoarum | -54.84222 | -20.755 |
| Paspalum guenoarum | -41.77805 | -12.21875 |
| Paspalum guenoarum | -47.4658 | -14.104 |
| Paspalum guenoarum | -55.3608 | -27.17127 |
| Paspalum guenoarum | -62.13 | -16.03 |
| Paspalum guenoarum | -50.61761 | -27.28734 |
| Paspalum guenoarum | -44.13 | -17.57 |
| Paspalum guenoarum | -56.1 | -28.4 |
| Paspalum guenoarum | -47.142 | -22.242 |
| Paspalum guenoarum | -49.23694 | -22.79167 |
| Paspalum guenoarum | -55.13333 | -30.78333 |
| Paspalum guenoarum | -51.05442 | -28.39204 |
| Paspalum guenoarum | -62 | -16.41666 |
| Paspalum guenoarum | -47.00082 | -15.0003 |
| Paspalum guenoarum | -62.13333 | -16.03333 |
| Paspalum guenoarum | -56.05 | -22.65 |
| Paspalum guenoarum | -55.61666 | -27.5 |
| Paspalum guenoarum | -62.08333 | -16.13333 |
| Paspalum guenoarum | -55.51666 | -27.35 |
| Paspalum guenoarum | -54.16666 | -26.76666 |
| Paspalum guenoarum | -55.58333 | -27.28333 |
| Paspalum guenoarum | -56 | -28.45 |
| Paspalum guenoarum | -55.66666 | -27.8 |
| Paspalum guenoarum | -55.11666 | -26.96666 |
| Paspalum guenoarum | -55.83333 | -27.53333 |
| Paspalum guenoarum | -57 | -25.5 |
| Paspalum guenoarum | -55.51666 | -27.3 |
| Paspalum guenoarum | -56.1 | -28.18333 |
| Paspalum guenoarum | -62 | -16.33333 |
| Paspalum guenoarum | -62 | -16 |
| Paspalum guenoarum | -47.96667 | -23.58333 |
| Paspalum guenoarum | -47.11944 | -22.18833 |
| Paspalum guenoarum | -58.60167 | -18.71083 |
| Paspalum guenoarum | -51.68694 | -30.09694 |
| Paspalum guenoarum | -47.7225 | -14.11417 |
| Paspalum guenoarum | -46.95111 | -14.89722 |
| Paspalum guenoarum | -47.97278 | -15.73944 |
| Paspalum guenoarum | -50.73917 | -29.32083 |
| Paspalum guenoarum | -49.98583 | -25.25833 |
| Paspalum guenoarum | -50.15722 | -24.76583 |
| Paspalum guenoarum | -51.065 | -28.67889 |
| Paspalum guenoarum | -51.53639 | -26.65806 |
| Paspalum guenoarum | -52.27667 | -28.10194 |
| Paspalum guenoarum | -49.98917 | -28.20194 |
| Paspalum guenoarum | -56.38333 | -19.48333 |
| Paspalum guenoarum | -56.94278 | -23.02417 |
| Paspalum guenoarum | -37.63333 | -10.91667 |
| Paspalum guenoarum | -41.06167 | -11.58333 |
| Paspalum guenoarum | -56.7333 | -27.5667 |
| Paspalum guenoarum | -47.92667 | -15.73139 |
| Paspalum guenoarum | -41.72722 | -4.08417 |
| Paspalum guenoarum | -49.33333 | -24.13333 |
| Paspalum guenoarum | -49.2375 | -22.81667 |
| Paspalum guenoarum | -51.40722 | -20.36111 |
| Paspalum guenoarum | -52.26556 | -20.62417 |
| Paspalum guenoarum | -47.71583 | -15.89833 |
| Paspalum guenoarum | -51.5717 | -28.6306 |
| Paspalum guenoarum | -47.85722 | -16.01111 |
| Paspalum guenoarum | -47.1525 | -22.28944 |
| Paspalum guenoarum | -47.5 | -15.5 |
| Paspalum guenoarum | -52.90417 | -27.97889 |
| Paspalum guenoarum | -39.25 | -12.9 |
| Paspalum guenoarum | -53.58139 | -31.73306 |
| Paspalum guenoarum | -54.67 | -22 |
| Paspalum guenoarum | -50.33 | -18 |
| Paspalum guenoarum | -62 | -16.33 |
| Paspalum guenoarum | -56.14972 | -22.67611 |
| Paspalum guenoarum | -49.35444 | -24.45361 |
| Paspalum guenoarum | -47.82 | -22.25 |
| Paspalum guenoarum | -62 | -16.17 |
| Paspalum guenoarum | -50.75 | -28.13 |
| Paspalum guenoarum | -46.34167 | -8.54167 |
| Paspalum guenoarum | -56.04833 | -28.12556 |
| Paspalum guenoarum | -53.75833 | -23.375 |
| Paspalum guenoarum | -56.40167 | -27.61361 |
| Paspalum guenoarum | -58.04389 | -27.32111 |
| Paspalum guenoarum | -57.72194 | -27.75861 |
| Paspalum guenoarum | -58.98611 | -28.44944 |
| Paspalum guenoarum | -55.72944 | -28.17083 |
| Paspalum guenoarum | -58.46028 | -28.33389 |
| Paspalum guenoarum | -63.23 | -17.77778 |
| Paspalum guenoarum | -63.29139 | -20.68444 |
| Paspalum guenoarum | -54.34889 | -27.99861 |
| Paspalum guenoarum | -54.44444 | -28.39667 |
| Paspalum guenoarum | -52.47528 | -28.31444 |
| Paspalum guenoarum | -59.3925 | -20.23194 |
| Paspalum guenoarum | -56.20722 | -25.4775 |
| Paspalum guenoarum | -55.3175 | -24.97222 |
| Paspalum guenoarum | -55.68917 | -31.53111 |
| Paspalum guenoarum | -53.99833 | -25.91222 |
| Paspalum guenoarum | -54.9175 | -26.87806 |
| Paspalum guenoarum | -57.93889 | -31.02833 |
| Paspalum guenoarum | -51.87417 | -26.92944 |
| Paspalum guenoarum | -55.64333 | -28.18222 |
| Paspalum guenoarum | -50.3 | -28.33333 |
| Paspalum guenoarum | -38.32 | -6.83 |
| Paspalum guenoarum | -59.26555556 | -28.17722222 |
| Paspalum guenoarum | -54.35527778 | -32.35138889 |
| Paspalum guenoarum | -58.36888890 | -33.33638890 |
Table S2. ODMAP (Overview, Data, Model, Assessment, and Prediction) metadata describing the methodology used in this study.
| Section | Subsection | Element | Value |
|---|---|---|---|
| Overview | Authorship | Study title | Predicting the potential distribution of Paspalum guenoarum (Poaceae) using environmental variables and BIOMOD2 ensemble models |
| Overview | Authorship | Author names | Francisco M.Luna; Florencia Marcón, Diógenes Cecchin Silveira, Miguel Dall´Agnol and Carlos A. Acuña |
| Overview | Model objective | Model objective | Mapping and Explanation |
| Overview | Model objective | Target output | Potential distribution/Habitat Suitability |
| Overview | Focal taxon | Focal taxon | Paspalum guenoarum Arechav. (Poaceae) |
| Overview | Location | Location | Central and northern Argentina, Uruguay, south-central Brazil, and Bolivia |
| Overview | Scale of analysis | Spatial extent | -66.7544, -34.1917, -34.7662, -0.8531 (xmin, xmax, ymin, ymax). Coordinate reference system: WGS84. |
| Overview | Scale of analysis | Spatial resolution | 2.5 arc-min |
| Overview | Scale of analysis | Temporal extent | Present day |
| Overview | Scale of analysis | Temporal resolution | Not applicable |
| Overview | Scale of analysis | Boundary | Minimum convex polygon with a 250 km buffer |
| Overview | Biodiversity data | Observation type | Preserved specimens; germplasm accessions; online databases |
| Overview | Biodiversity data | Response data type | Point occurrence |
| Overview | Predictors | Predictor types | Bioclimatic; edaphic; topographic |
| Overview | Hypotheses | Hypotheses | Precipitation and temperature are the main drivers of the species’ distribution, with a preference for humid environments and marked seasonal temperature variation |
| Overview | Assumptions | Model assumptions | Species–environment equilibrium; inclusion of the most relevant environmental predictors; reduction of spatial autocorrelation through geographic filtering |
| Overview | Algorithms | Modelling techniques | Generalized boosting models (GBM), random forest (RF), maximum entropy (MAXNET), generalized linear models (GLM), and flexible discriminant analysis (FDA) |
| Overview | Algorithms | Model complexity | MaxEnt (MAXNET) models were calibrated by tuning feature classes and regularization multipliers using the ENMeval package, selecting a configuration with linear, quadratic, hinge, and product features (LQHP) and a regularization multiplier of 0.5 based on the lowest AICc. The remaining algorithms (GBM, RF, GLM, and FDA) were optimized using the “bigboss” strategy implemented in BIOMOD2. Models were trained under a random cross-validation framework with ten replicates, using 70% of the data for calibration and 30% for evaluation. Model performance was assessed using AUC, TSS, and Kappa, and only models with TSS > 0.75 and AUC > 0.9 were retained for ensemble construction. |
| Overview | Algorithms | Model averaging | We built two types of ensemble models, i.e., a bioclimatic model and a combined model including edaphic variables and elevation, using an ensemble forecasting approach (weighted mean of selected models) implemented in BIOMOD2. In both ensemble models, we combined predictions from five SDM algorithms (GLM, FDA, GBM, RF, and MAXNET) and retained only models with high predictive performance (TSS > 0.75 and AUC > 0.9). The final ensemble predictions were obtained by aggregating individual model outputs to improve robustness and reduce uncertainty. |
| Overview | Workflow | Model workflow | We compiled occurrence records from herbarium collections, global biodiversity databases (GBIF and SpeciesLink), and germplasm accessions. We reduced spatial sampling bias by applying a geographic rarefaction filter, retaining one record within a 30 km radius. Environmental predictors were selected based on ecological relevance and collinearity analyses (Pearson correlation and VIF). We fitted SDMs using GLM, FDA, GBM, RF, and MAXNET within the BIOMOD2 framework. Model performance was evaluated using repeated random cross-validation (70/30 split, 10 replicates) and metrics including AUC, TSS, and Kappa. Finally, we predicted the potential distribution of Paspalum guenoarum under current environmental conditions. |
| Overview | Software | Software | We performed all analyses in R (version 4.3.3), primarily using the BIOMOD2, ENMeval, raster, and related spatial analysis packages. |
| Overview | Software | Data availability | Occurrence data were obtained from public repositories (GBIF and SpeciesLink) and herbarium/genebank collections, while environmental variables are available from WorldClim and ISRIC databases. |
| Data | Biodiversity data | Taxon name | Paspalum guenoarum Arechav. (Poaceae) |
| Data | Biodiversity data | Taxonomic reference system | Zuloaga, F. O., & Morrone, O. (2005). Revisión de las especies de Paspalum para América del Sur austral (Argentina, Bolivia, sur del Brasil, Chile, Paraguay y Uruguay). Monographs in Systematic Botany from the Missouri Botanical Garden 102: 1-297. |
| Data | Biodiversity data | Ecological level | Species |
| Data | Biodiversity data | Data sources | We compiled occurrence records from herbarium collections (CTES, SI, MVJB), global biodiversity databases (GBIF and SpeciesLink), and germplasm databases (Genesys and BGCTES genebank). |
| Data | Biodiversity data | Sampling design | Occurrence data were compiled from multiple sources and subjected to spatial filtering to reduce sampling bias. |
| Data | Biodiversity data | Sampling size | A total of 518 occurrence records were compiled from GBIF, SpeciesLink, herbarium collections (CTES, SI, MVJB), and germplasm databases (Genesys, BGCTES). After quality control (removal of duplicates and spatial errors) and spatial thinning (30 km radius), 124 records were retained for modeling. |
| Data | Biodiversity data | Clipping | Occurrence records were restricted to the accessible area (M) defined by a minimum convex polygon with a 250 km buffer. |
| Data | Biodiversity data | Scaling | No |
| Data | Biodiversity data | Cleaning | To ensure the accuracy of occurrence records and reduce spatial clustering, records were carefully checked using Google Earth and duplicates or inconsistent points were removed. A spatial rarefaction filter was applied, retaining a single occurrence within a 30 km radius. |
| Data | Biodiversity data | Absence data | No true absence data; pseudo-absences were generated using the surface range envelope (SRE) strategy. |
| Data | Biodiversity data | Backgound data | Pseudo-absence data were generated using the surface range envelope (SRE) strategy, producing 2,000 pseudo-absences across the calibration area, with 10 replicates. |
| Data | Biodiversity data | Error and biases | Sampling bias was reduced by applying spatial filtering (rarefaction) of occurrence records. |
| Data | Data partitioning | Training data | Random cross-validation using 70% of the data for model calibration, repeated 10 times. |
| Data | Data partitioning | Validation data | The remaining 30% of the data were used for model evaluation. |
| Data | Data partitioning | Test data | |
| Data | Predictor variables | Predictor variables | We considered eight environmental predictors, including four bioclimatic variables (mean diurnal temperature range, annual precipitation, precipitation of the wettest month, and precipitation of the warmest quarter), three edaphic variables (soil pH, cation exchange capacity, and soil water content), and elevation. |
| Data | Predictor variables | Data sources | Bioclimatic variables were obtained from WorldClim version 2.1, edaphic variables from the ISRIC database, and elevation from WorldClim. |
| Data | Predictor variables | Spatial extent | The spatial extent corresponded to the accessible area (M) defined by a minimum convex polygon around occurrence records with a 250 km buffer. |
| Data | Predictor variables | Spatial resolution | 2.5 arcm-min. |
| Data | Predictor variables | Spatial extent | The spatial extent corresponded to the accessible area (M), defined by a minimum convex polygon around occurrence records with a 250 km buffer in southern South America. |
| Data | Predictor variables | Coordinate reference system | WGS84 (EPSG:4326) |
| Data | Predictor variables | Temporal extent | Current climatic conditions (WorldClim version 2.1 baseline period) |
| Data | Predictor variables | Data processing | Collinearity among predictors was assessed using Pearson’s correlation coefficient (|r| > 0.7), and further evaluated using the Variance Inflation Factor (VIF), retaining only variables with VIF < 10. |
| Data | Predictor variables | Error and biases | Potential biases may arise from differences in spatial resolution and accuracy among environmental datasets, particularly between bioclimatic and edaphic variables. |
| Data | Predictor variables | Dimension reduction | Based on collinearity analyses, we selected a subset of non-collinear variables using Pearson’s correlation (|r| > 0.7) and Variance Inflation Factor (VIF < 10). |
| Data | Transfer data | Data sources | Not applicable |
| Data | Transfer data | Spatial extent | Not applicable |
| Data | Transfer data | Spatial resolution | Not applicable |
| Data | Transfer data | Temporal extent | Not applicable |
| Data | Transfer data | Temporal resolution | |
| Data | Transfer data | Models and scenarios | Not applicable |
| Data | Transfer data | Data processing | Not applicable |
| Data | Transfer data | Quantification of Novelty | Not applicable |
| Model | Variable pre-selection | Variable pre-selection | Environmental variables were pre-selected based on ecological relevance and collinearity analyses. |
| Model | Multicollinearity | Multicollinearity | We quantified collinearity among predictors using Pearson’s correlation coefficient (|r| > 0.7) and excluded highly correlated variables. This was further evaluated using the Variance Inflation Factor (VIF), retaining variables with VIF < 10. |
| Model | Model settings | Model settings (fitting) | GLM, FDA, GBM, and RF models were calibrated using the BIOMOD2 platform with the “bigboss” optimization strategy. MAXNET models were calibrated using the ENMeval package, selecting feature classes (LQHP) and a regularization multiplier of 0.5 based on AICc. |
| Model | Model settings | Model settings (extrapolation) | TRUE |
| Model | Model estimates | Coefficients | |
| Model | Model estimates | Parameter uncertainty | |
| Model | Model estimates | Variable importance | Variable importance was estimated using a permutation procedure implemented in BIOMOD2. |
| Model | Model selection | model averaging - ensembles Model selection | Only models with high predictive performance (TSS > 0.75 and AUC > 0.9) were selected. |
| Model | Model selection | model averaging - ensembles Model averaging | Ensemble predictions were obtained by combining selected individual models to improve robustness. |
| Model | Model selection | model averaging - ensembles Model ensembles | Ensemble forecasting implemented in BIOMOD2. |
| Model | Analysis and Correction of non-independence | Spatial autocorrelation | Spatial autocorrelation was reduced by applying a geographic rarefaction filter (30 km). |
| Model | Analysis and Correction of non-independence | Temporal autocorrelation | |
| Model | Analysis and Correction of non-independence | Nested data | |
| Model | Threshold selection | Threshold selection | Continuous predictions were converted into binary maps using the threshold that maximized the TSS metric. |
| Assessment | Performance statistics | Performance on training data | Model performance was evaluated using AUC, TSS, and Kappa metrics. |
| Assessment | Performance statistics | Performance on validation data | AUC;TSS;Kappa |
| Assessment | Performance statistics | Performance on test data | |
| Assessment | Plausibility check | Response shapes | Response curves were analyzed to assess ecological plausibility of predictor–response relationships. |
| Assessment | Plausibility check | Expert judgement | Model outputs were interpreted based on published ecological knowledge of the species and its observed distribution patterns. |
| Prediction | Prediction output | Prediction unit | Habitat suitability; predicted presence/absence |
| Prediction | Prediction output | Post-processing | Continuous suitability maps were converted into binary maps using TSS-based thresholds. |
| Prediction | Uncertainty quantification | Algorithmic uncertainty | Reduced by using ensemble modeling across multiple algorithms. |
| Prediction | Uncertainty quantification | Input data uncertainty | Uncertainty may arise from differences in spatial resolution and quality among environmental datasets, especially for edaphic variables. |
| Prediction | Uncertainty quantification | Parameter uncertainty | Reduced through model calibration and repeated cross-validation. |
| Prediction | Uncertainty quantification | Scenario uncertainty | |
| Prediction | Uncertainty quantification | Novel environments |