Data: Climate change threatens the persistence of forest lizards in the Brazilian Cerrado

Description of the data and file structure From ten forest-habitat specialist species endemic to the Cerrado initially identified in the literature, only these five had sufficient occurrence records available in scientific collections to meet the minimum criteria for modeling, with at least 20 distinct occurrence localities. We adopted the threshold of 20 records because the predictive performance of species distribution models declines substantially below this value (Sampaio and Cavalcante 2023; Shim et al. 2023; Wisz et al. 2008). Occurrence records span the period from 1921 to 2022 and were obtained through Sistema de Avaliação do Risco de Extinção da Biodiversidade – SALVE (https://salve.icmbio.gov.br). We removed duplicate records using the dismo package (Hijmans, Phillips, Leathwick, & Elith, 2017), and maintained all coordinates in the WGS84 reference system. Environmental predictors We characterized current (1970-2000) and future (2021-2040, 2041-2060, 2061-2080, and 2081-2100) climatic conditions using total monthly precipitation, monthly maximum and minimum temperatures, and the 19 bioclimatic variables from WorldClim 2.1 (Fick and Hijmans 2017; Hijmans et al. 2005; Hijmans et al. 2017) at a spatial resolution of 10 arc-minutes. Climatic layers were clipped to South America (10° N–40° S, 80° W–30° W). We selected a 10 arc-minute resolution as it captures the broad-scale climatic gradients driving species distributions at the continental extent, while maintaining computational feasibility. Furthermore, previous studies indicate that increasing spatial resolution does not necessarily improve the predictive accuracy of bioclimatic models when the goal is to characterize regional to continental ranges rather than local microhabitats (Connor et al. 2018; Guisan et al. 2007; Seo et al. 2009). For future climates, we used two distinct Shared Socioeconomic Pathway scenarios (SSP2–4.5 and SSP5–8.5), which reflect different greenhouse gas emission trajectories (Riahi et al. 2017). The SSP2–4.5 scenario is characterized by intermediate sustainability and moderate mitigation, indicating a pathway in which policies and strategies are implemented in a balanced manner. In contrast, the SSP5–8.5 scenario represents a trajectory with high inequality and limited mitigation policies, leading to increased greenhouse gas emissions and radiative forcing (Riahi et al. 2017). We employed the general circulation model (GCM) EC-Earth3-Veg, a variation of EC-Earth3, which was identified as one of the top-performing CMIP6 models for representing the South American climate (Almazroui et al. 2021; Bazzanela et al. 2024). While multi-model ensembles are often used to address uncertainty, the inclusion of models with significant regional biases can degrade the accuracy of projections, particularly in South America where observational stations are sparse. Therefore, we prioritized a model with demonstrated high skill in reproducing regional climate features over a potentially biased full-ensemble approach (Almazroui et al. 2021; Bazzanela et al. 2024). Furthermore, EC-Earth3-Veg includes vegetation dynamics by coupling the second-generation dynamic global vegetation model LPJ-GUESS. This coupling helps reduce uncertainties and biases in simulated circulations and enables the simulation of climatic parameters using consistent physical processes and spatial approaches (Döscher et al. 2022). Climatic suitability modeling We estimated climatic suitability for each species using the Random Forest algorithm implemented in the biomod2 package (Thuiller et al. 2020). We selected Random Forest because it presents consistent predictive performance in species distribution models and high computational efficiency compared to other widely used algorithms, in addition to adequately handling complex interactions and nonlinear relationships between environmental predictors (Valavi et al. 2022). Prior to modeling, collinearity among the 19 bioclimatic variables was assessed using the variance inflation factor (VIF), employing the usdm package (Naimi 2015). Variables with VIF values greater than 10 were iteratively removed, resulting in the retention of nine relatively independent variables (BIO2, BIO3, BIO8, BIO9, BIO13, BIO14, BIO15, BIO18, and BIO19), reducing statistical redundancy and avoiding overestimation of correlated variable importance. For each species, 20 independent pseudo-absence datasets were generated, each containing the same number of points as the presence records, ensuring balanced prevalence in the Random Forest models (Barbet‐Massin et al. 2012; VanDerWal et al. 2009). Models were fitted with 2,000 decision trees, a number sufficient to stabilize internal algorithm estimates and reduce variation associated with sampling randomness. Calibration and validation were performed using k-fold cross-validation, with five partitions and ten repetitions (Buisson et al. 2010). Model performance was quantified using the True Skill Statistic (TSS), a metric widely recommended for being independent of prevalence and for combining sensitivity and specificity (Allouche et al. 2006). Only models with TSS ≥ 0.7 were retained for the final model. Selected models were combined through a weighted ensemble, in which each model contributed proportionally to its TSS value (Qiao et al. 2015). Continuous suitability predictions were converted into binary presence–absence maps based on the maximum sensitivity and specificity threshold (maxSSS) (Freeman and Moisen 2008; Liu et al. 2016). To estimate the spatial pattern of richness under current climatic conditions, the binary distribution maps of the five modeled species were summed to generate a richness map representing the number of modeled species predicted to occur in each grid cell. Models calibrated under current conditions were then projected onto the different future climatic scenarios and time periods, maintaining the same set of environmental predictors. Changes in climatically suitable area We quantified changes in species distribution between current and future climatic conditions using the BIOMOD_RangeSize function, which estimates areas of contraction (rear: lost, no longer suitable in the future), stability (stable: retained, currently suitable and remaining suitable), or expansion (lead: gained, currently unsuitable and becoming suitable). These metrics were calculated separately for each species, climate scenario, and time period. We conducted all analyses in the R environment (R Core Team 2020). Files and variables File: extinction_risk_calculation_BAS.R Description: This file contains the R script used to quantify changes in geographic range size and extinction risk proxies for forest-associated lizard species in the Brazilian Cerrado under current and future climate scenarios. The workflow integrates species distribution model (SDM) outputs generated in BIOMOD2 with spatial analyses performed using the terra and sf packages. Initially, the script defines the study area by assembling spatial polygons for South America, Brazil, Brazilian states, and the Cerrado biome. These spatial layers are used to standardize projections and to support map visualization. For each focal species (e.g., Coleodactylus brachystoma, Colobosaura modesta, Hoplocercus spinosus, Norops brasiliensis, and Tupinambis quadrilineatus), the script: Loads binary habitat suitability maps (TSS-based thresholds) for present and future climate scenarios. Iterates over future projections (different SSPs and time periods). Calculates changes in range size using the BIOMOD_RangeSize function, classifying pixels into loss, stability, unsuitable areas, and gain. Exports spatial outputs as maps (PDF) and tabular summaries (TXT and RDS files). Subsequently, the script consolidates all species-specific outputs into a single dataset, converts pixel-based metrics into area (km²), and calculates descriptive statistics (mean and standard deviation) by scenario and time period. Additional metrics include percentage range loss, gain, and net range change. The script also generates multiple visualizations, including: Percentage area loss and gain across scenarios and periods; Net distribution changes (percentage and area); Comparisons between current and future range sizes (with and without dispersal assumptions); Species richness maps derived from stacking binary distributions (present and future). Finally, richness maps are produced by summing binary suitability layers across species and reclassifying values into discrete richness categories (0–5 species), allowing visualization of biodiversity patterns and their projected shifts under climate change. Overall, this script provides a fully reproducible pipeline for assessing climate-driven changes in species distributions, quantifying range contraction/expansion, and generating spatial products suitable for conservation analyses. File: dados_rangesize_mean_sd_km2.csv Description: This dataset contains quantitative estimates of geographic range dynamics for forest-associated lizard species in the Brazilian Cerrado under current and future climate scenarios. The data were derived from species distribution models (SDMs) projected using BIOMOD2 and subsequently processed to quantify changes in suitable habitat area. Range change metrics were calculated by comparing present-day binary suitability maps with future projections across multiple climate scenarios (Shared Socioeconomic Pathways – SSPs) and time periods. Pixel-based outputs were converted to area (km²), assuming a spatial resolution of 18.5 km (342.25 km² per pixel). The dataset includes measures of habitat loss, gain, and stability, as well as total range size under different dispersal assumptions (no dispersal vs. full dispersal). Additionally, percentage-based metrics and derived variables summarizing net range change are provided. Summary statistics (mean and standard deviation) are also included for each scenario and period. This dataset can be used to assess extinction risk proxies, evaluate species vulnerability to climate change, and support conservation planning in the Cerrado biome. Variables Loss_km2:Area (km²) of currently suitable habitat projected to become unsuitable under future climate conditions. Stable0_km2:Area (km²) classified as unsuitable in both current and future conditions (i.e., consistently unsuitable habitat). Stable1_km2:Area (km²) of habitat that remains suitable in both present and future projections (i.e., stable suitable habitat). Gain_km2:Area (km²) projected to become newly suitable under future climate conditions (i.e., expansion of potential distribution). CurrentRangeSize_km2:Total area (km²) of suitable habitat under current climatic conditions. FutureRangeSize.NoDisp_km2:Projected total suitable area (km²) under future climate conditions assuming no dispersal (species can only persist in currently suitable areas that remain suitable). FutureRangeSize.FullDisp_km2:Projected total suitable area (km²) under future climate conditions assuming full dispersal (species can occupy all newly suitable areas). RangeChangeArea_km2:Net change in suitable area (km²), calculated from the percentage change relative to the current range size. PercLoss_pct:Percentage of current suitable habitat projected to be lost under future conditions. PercGain_pct:Percentage of area gained relative to the current distribution. SpeciesRangeChange_pct:Net percentage change in species distribution area, combining losses and gains. Especie:Species name (abbreviated format; e.g., C.brachystoma, H.spinosus). Cenario:Climate change scenario (e.g., SSPs such as ssp245, ssp585). Periodo:Future time period of projection (e.g., 2041–2060, 2081–2100). File: Biomod.zip Description: The file biomod.zip contains the full directory structure of BIOMOD2 outputs used to generate species distribution projections for Cerrado forest lizards. This folder includes model outputs for each focal species, organized into species-specific subdirectories: C.brachystoma C.modesta H.spinosus N.brasiliensis T.quadrilineatus richness_outputs (derived products) Each species folder contains the results of ensemble species distribution models calibrated and projected using BIOMOD2. These outputs include: Current projections (proj_*_ensemble_all):Binary suitability maps (TSS threshold) representing present-day potential distribution. Future projections (proj_*_ensemble_all_futuro_*):Binary maps for multiple climate scenarios (e.g., SSPs) and time periods, used to assess range shifts under climate change. Raster files (.tif):Spatial predictions of habitat suitability (binary presence/absence), which are used as input for range size calculations. Model evaluation and ensemble outputs:Intermediate and final results generated by BIOMOD2, including ensemble forecasts. The folder richness_outputs contains stacked raster outputs derived from individual species projections, used to calculate and map species richness (number of species per pixel) under present and future conditions. Overall, this compressed folder provides all spatial inputs required to: Reproduce range size change analyses; Calculate habitat loss, gain, and stability; Generate species richness maps; Support extinction risk assessments based on projected distribution changes. Code/software R version 4.5.0 (2025-04-11 ucrt) Platform: x86_64-w64-mingw32/x64 Running under: Windows 11 x64 (build 26200) Matrix products: default LAPACK version 3.12.1 locale: [1] LC_COLLATE=Portuguese_Brazil.utf8 LC_CTYPE=Portuguese_Brazil.utf8 LC_MONETARY=Portuguese_Brazil.utf8 [4] LC_NUMERIC=C LC_TIME=Portuguese_Brazil.utf8 time zone: America/Sao_Paulo tzcode source: internal attached base packages: [1] stats graphics grDevices utils datasets methods base other attached packages: [1] lubridate_1.9.4 forcats_1.0.0 tidyr_1.3.1 tibble_3.2.1 tidyverse_2.0.0 [6] units_0.8-7 readr_2.1.5 purrr_1.0.4 stringr_1.5.1 tidyterra_0.7.2 [11] ggspatial_1.1.9 sf_1.0-20 dplyr_1.1.4 viridisLite_0.4.2 ggplot2_3.5.2 [16] randomForest_4.7-1.2 gbm_2.2.2 mgcv_1.9-1 nlme_3.1-168 rpart_4.1.24 [21] nnet_7.3-20 biomod2_4.2-6-2 terra_1.8-42 loaded via a namespace (and not attached): [1] gtable_0.3.6 lattice_0.22-6 tzdb_0.5.0 vctrs_0.6.5 tools_4.5.0 [6] generics_0.1.3 proxy_0.4-27 PresenceAbsence_1.1.11 pkgconfig_2.0.3 Matrix_1.7-3 [11] KernSmooth_2.23-26 RColorBrewer_1.1-3 lifecycle_1.0.4 compiler_4.5.0 farver_2.1.2 [16] codetools_0.2-20 class_7.3-23 pillar_1.10.2 classInt_0.4-11 iterators_1.0.14 [21] abind_1.4-8 foreach_1.5.2 tidyselect_1.2.1 stringi_1.8.7 reshape2_1.4.4 [26] splines_4.5.0 rnaturalearth_1.0.1 grid_4.5.0 cli_3.6.4 magrittr_2.0.3 [31] survival_3.8-3 e1071_1.7-16 withr_3.0.2 scales_1.4.0 sp_2.2-0 [36] timechange_0.3.0 httr_1.4.7 hms_1.1.3 rlang_1.1.6 Rcpp_1.0.14 [41] glue_1.8.0 DBI_1.2.3 pROC_1.18.5 rstudioapi_0.17.1 reshape_0.8.9 [46] jsonlite_2.0.0 R6_2.6.1 plyr_1.8.9