Tecomella undulata under threat: The impact of climate change on the distribution of a valuable tree species using a machine learning model

PLOS One | https://doi.org/10.1371/journal.pone.0326609 July 9, 2025 6 / 18 Model construction We employed the Maximum Entropy (MaxEnt) algorithm [45] in R version 4.3.1, using the sdm package [46], to make pre- dictions of T. undulata distributions. The model was run with 10,000 random background points. MaxEnt generates robust and precise predictive and distribution models comparable in accuracy to ensemble methods [47]. This well-established modeling approach offers a flexible response and strong transferability capabilities. Such attributes render it an appropri- ate model for projecting potential species distributions under varied environmental scenarios, making it a valuable asset for conservation and ecology-related applications [26]. We performed ten runs of our algorithm and employed a four-fold cross-validation technique to evaluate its accuracy. The dataset was randomly partitioned into four parts, with three parts used for model training and the remaining part held out for evaluation. We utilized the area under the receiver operating characteristic curve (AUC) [48] and true skills statis- tic (TSS) [49] methods to measure model performance. The ROC curve is not influenced by the prevalence of species, and the AUC indicates the model's ability to correctly distinguish between presence and absence points. The AUC value ranges from 0 to 1, with 0.5 indicating a random model. AUC values above 0.7 are considered valid and acceptable, while values above 0.9 indicate excellent results [26]. TSS scores range from +1 to −1, where a score closer to 1 signi-

fies a more accurate model, while a score closer to zero or negative suggests a model performing no better than random chance [49]. We computed the relative variable importance metric by using Pearson correlation within the SDM pack- age. Furthermore, we projected the future distribution of species using the models and considered both the SSP245 and SSP585 scenarios for 2030, 2050, 2070, and 2090. In addition, we classified the present and future suitability maps into four categories: unsuitable (< 0.1), low (0.1–0.4), medium (0.4–0.7), and high (> 0.7) suitability. To evaluate the impact of climate change on species distribution, we conducted comparisons between each future species distribution map with the current distribution maps. Table 1. Bioclimatic variables used in the study. Variable Code Unit Annual mean temperature Bio1 °C Mean diurnal range Bio2 °C Isothermality (Bio2/Bio7) (×100) Bio3 Index Temperature seasonality Bio4 Index Max temperature of the warmest month Bio5 °C Min temperature of the coldest month Bio6 °C Temperature annual range Bio7 °C Mean temperature of the wettest quarter Bio8 °C Mean temperature of the driest quarter Bio9 °C Mean temperature of the warmest quarter Bio10 °C Mean temperature of the coldest quarter Bio11 °C Annual precipitation Bio12 mm Precipitation of the wettest month Bio13 mm Precipitation of the driest month Bio14 mm Precipitation seasonality Bio15 Index Precipitation of the wettest quarter Bio16 mm Precipitation of the driest quarter Bio17 mm Precipitation of the warmest quarter Bio18 mm

Precipitation of the coldest quarter Bio19 mm https://doi.org/10.1371/journal.pone.0326609.t001