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 11 / 18 that future climatic conditions could negatively impact on the distribution of T. undulata, emphasizing the need for proac- tive measures to conserve this species. Response curves of environmental variables The habitat suitability modeling of T. undulata, provided valuable insights through the analysis of response curves for several variables (Fig 6). The species showed preferences for certain environmental conditions, and these findings can inform conservation strategies and efforts. The variables analyzed included Bio4, Bio8, Bio15, Bio9, Bio19, Bio14, Bio2, Bio3, and DEM. The species shows a preference for temperature ranges between −2–27°C in the driest quarter (Bio9) with declining suitability beyond 27°C. A moderate range of seasonal temperature variation (Bio4) is optimal, with a peak preference at a value of 720, beyond which suitability declines, underscoring its sensitivity to wide temperature fluctuations. T. undulata also favors warmer conditions during the wettest quarter (Bio8), indicative of its tropical adaptation, and exhibits a strong preference for environments with high precipitation seasonality (Bio15). The species shows optimum desirability for habi- tats with minimal diurnal temperature variation (Bio2), peaking at a value of 10.5. Regarding precipitation during the driest (Bio14) and coldest quarters (Bio19), habitat suitability decreases with increased wetness beyond certain thresholds,

demonstrating a preference for drier conditions and a particular precipitation range during colder times. Elevational prefer- ences (DEM) peak at intermediate levels, suggesting adaptability within a specific altitude range. Lastly, there is a declin- ing habitat suitability with increased isothermality (Bio3), indicating a preference for low uniform temperature distribution. Discussion Considerations and limitations Species distribution models (SDMs) are not without their limitations and underlying assumptions. For instance, they can be susceptible to biases arising from sampling methods, sample size, multicollinearity, and spatial autocorrelation. Additionally, while SDMs can be employed to project species distributions under future conditions, the uncertainty asso- ciated with these projections can be substantial. Furthermore, it is important to acknowledge that SDMs, particularly those focused primarily on climate variables, may not fully capture the influence of rapidly changing factors such as land use and human interventions. While our study utilizes SSPs scenarios to project future climate conditions, which partially cover these aspects by incorporating socioeconomic factors that drive land use patterns, they still represent a simplification of complex real-world dynamics. Therefore, future research could benefit from integrating land use change models and human impact assessments alongside climate projections to provide a more comprehensive understanding

of potential species distributions. Despite these limitations, MaxEnt has become a widely used tool for habitat evaluation studies. However, it is important to note that MaxEnt’s presence-only models generally perform less effectively in species distribution modeling compared to presence-absence models, as absence data can provide valuable insights into spe- cies’ absence points. In contrast, MaxEnt has emerged as a superior tool compared to other species distribution models, especially in the context of evaluating the extensive uncertainties associated with modeling [28]. Therefore, the application of such modeling approaches is appropriate for Iran, considering the limited availability of data on both species distribution and climate [50]. The MaxEnt model's high AUC and TSS values affirm its robustness in delineating the potential habitat for T. undulata, providing us with a credible foundation to discuss the implications of our findings. A common challenge in ecological studies, including species distribution modeling, is obtaining high-resolution climate data that accurately reflect the conditions experienced by organisms in their specific habitats. This is particularly true in regions with limited monitoring infrastructure, like Iran, where data on species occurrences and local climate are often scarce. Gridded climate datasets, such as WorldClim used in this study, help overcome this limitation by providing spa- tially continuous estimates of climate variables [33,51]. However, these datasets have inherent uncertainties due to factors