2.4. Climate niche modeling
Three distinct algorithms were employed within an ensemble modeling framework to assess the current and future climate niche of the species. These algorithms included the Generalized Linear Model (GLM) proposed by Nelder and Wedderburn in 1972, the Generalized Boosted Model (GBM) introduced by Friedman in 2001, and the Maximum Entropy (MaxEnt) method developed by Phillips and others in 2006, as outlined by Breiner et al. in 2015.
To generate pseudo-absence data, 10,000 background points were created by randomly sampling at a specific distance from the presence points, following the methodology suggested by Naimi et al. in 2022 and Collart et al. in 2023.
The evaluation process involved splitting the occurrence data into two sets: 70% for calibration and 30% for evaluation, following the guidelines set by Ngila et al. in 2023 and Collart et al. 2023. This entire procedure was repeated ten times to ensure the creation of predictions independent of the training data, as recommended by Guisan et al. in 2017.
For all three modeling algorithms, the default parameters provided by the biomod2 R package, as detailed by Thuiller et al. in 2009, were utilized.
The performance of the models was assessed using two distinct evaluation metrics: firstly, the Area Under the Curve (AUC) of the Receiver Operating Characteristic (ROC) curve, as introduced by Jiménez‐Valverde in 2012, and secondly, the True Skill Statistics (TSS) method developed by Allouche et al. in 2006.
Finally, an ensemble habitat suitability map was created using a weighted average approach for two species, following the procedure described by Ahmadi et al. in 2023. The predictions derived from the ensemble models were classified into five categories: (1) unsuitable (≤ 0.2), (2) low (0.21-0.4), (3) moderate (0.41-0.6), (4) high (0.61-8), and (5) excellent (≥ 0.81). Notably, the high and excellent categories were considered acceptable thresholds for the analysis.Top of Form