Climatic niche model
I developed MaxEnt climatic niche model in R using the ENMeval package (Muscarella et al., 2014) by considering eight climatic variables. I validated each model with 10-fold cross-validation, tuned it using six feature class combinations (”L”, ”LQ”, ”H”, ”LQH”, ”LQHP”, ”LQHPT”) and several candidate regularisation multipliers. While choosing the best model, I chose the one with the highest AUC (the Area Under the Curve) score. I used the ‘checkerboard2’ evaluation method to handle model overinflation resulting from biased sampling (Muscarella et al., 2014; Chowdhury et al., 2021b,e), which partitions both geospatial records and background points into evaluation bins to reduce spatial autocorrelation between points in the testing and training bins (Muscarella et al., 2014). I used the best-fitted model (with the current climatic data) and predicted the future suitability (with the future climatic data) using the ‘dismo’ R package (Hijmans et al., 2017).
Based on the maximum sum and sensitivity statistics, I did threshold the suitability map (Liu et al.,2016) and converted it into binary (1 (presence): suitability value > threshold; 0 (absence): suitability value ≤ threshold). To threshold the future suitability map, I used the same threshold value obtained from the best-fitted model under the current climatic condition. Overall, I obtained five suitability maps (current, ssp126, ssp245, ssp370, and ssp585) for each of the 242 butterfly species.