Comparison of different models in predicting habitat suitability of rare species in Uzbekistan: 8 rare Tulipa species case-study

Abstract

Species distribution models (SDMs) have become an essential tool in conservational biology, biogeography and ecology. But there is no consequence in what SDM method is the most efficient in predicting suitable habitat distribution of rare species. To explore this issue, we chose 8 rare Tulipa species in Uzbekistan as case study to test 8 common Machine Learning (GLM, GBM, MARS, CTA, SRE, FDA, RF, MaxEnt) and Deep Neural Network (DNN) SDM models, using three different methods of pseudo-absence data generation (random sampling, random sampling with exclusion buffer, random sampling with environmental profiling). To compare the effectiveness of each model 3 common metrics (Area under ROC (AUC), True skill statistics (TSS) and Cohen Kappa (K)) were used. We have found that RF and GBM combined with RSEP strategy are superior to other modeling methods.