Comparison of Regression and Machine Learning techniques in Determination of Geographical Range of Onobrychis cornuta L. under Environmental Characteristics and Climate Change using the IPSL-CM6A-LR Model

Predicting the effect of climate change on native ecosystems is one of the longstanding goals of ecologists and is essential for their conservation and management. Species distribution models (SDMs) are the most widely used tools to predict the effects of climate change on the geographical range of plants. In this study, two regression techniques (GLM and MARS) and two machine learning techniques (ANN and RF), along with environmental factors were used to predict the distribution of Onobrychis cornuta L. The species response to future climate (2050-2070) was investigated under optimistic (SSP1-2.6), pessimistic (SSP3-7.0) and very pessimistic (SSP5-8.5) emission scenarios of the IPSL-CM6A-LR climatic model from CMIP6 models. Based on results, the ensemble model and then MARS presented the most accurate prediction. ANN had the lowest prediction accuracy with a significant difference from other models (p<0.05). The sensitivity analysis revealed altitude (24.64%), maximum temperature of the warmest month (20.31%), temperature seasonality (16.57%) and diurnal range of mean temperature (16%) as the most effective variables on the distribution. According to the ensemble model, the suitable habitat occupies about 27% of the area, but its distribution will be decrease under the future climate. The SSP5-8.5 scenario will have the greatest impact on the displacement of the species distribution range. The resulting prediction maps provide valuable information for conservation strategies, including identifying suitable places for its reintroduction and cultivation in the framework of rangeland management plans.