Modeling spatial distribution of plant species using autoregressive logistic regression method-based conjugate search direction

Modeling plant habitat range distributions is critical for monitoring and restoring species in their natural habitat. The classical logistic regression (LR) model for plant habitat distribution has several drawbacks such as neglecting the effects of the important variables and sensitivity to non-correlation variables. In this paper, an autoregressive logistic regression (ALR)-based conjugate gradient training approach was proposed to improve the drawbacks of LR in predicting the presence and absence of spatial habitat distribution based on input attributes including soil gypsum amount (gyps), lime content, soil available moisture (AM), soil electrical conductivity (EC), clay, and gravel amounts in Poshtkouh rangelands of Yazd Province, Iran. The conjugate gradient approach to calibrate logit model is extended by an iterative formulation using a limited scalar factor and adaptive step size. The predicted results of the classical LR and ALR were validated for nine plant habitats based on several comparative error statistics. The results illustrated that different coefficients were obtained for LR and ALR models but the proposed ALR performed better than the LR in estimating the occurrence probability of plant species.