Evaluating the predictive performance of habitat models for Yarrow (Achillea millefolium)

Predicting geographical species distributions has become essential in several aspects of the biogeographical, environmental, and biological sciences. One of its most important applications is related to conservation and control of species, especially endangered species. A variety of statistical techniques has been used in species distribution modeling that attempt to predict occurrence of a given species in respect to environmental conditions. The objective of this study is to compare the performance of some of the most common methods of presence-absence models to mapping geographic distribution of a non-woody species, Yarrow (Achillea millefolium), in central Germany. This study compared the performance of three machine-learning algorithms, i.e., Random Forest (RF), Artificial neural network (ANN), and Gradient Boosting Machine (GBM). Accordingly, eight topography-climate predictors, i.e., slope, aspect, elevation, topographic wetness index, mean summer temperature, sum summer precipitation, summer solar radiation, and soil moisture index were selected and prepared by GIS techniques for model calibration. Also, ۱۰۲ occurrences of Yarrow were selected as species-presence data. The predictive maps were developed for the target species after calibration of the models. Finally, model accuracy was evaluated using the area under Receiver Operating Characteristics curve (AUC) and true skill statistics (TSS). The results show that the mean summer temperature followed by the sum summer precipitation and the slope are most important predictors for modeling the distribution of Yarrow, and soil moisture index was the least important predictor for this plant species. Also, the results show that the mean AUC values of the three models is ۰.۷۴ (ANN), ۰.۸۴ (RF), ۰.۸۹ (GBM), and ۰.۶۱ (ANN), ۰.۶۹ (RF), ۰.۷۲ (GBM) for TSS index. Both indices illustrate that the GBM model has better performance than the other two models, and the produced distribution map by that way has higher accuracy.

کلیدواژه ها:

Random Forest ، Artificial neural network ، Gradient Boosting Machine