In recent years, there is a growing concern related to the effects of climate change on species' distributions. Thus, potential change of distribution of species is rising as a critical issue for ecologists and conservation biologists.

A popular machine learning method, artificial neural networks, is implemented to model the geographic distribution of tree species in North America. An artificial neural network (ANN) with backpropagation algorithm is used to learn relationships between climatic parameters and species distribution. Three main climatic variables from CRU 1.0 Dataset at 0.5° latitude/longitude resolution containing 1961-1990 climatology are used: mean temperature, diurnal temperature range and precipitation. Five plant species from digital representation of "Atlas of United States Trees" (Elbert et al., 1999) are chosen as having different geographic distributions: Amelanchier alnifolia, Betula papyrifera, Prosopis juliflora, Asimina triloba, Zanthoxylum fagara.

The ANN is constructed accepting at input nodes climatic variables for each month (i.e. 36 input nodes) and giving output as binary value of presence/absence of species at selected point of the 0.5° grid. Threshold values achieving maximum accuracy on test data are chosen to classify outputs as absent or present. Then, the network is used to predict presence/absence of species with respect to changing climatic conditions which are obtained from 0.5° global 2001-2100 Climate Change Scenarios -- TYN SC 2.0 Dataset (Tyndall Centre, Mitchell et al, 2003) having four SRES emissions scenarios (A1FI, A2, B2, B1) and five models (CGCM2, CSIRO mk 2, DOE PCM, HadCM3, ECHam4). The model predicts that, in last quarter of 21^st century, plants native to arid climates will expand towards regions currently with temperate or humid climate.