Evaluation artificial neural network method for spatial mapping of species potential habitat (Case study: Rangeland Siah bisheh, Mazandaran)

Prediction of the spatial distribution of Festuca Ovina and Bromus briziformis in Siahbisheh Rangelands using artificial neural network was the purpose of this study. Random classification sampling was done for vegetation in 29 homogenous units. 290 plot 1 m² were established in the area and was recorded percent of canopy cover. 3 soil samples were collected from a depth of 0-30 in any homogenous unit. In this study, 20 Environmental factors (Slope, aspect, elevation, distance from road, distance from river, precipitation, distance from livestock, geology, percent of silt, clay, sand, moisture, carbon, organic matter, ph, EC and N.P.K) were independent variables and species presence data of Festuca Ovina and Bromus briziformis was dependent variable. The information layers of each these factors prepared in Arc GIS and were classified using the frequency of each these factors. The results showed that the most important environmental variables affecting the distribution of the studied species were elevation, soil texture and nutrients. Then 70 and 30 percent of the data were used for training and test network respectively. In this study, artificial neural network structure with the 20 neurons in the input layer and the hidden layer and one neuron in the output layer, values of MSE were calculated for festuca 0.75 and Bromus 0.72. Then zoning maps of plant species were prepared with 4 zones including absence and presence of low, medium, high. Zoning maps were evaluated using ROC curves and Kappa coefficient that accuracy with ROC curves were 97.10, 84.10 and with kappa coefficient were 0.78, 0.66 percent for Festuca ovina, and Bromus briziformis respectively that represents a good evaluation of model.