Comparison of artificial neural network and multivariate linear regression (MLR) models to predict cover percentage Artemisia aucheri from some soil properties

Subject Areas :

Mansoreh Kargar ¹ , Zeynab Jafarian ²

1 - Department of Rangeland and Watershed, College of Natural Resources, Sari Agricultural Science and Natural Resources University, Sari, Iran.
2 - Department of Rangeland and Watershed, College of Natural Resources, Sari Agricultural Science and Natural Resources University, Sari, Iran.

Received: 2013-04-24 Accepted : 2014-06-20 Published : 2014-07-23

Keywords: Artificial Neural Network (ANN), Vavsar rangeland, Artemisia aucheri, multivariate linear regression (MLR),

Abstract :

Soil ecosystems management for different purposes requires accurate and quantitative understanding of the soil characteristics and their processes. This study was aimed to predict Artemisia aucheri cover though some soil physical and chemical properties in Vavsar rangeland, Kiasar, Mazandaran province. Random systematic sampling was used. Five transects with 100 m length and 10 plots 4 m2 on each transect were established. Then cover (%) of A. aucheri and 50 soil sample from 0-15 cm depth was estimated in each plot. Soil properties including soil organic carbon, total nitrogen, EC, water percentage, CaCo3 percentage, soil texture, and pH were measured. Data were divided in two series: a series for analysis including 70% of the data for and 30% for evaluation of customized models. Result showed that soil water, silt and sand percentages were the most important soil properties for prediction A. aucheri cover in the study area. Prediction of the statistical models in the study area resulted in mean error and root mean square error values of 0.25, 0.06 for ANN equation and 0.43, 0.12 for MLR, respectively. Therefore, the ANN model could provide superior predictive performance when was compared with MLR model.