Applying Artificial Neural Networks and Multiple Linear Regression models to estimate Forest density in Marivan forests

Document Type : Research Paper

Author

Abstract

Studying and modeling quantitative characteristics of forest to develop and direct the ecosystem toward optimal aims and conservative activities is considered as an eminent operation. For this purpose, tree density and forest canopy cover which support each other as important criteria of forest density using linear regression models and artificial neural networks by various variables including; topographic attributes, soil properties, climatic parameters and remote sensing data in some parts of the Baghan forests in Marivan region. The topographic attributes maps were derived from DEM. Climatic parameters and soil properties were extracted using climatic maps and soil Analysis. In order to use satellite imagery data, Landsat 5 images and NDVI index were used. Forest inventory was performed in order to determine its quantitative characteristics based on obtained data from 89 sample plots (0.1 hectare area). The relationship between the forest characteristics and these attributes was analyzed and modeled using Multiple Linear Regression and Artificial Neural Network models. R2 and RMSE for the Neural Network method to predict canopy cover and trees density were as follows: R2=0.92, RMSE=10.20% and R2=0.84, RMSE=11.32% for canopy cover and tree density, respectively. The amounts of the mentioned parameters for estimation with multiple linear regression model were: R2=0.81, RMSE=15.02% and R2=0.68, RMS=16.52%, respectively. Results indicated that there is an appropriate potential of combination the topographic attributes, soil properties, climatic parameters and remote sensing data in estimating the forest density and the linear regression model can be replaced by artificial neural networks model regarding to its high performance.

Keywords

Main Subjects