Estimating changes in forest cover in the Rudsar county by using neural network and maximum likelihood methods

Subject Areas : Geospatial systems development

Seyed Reza Fatemti Talab ¹ , Morteza Madanipour Kermanshahi ² , Seyed Armin Hashemi ³

1 - Assis. Prof. Department of Environment, Abadeh Branch ,Islamic Azad University
2 - Assis. Prof. Department of Forestry, Shabestar Branch, Islamic Azad University
3 - Assis. Prof. Department of Forestry, Lahijan Branch, Islamic Azad University

Received: 2014-06-15 Accepted : 2015-02-21 Published : 2015-08-23

Keywords:

Abstract :

The acquisition of knowledge about the vegetation plays an important role in soil management.  However, vegetation estimating in the usual way, including an overall assessment of the vegetation is  time consuming and does not also provide accurate enough information. Therefore, remote sensing technology is a desirable way for reducing time and cost compared to other usual methods. In this study, forest cover maps were prepared using remote sensing techniques and  LandSat ETM+ imagery of year 2000 and LandSat 8 of year 2013. The classification of the study area digital images was performed  to prepare  land use map classification using maximum likelihood and neural network with participation of different bands. The results showed that the best overall accuracy of image classification using neural networks ETM+ in 2000 and LandSat 8  in 2013  was 0.95 and 0.95 respectively. It was also indicated that the kappa coefficient was estimated 0.91 and 0.91 respectively. The overall accuracy of maximum likelihood method of the collected images of  2000 and 2013 was  0.95 and 0.85, but it was 0.86 and 0.84 for Kappa statistics method. The results also showed a 1054.507 and 635.319 hectares decreasing of forest cover using  neural network classification  and maximum likelihood classification methods respectively. According to classification accuracy and Kappa statistics, it was observed that the accuracy and kappa coefficient of neural network classification was higher than accuracy and the Kappa coefficient of maximum likelihood method.