Assessing the desertification trend using neural network classification and object-oriented techniques (Case study: Changouleh watershed - Ilam Province of Iran)

• Main Authors: Abdolreza Mohamadi, Zahedeh Heidarizadi, Hadi Nourollahi
• Format: Article
• Language: English
• Published: İstanbul University 2016-07-01
• Series: İstanbul Üniversitesi Orman Fakültesi Dergisi
• Subjects: Neural network classification; object-oriented classification; land use changes; Changouleh watershed
• Online Access: http://dx.doi.org/10.17099/jffiu.75819
• ISSN: 0535-8418; 0535-8418

Desertification consists of decline in production and ecological activities, which may happen due to either natural or unnatural (human) factors.This phenomenon is more evident in arid and semi-arid areas. The aim of this study is to assess the desertification trend using neural network classification and object-oriented techniques in Changouleh watershed which covers an area of 9949 hectare and is located in the south of Ilam province. For this study, TM and ETM+ satellite images of 1984 and 2013 were used. After conducting geometric and atmospheric corrections, images were classified using two neural network and object-orientedalgorithms. Moreover, to evaluate the accuracy and control the correctness of the obtained maps, typical parameters such as Kappa coefficient, the Confusion matrix, and stability of the classification were extracted for assessing the accuracy. The results show that most changes are related to increase in bare lands and decrease in poor and fair rangelands; therefore, approximately 18% of these areas has turned into desert. The results of evaluation of maps correctness show that these two methods are of high accuracy, but the object-oriented approach with Kappa coefficient (94%) and overall accuracy (96.26 %); in addition to being able to detect and categorize more classes, has a high accuracy compared to neural network method.