The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Publications Copernicus
Articles | Volume XLIII-B4-2022
01 Jun 2022
 | 01 Jun 2022


S. Mehri and A. A. Alesheikh

Keywords: Artificial intelligence, Classification, Feature selection, Neural networks, Oak decline

Abstract. Oak decline is a complex phenomenon. The classification of oak decline potential could be a valuable tool for forest management. This paper identified seven factors that influence oak decline: height, slope, aspect, temperature, perception, soil type, and aerosol. Then, factor analysis is used to determine factors that should be included in oak decline potential classification and reduce data complexity.

As a result, five components explaining 92.49% of total variance are selected. The first component explains 40.34% of the variance, and three factors, including perception with positive and temperature and aerosol with negative load, have contributed to its construction. The second component is composed of a positive load of aspect, and soil type explains 14.89% of the variance. By explaining 14.10% of the variance, the third component consists of soil type and aspect with positive and negative loads, respectively. Slop and height have a positive load in constructing the fourth and fifth components.

Five extracted components are used as input sets of PNN, MLC and SVM methods. 80% of samples are used for training methods, and 20% are used for testing purposes. Results are compared based on the overall accuracy of the methods.

These components are used as an input set of three classification methods, including Probabilistic Neural Network (PNN), Maximum Likelihood Classification (MLC) and a Support Vector Machine (SVM). Based on the results, the SVM, with an overall accuracy of 0.87%, has proved its capability in oak decline potential classification.