Estimation of Semiarid Forest Canopy Cover Using Optimal Field Sampling and Satellite Data with Machine Learning Algorithms

Forest canopy cover represents different characteristics of forest stands. However, especially in semiarid forest, the estimation of canopy cover by field measurements is too expensive. Therefore, it is necessary to develop appropriate techniques to estimate forest canopy cover for forest management in semiarid areas. In this research, a robust procedure to estimate canopy cover using stratification field sampling and AdaBoostM(1) machine learning algorithm with Landsat 8 OLI imagery is suggested. Approximately 29,000 ha of semiarid forest (Manesht- and Ghelarang-protected area) in west of Iran was selected as the study area. The unsupervised classification was used on NDVI layer extracted from OLI data, and Neyman method was applied for allocation, in stratified areas. The crown cover was measured in percentage in each plot. In inaccessible plots, the optical satellite imagery of Worldview-2 from Google Earth database was used (0.46 m spatial resolution). For the classification of canopy cover, the AdaBoostM1 algorithm with random forest classifier was trained by 75% split original data, while 25% remaining data were used for accuracy assessment using ROC curve, true positive (TP), false positive (FP), overall accuracy (OA) and kappa coefficient (K). The results showed the overall accuracy and kappa coefficient of 91% and 0.88, respectively. Based on the results, the methodology developed in this study is suitable to estimate canopy cover in semiarid forests.