PLOT-LEVEL FOREST VOLUME ESTIMATION USING QUICK BIRD SATELLITE DATA IN THE HYRCANIAN FOREST

摘要

Awareness of the stand volume on forest is essential for management. The objective of study was estimation of forest stand volume using Quickbird data in the Hyrcanian forests of Iran. The images were orthorectified using DEM and ground control points. The proper processing analyses including principal component analysis, pansharp merging, band rationing and vegetation indices were done using main bands. The different texture analyses by different kernel sizes were also applied on the multispectral and panchromatic bands. By random cluster sampling method, 112 plots with size of 10*10 meters, the diameter at breast height (DBH) and height of some trees were measured in natural stand and hand transplant. The volume per hectare for each sample plot was calculated using DBH and height of trees. The positions of center of plots were registered by DGPS in a processing kinematics method. The plot based spectral values of main and processed bands were extracted. After normalizing the data, the genetic algorithm was used to select the more correlated bands among 227 main and processed bands. The results showed that the homogeneity, correlation, GLDV mean (equivalent to dissimilarity), variance and dissimilarity texture analyzed bands with sizes of 4*4 and 8*8 kernel sizes and homogeneity with kernel size of 20*20 on panchromatic band were more correlated with volume variable. Correlation analysis between volume and the best selected spectral bands were investigated by linear and non-linear regression methods. The performances of estimations of best models with highest R~2adj were examined using 20 percent of unused plots by relative RMSe and Bias measures. The results showed that linear regression model could estimated stand volume by R~2adj=0.26, RMSE=68.04% and Bias=25.01%, but the best non-linear logarithm model could estimate the volume with R~2adj=0.31, RMSE=65.91% and Bias=20.77%. For improving the estimations, using of the non-parametric algorithms may be produce better result in a future work.