Plot-level Forest Volume Estimation Using Airborne Laser Scanner and TM Data, Comparison of Boosting and Random Forest Tree Regression Algorithms

摘要

The optical remote sensing data is often providing two-dimensional spectral information, but airborne laser scanner (ALS) provides three-dimensional vegetation structure data. Combining the spectral responses and structure vegetation information could be more useful to estimate forest attributes compared to using only one of data sources. In a small case study in the south western of Germany, the TM and ALS data was investigated for plot-level volume estimation using regression tree-based method of random forest (RF) and boosting regression tree (BRT). The volume per hectare of 411 field plots were calculated using DBH and height of trees. The plot base of height statistics and canopy cover images were extracted from ALS data corresponding to grid cell size of TM images and plots. The TM data using orthorectified CIR image was georeferenced by image-to-image registration methods. The proper vegetation indices, tasselled cap, and principal components were generated on the TM bands. The RF and BRT were examined using a bootstrap learning method on the independent data sets of TM, ALS and combined images. The performances of implementations were examined using relative RMSe and Bias measures. Results showed that employing the BRT using combined ALS and TM data sets could produce more performance with RMSe of 40.56% compared to 42.93% for RF. Combining TM and height-based ALS data could also predict the most accurate volume with 40.56% RMSe than using only one of them with 42.76% and 52.82% RMSe respectively.