Combining tree height samples produced by airborne laser scanning and stand management records to estimate plot volume in Eucalyptus plantations

Regular stand structure and availability of precise silvicultural management data produce a special situation regarding remote sensing based assessments of plantation forests. This study tested the use of stand management records to improve single-tree detection in a Eucalyptus plantation. Combined airborne laser scanning (ALS) and planting distance data were used to detect trees and extract their heights. The extracted heights were used as an input for volume estimation using both existing plot-level functions and new tree-level models. The accuracies were evaluated in a test data set of 191 field reference plots in which the diameters of the Eucalyptus urograndis (E. grandis (Hill) Maiden x E. urophylla S.T. Blake) trees varied from 6 to 41 cm and tree heights varied from 12 to 41 m. The constructed mixed-effects model that predicted stem volume from tree height resulted in a root mean squared error (RMSE) of 68 dm(3) (15%) in a cross validation of the modeling data. The tree detection produced estimates of stem number with low bias (i.e., average difference between measured and estimated) and an RMSE of 6% of the mean, whereas plot-level mean and dominant heights were estimated with RMSEs of 1.5 m (5%) and 2 m (6%), respectively, using ALS data alone. The difference of about 60 cm observed between the ALS-based and field-measured dominant height was most likely caused by the penetration of the laser pulses through the canopy. A system of plot-level models that employed a small sample of calibration field data gave RMSEs of 1 m (3%) and 2.2 m(2)/ha (9%) for site index and basal area, respectively. The plot volume was estimated with an RMSE of 44 m(3)/ha (12%) at best. A similar residual variation was observed in the volume estimates of an area-based method applied to the same data set. The combined results suggest the feasibility of the proposed methodology in a plantation inventory using ALS data with a density of only 1.5 pulses/m(2).