Characterising forest structure using combinations of airborne laser scanning data, RapidEye satellite imagery and environmental variables

摘要

The objective of this study was to compare the utility of combinations of data from airborne laser scanning (ALS), RapidEye satellite imagery and auxiliary environmental data to predict stand structure in a plantation forest. Both parametric and non-parametric modelling techniques that could simultaneously predict a multivariate response were employed and found to produce predictions with similar levels of accuracy. Response variables were derived from 463 field measurement plots that were used during model development; a further 60 randomly selected plots were set aside for validation of model performance. Candidate predictor variables were extracted from the ALS data, satellite data and auxiliary environmental data, and the variables with the greatest explanatory power were used to create six separate models based on combinations of the data sources. Model validation showed that models using RapidEye data only were the least precise and that adding auxiliary environmental data only led to a moderate improvement in model precision. The model precision observed was similar to those reported previously from studies using satellite data to predict stand structure. Models developed using data from ALS were by far the most precise and adding information from satellite data or auxiliary environmental data led to negligible improvement in the prediction of stand structure. Although the outputs of both model types were similar, the practical efficiencies of using the non-parametric approach make it appealing to meet the demands of managers of industrial plantation forest managers.