Predicting stem total and assortment volumes in an industrial Pinus taeda L. forest plantation using airborne laser scanning data and random forest

Abstract

Improvements in the management of pine plantations result in multiple industrial and environmental benefits. Remote sensing techniques can dramatically increase the efficiency of plantation management by reducing or replacing time-consuming field sampling. We tested the utility and accuracy of combining field and airborne lidar data with Random Forest, a supervised machine learning algorithm, to estimate stem total and assortment (commercial and pulpwood) volumes in an industrial Pinus taeda L. forest plantation in southern Brazil. Random Forest was populated using field and lidar-derived forest metrics from 50 sample plots with trees ranging from three to nine years old. We found that a model defined as a function of only two metrics (height of the top of the canopy and the skewness of the vertical distribution of lidar points) has a very strong and unbiased predictive power. We found that predictions of total, commercial, and pulp volume, respectively, showed an adjusted R2 equal to 0.98, 0.98 and 0.96, with unbiased predictions of -0.17%, -0.12% and -0.23%, and Root Mean Square Error (RMSE) values of 7.83%, 7.71% and 8.63%. Our methodology makes use of commercially available airborne lidar and widely used mathematical tools to provide solutions for increasing the industry efficiency in monitoring and managing wood volume.