Estimativa do estoque de carbono em plantios de Eucalyptus spp. a partir de métricas LiDAR

Abstract

Currently, the emphasis given to the climatic issues is due to the pronounced increase in the concentrations of carbon dioxide in the atmosphere and its effects over the environment. Among the many environmental services provided by forests, we highlight the sequestering and stocking of carbon. Forest plantations with eucalyptus species represent 5.5 million hectares in Brazil. These species are characterized by the rapid growth and consequent speed in absorbing and capturing carbon available in the atmosphere. The quantification of the carbon stock in the forest biomass is a laborious, time-consuming and onerous activity. New technologies have been developed in order to remedy such limitations, among which is the LiDAR (Light Direction and Raning) technology. The precision of the surveys using LiDAR depends on the traits intrinsic to the system (sensor, point density, flight platform altitude, pulse frequency, footprint, scanning angle, among others). The estimation of biomass and carbon stock obtained from the LiDAR occur via modeling by means of parametric (multiple linear regression) or non-parametric (Random Forest) methods. In this context, the objective of this work was to evaluate the use of LiDAR metrics in the estimation of the carbon stock of Eucalyptus spp plantations, as well as to study the influence of the flight traits and type of modeling. We conducted two LiDAR flights over eight farms owned by the Fibria company. With the cloud of points derived from each flight, we obtained the LiDAR metrics, which were pre-selected according to the correlation with the carbon stock and multicollinearity between them, resulting in six independent variables that comprised the model. The dependent variable of carbon stock was obtained by means of the Schumacher & Hall logarithm model (1993) adapted for the study area. We adjusted models from flight 1 via multiple linear regression (1), flight 2 via multiple regression (2), flight 1 via Random Forest (3) and flight 2 via Random Forest (4). The best models of 1, 2, 3 and 4 obtained R2 ajd of 83%, 84%, 82% and 79%, and RMSE of 7.82, 7.71, 8.02 and 8.72 Mg ha-1 , respectively. In all cases, the independent variables comprising the final models were hp50 and stratum VI. Therefore, with the LiDAR metrics, it is possible to obtaine accurate estimates of the carbon stocks in Eucalyptus planted forests. There was no significant difference between the carbon stock estimates obtained from flights 1 and 2, as well as between those obtained from modeling via multiple linear regression and Random Forest.