Modelagem volumétrica de plantios de eucalipto por meio de dados do Lidar Gedi, Sentinel e Ambientais

Abstract

Forest ecosystems are considered a critical component of the world's biodiversity, being an important indicator of Sustainable Development Goals (SDG) n. 15, Life on Earth – Protect, restore, and promote sustainable use of Earth's ecosystems, sustainably manage forests, combat desertification, halt and reverse land degradation and halt biodiversity loss. It is imperative to know the stock of woody material present in forest plantations so that the management of these areas can meet market demand and respect the sustainable use of terrestrial ecosystems. For this purpose, forest inventories based on field measurements of parameters such as diameter at breast height and total height of trees are used. New techniques for integrating these variables with data obtained by remote sensors, such as optical and radar images, collected by orbital sensors, cameras on board Remotely Piloted Aircraft or LiDAR, are constantly being improved, aiming at reducing acquisition costs and greater efficiency. In this sense, the main objective of this work is to generate volumetric estimates of eucalyptus timber from data from the Global Ecosystem Dynamics Investigation (GEDI) orbital sensor and random forest regression. The study area is located on the North Coast, Agreste and Reconcavo of the Bahia State. Data from forest inventories collected in the field, attributes from the climatological water balance using Terraclimate data, terrain attributes extracted from the digital elevation model of the Shuttle Radar Topography Mission (SRTM) and radar data obtained by the Sentinel-1 satellite were used. The database used consisted of 412 observations, randomly split between training and validation in the proportion 70-30%, which were submitted to tests of normality, homoscedasticity, analysis of variance and mean test. The Wilcoxon test indicated that there is no significant difference at 5% probability between the dominant height of the continuous forest inventory and the relative heights rh100 and rh99 derived from the GEDI. Total height, in turn, did not show significant differences at 5% probability with rh97, rh96 and rh95. Height relative to 97th quantile (rh97), foliage height diversity normalized by plant area index (fhd_norm), and vertical distribution of canopy failure probability (pgap_theta) showed high correlation with volume (correlation coefficients of 0.907, 0.824 and -0.432, respectively). The model that used only the height and vertical structure metrics from the GEDI showed greater accuracy for volume prediction (root mean squared error of 17.37 m3 ha-1 and mean absolute error of 10.72 m3 ha-1).