Evolução na modelagem da distribuição espacial do estoque de carbono arbóreo no Cerrado brasileiro

Abstract

Understanding the carbon stock spatial distribution allows for the definition of priority areas for conservation. Furthermore, the understanding of the carbon stock distribution within priority areas is critically important for public forest management efficiency. The development of an approach that allows for the precise carbon stock distribution mapping in the Brazilian Savanah, as well as an approach that allows for the precise carbon stock estimates with a low level of uncertainty for different forest fragments is relevant to the current forest context. This dissertation presents a hybrid approach for the spatial modeling of aboveground carbon stock in the Savanah of Minas Gerais (Article 1), and a growth and prediction (CP) approach for the precise carbon stock spatial mapping with low degree of uncertainty (Article 2). A database from the Forest Inventory of Minas Gerais was used in this dissertation. In addition, remote sensing variables were extracted from Landsat TM while bioclimatic variables were extracted from WorldClim for all forest fragments. Overall, Article 1 highlights how the hybrid approach, which combines Lasso (least absolute shirinkage selection operator) for variable selection, maximum likelihood for fitting a linear model, and the ordinary kriging of the residuals based on the linear model, was substantially more precise than conventional approaches applied for this type of study. In particular, the hybrid approach allows for a precise carbon stock distribution mapping in the Savanah across the environmental gradient. Article 2 emphasizes the importance of all auxiliary variables, especially the past inventory results. Forest inventory, bioclimatic and remote sensing variables were combined and used to fit a predictive model of carbon stock distribution. The use of several auxiliary variables allowed for the development of an approach that not only precisely maps the carbon stock distribution, but also provide precise forest fragment (site)-specific carbon stock estimates. The different approaches presented in this dissertation add great value on the way to quantify the carbon stock in native forests in Brazil.