Predição da biomassa aérea em plantações de Pinus taeda L. por meio de dados LiDAR aerotransportado

Abstract

ResumoEste trabalho teve como objetivo predizer a biomassa acima do solo (AGB) em plantações de Pinus taeda L., localizados na região sul do Brasil. A base de dados utilizada no estudo foi originada de levantamentos a laser aerotransportados (LiDAR), complementados por informações de campo. Os modelos preditores da biomassa foram ajustados por modelagem não paramétrica, Random Forests (RF), implementado no ambiente R. Para compor os dados de campo foram inventariadas 50 parcelas de área fixa, nas quais foram mensurados os diâmetros de todas as árvores e parte das alturas. Posteriormente o volume individual foi estimado por modelos polinomiais de quinto grau e serviu de base para o cálculo dos valores de AGB. O modelo final, preditor da biomassa, foi composto pelas métricas LiDAR referente a Altura no percentil 99 (H99TH), Coeficiente de Variação da Altura (HCV) e Assimetria da altura (HSKEW) por proporcionarem baixa correlação entre si, e fornecerem os maiores valores de Razão de Melhoria do Modelo (MIR). O modelo final apresentou um o coeficiente de determinação ajustado (R 2 aj. ) de 0,98, Raiz Quadrada do Erro Médio (RMSE) de 5,98%, enquanto que para a validação, esses valores foram de 0,93 e 12,64%, respectivamente. Portanto conclui-se que o modelo gerou resultados satisfatórios na predição da biomassa aérea em plantações de P. taeda, podendo ser considerado como uma ferramenta eficaz no manejo de florestas plantadas. Palavras-chave:Inventário Florestal, Modelagem Não-paramétrica, Plantações Florestais, Métricas LiDAR. AbstractThe aim of this study was to predict aboveground biomass (AGB) from Pinus taeda L. plantations, located in South of Brazil, using LiDAR data, in-situ measurements and Random Forests (RF) modeling. Fifty regular sample plots were used, in which the diameter at the breast height (DBH) for all trees and about 15% of the heights were measured. Afterwards. forest stem volume was predicted using a fifth degree polynomial model, and used to calculate the field AGB values. To create the RF model we selected the H99TH, HCV and HSKEW LiDAR metrics, because they were not highly correlated to each other and presented the higher calculated value of Model Improvement Ratio (MIR). The estimative model of AGB presented a coefficient of determination (Adj.R 2 ) of 0.98 and RMSE of 5.98%, while for the validation these values were 0.93 and 12.64%, respectively. It was possible to conclude that the RF and LiDAR-derived metrics were able to predict precisely the values of AGB in P. taeda plantation, therefore, it can be used as a helpful tool to forest management.

ESTIMATION OF BIOMASS, CARBON AND CO2 OF Khaya senegalensis (Desr.) A. JUSS

“…The average amount of carbon estimated per tree was 7.82 kg (Table 6), a value lower than that found by Silva et al (2015), when they estimated carbon stock in the aboveground biomass in plantations of Eucalyptus spp. aged 2.3 years (27 months) in São Paulo, and found a total of 12,45 kg.árv -1 , average DBH of 8.35 cm.…”

Aboveground biomass stock and change estimation in Amazon rainforest using airborne light detection and ranging, multispectral data, and machine learning algorithms

“…Among the LiDAR metrics, the mean height (Hmean) had the highest correlation with AGB. Silva et al 36 highlighted that LiDAR height metrics correlate with forest attributes used in estimation models. In the same context, Silva et al 28 used the metric mentioned above to estimate the stock and changes in AGB in different digital terrain model (DTM) scenarios and pulse densities in the same study area of this work, obtaining satisfactory results.…”

“…Initially, the significance criteria directed the variables selection to estimate AGB. According to Silva et al., 36 selecting the most significant variables for modeling is crucial to obtain accurate estimates. Thus, due to the high number of LiDAR variables, the selection of variables that presented a significant Pearson’s correlation (p-value < 0.001) with the AGB.…”

“…In possession of the most significant variables, according to the RF, Pearson’s correlation was analyzed to identify the highly correlated (r≥0.9) with each other, as proposed by Hudak et al 22 . After identifying the high correlation between two or more variables, only the one with the highest correlation with AGB remained, eliminating the others 36 . This process is necessary to eliminate multicollinearity in modeling, which arises when at least two highly correlated predictors are assessed simultaneously in a regression model, leading to biased errors and misleading conclusions about the role of each of the predictors 40 .…”

Determination of the biomass stock in a mixed plantation of Pinus taeda L. and Pinus elliottii Engelm.

“…Of note, forest biomass production is affected by multiple factors, such as the site quality, genetic ability of the species, and age and density of the plantation. In this context, numerous studies have evaluated biomass production of forests with diverse species and at different sites (Watzlawick et al 2013;Lisboa et al 2015;Lima et al 2016;Silva et al 2017;Sanquetta et al 2019;Péllico Netto and Behling 2019) to gather data on forest plantations. Upon the identification of such factors, biomass production of plantations can be improved.…”