Discrete-return LiDAR data and model development to predict aboveground carbon stocks in Eucalyptus spp. Plantations in Brazil
In the context of global climate change,the quantification of carbons stocks in areas containing commercial forest plantations has been receiving great attention.This is mainly due to the fact that forests play an important role within the global carbon budget.Light Detection and Ranging (LiDAR) technology has surged as a good alternative for measuring carbon stocks in plantation forests,and it may enable precise data collection at high spatial resolution and in relatively short time compared to traditional methods.In this investigation,discrete-return LiDAR data were evaluated to estimate total Above Ground Carbon (AGC) in different plantation forests containing Eucalyptus spp.Field data collect areas were stratified into different ages,representing both genetic and forest site productivity aspects.A total of 141 sample plots were established.Individual trees were harvested and their biometric parameters measured.Afterwards,carbon measurements were performed in the laboratory.The AGC measured in the field was considered as the dependent variables and the independent variables were metrics obtained from the LiDAR point cloud.These metrics were correlated with the in-situ field AGC measurements to develop predictive carbon models.We developed a multiple linear regression model from a suite of 28 candidate predictor variables derived from LIDAR data to create eight AGC models using best subsets regression.The AGC models differed in the number of predictor variables,ranging from a model with 1 predictor variable to an model with 8 predictor variables.The best AGC model was selected through corrected Akaike Information Criterion (AICc) and evaluated through the Root Mean Square Error (RMSE),Coefficient of determination (R2) and Pearson’s correlation (r).All the data processing and analysis were performed using open-source and freeware software.In the findings,the AICc ranged from 211.33 to 244.14,and the best model used five predictor variables,whereas the worst fitted model used one predicted variable.The predicted variables selected by the best AGC model were the 80th percentile for intensity,cloud density above 2 m (canopy),standard deviation of height,coefficient of variation of height and the 75th percentile height (Ip80,dc,hdesv,hcv,hp75,respectively) and the performance was proved by a R2 of 0.89,a RMSE of 7.48 Mg ha-1 (RMSE% 13.17) and a Pearson’s r of 0.92.The results obtained in this investigation show that LiDAR data can be used to accurately predict AGC in Eucalyptus spp.plantations in Brazil.
Carlos Alberto Silva Carine Klauberg Samuel de P(a)dua Chaves e Carvalho Andrew T.Hudak Luiz Carlos Estraviz Rodriguez
Department of Forest Resource, University of S(a)o Paulo - ESALQ/USP, Piracicaba-SP, Brazil Forest Inventory Specialist - Fibria Celulose S/A, Cap(a)o Bonito SP, Brazil United States Forest Service Rocky Mountain Research Station, Moscow, ID 83843, USA
国际会议
北京
英文
239-247
2013-10-09（万方平台首次上网日期，不代表论文的发表时间）