Modelagem volumétrica da necromassa lenhosa em floresta manejada e não manejada na Amazônia Central

Ciencia Florestal

article

Celso Paulo de Azevedo¹  Cintia Rodrigues de Souza¹  Natalino Calegario²  Filipe Campos de Freitas²  Daniel Dantas² 

[1] Empresa Brasileira de Pesquisa Agropecuária;Universidade Federal de Lavras

关键词: Carbon;    Dead wood;    Necromass;   

DOI  :  10.5902/1980509848108

学科分类：农业科学（综合）

来源: Universidade Federal de Santa Maria * Centro de Pesquisas Florestais

【 摘 要 】

Woody necromass represents about 20% of the carbon available in above-ground biomass in Amazonian forests, however its quantification is not a common activity in forestry studies. The aim of this paper was to modeling necromass volume in order to provide a tool for quantifying this component of vegetation in Amazonian forests. Data collect was carried out in 15 permanent plots allocated in a logged forest and 5 plots in an unlogged forest, both at Amazonas state, Brazil. The volume determination was done for fallen logs and piece of fallen logs from dead tree identified within the boundaries of the plot with a minimum diameter of 10 cm. The modeling was done by non-linear regression (Schumacher-Hall model), training of Artificial Neural Networks (ANN) and Support Vector Machine (SVM) in which the data were divided into 80% for fitting/training and 20% for testing. The modeling accuracy was assessed by these following indicators: correlation between estimated and observed volume, root of the mean square error in percentage and residual plot. A total of 1049 logs or pieces of fallen logs from dead trees were measured, 848 in the logged forest area and 201 in the unlogged forest. The three tested methods obtained a correlation between observed and estimated volume close to 1. The lowest RMSE% were 33,21% for logged forest (ANN) for training database and 22,38% for unlgged forest (Schumacher-Hall) for testing database. The ANNs had better performance during the training stage, however there was no good extrapolation of their results to the testing database. The best volume estimates for testing database were obtained from Schumacher-Hall model. The modeling of the individual volume of fallen dead logs presented great difficulty in minimizing the estimation errors due to the data characteristics.

【 授权许可】

CC BY-NC