REGRESSION MODELLING FOR FOREST TIMBER VOLUME ESTIMATION:
AN INITIAL FEASIBILITY STUDY
David Chung-Liang Lee
Yosio Edemir Shimabukuro
Thelma Krug
Bernardo Friedrich Theodor Rudorff
Instituto Nacional de Pesquisas Espaciais
Av. dos Astronautas, 1758
12227-010 Sào José dos Campos, Sào Paulo, Brasil
ISPRS Commission VII / Working Group 3
KEY WORDS: Timber Volume Estimation, Regression Modelling, Dummy Variable, Unbalanced Data.
ABSTRACT
A new concept to estimate timber volume is introduced, based on a 3-way crossed classification linear
regression model on dummy variables for unbalanced data. Qualitative forest parameters (species,
canopy density, and stand age) were used as independent variables in the model. Ground data collected
over 43 field sample plots generated the estimates of the model. The timber volume estimation was
obtained applying the model on a TM-Landsat ima ge, on a pixel by pixel basis over a plantation study
area located in Mogi Guaçu, Säo Paulo, Brazil. Information for species was obtained from digital
classification of the image, whereas data for stand age and canopy density were digitalized into the
system from maps generated from previous studies. Hence, for each pixel, information was available
for the three qualitative variables under consideration. The model was able to explain 90% (r’=0.90) of
the timber volume variation of the field sample plots, therefore encouraging its application for forest
survey over extensive areas.
038