Abstract

A spatial analysis of the site index used for the classification of Pinus taeda production forests was performed using dominant height data from 402 continuous inventory plots. The data were examined with simple descriptive statistics and fit with four semivariogram models by the GS + program. The best model was then used to predict the site index in unsampled areas by ordinary kriging in ArcView. All models showed that site index values exhibited spatial dependence, with the degree of spatial dependence ranging from strong to moderate. The spherical model was used for kriging. In this model, the degree of spatial dependence was 29% and the range was 5,330 m, with a residual sum of squares (RSS) of 3.00 and coefficient of determination (r²) of 0.776. Measured and predicted values were compared by cross-validation, which produced a linear regression of observed versus predicted value with a slope coefficient of 1.068, slope standard error of 0.070, and intercept coefficient of -1.45. The site classification map generated by kriging divided the studied forests into five classes. Before kriging, all of the forest stands had one global average value for the site index, but after kriging this was changed to there being two or three values of the site index for many stands. Ordinary kriging proved to be an optimal method for interpolating the site index of unsampled areas to permit their classification, as is the case for young plantations for which inventory samples have not yet been taken.

This work is licensed under a Creative Commons Attribution 4.0 License.