| Summary: | The present study was carried out with the objective of obtaining regression equations and Artificial Neural Networks (ANNs) for the prognosis of Pinus caribaea var. caribaea survival in Macurije Forest Company, province of Pinar del Río - Cuba. The data used in the modeling comes from the measurement of the variables age (years) and survival (density) in circular permanent plots of 500 m² established in P. caribaea var. caribaea plantations. The study was divided into three stages: i) Adjustment of survival traditional regression models; ii) Training of ANNs for survival prognosis, including categorical variables «site» and «Basic Units of Forest Production»; iii) Comparison of regression equations performance with those of ANNs in survival prognosis. The best models and ANNs were selected based on: adjusted determination coefficient - R2aj (%), square root of the mean square error - RMSE (%) and residue distribution analysis. The evaluation of the models goodness of fit also included the verification of the assumptions of normality, homocedasticity and absence of serial autocorrelation in the residues by Kolmogorov-Smirnov, White and Durbin-Watson tests, respectively. The model of Pienaar and Shiver (1981) turned out to be the best fit in survival prognosis. The ANN MLP 13-10-1 was the one with the best generalization capacity and presented a performance similar to that of Pienaar and Shiver equation.
|
|---|
