Nonparametric method: Kernel density estimation applied to forestry data

Abstract

Probability density function can be fitted through parametric or non-parametric methods. The use of a non-parametric method is interesting and appropriate, considering its flexibility and better adjustment to multimodal data. The objective of the present study was to compare the performance of the non-parametric distribution in relation to the parametric distribution in these cases. We used six separate sets of forest database with bimodality and asymmetric characteristics. The probability density functions were estimated for each set of data using the KDE method. To evaluate the effectiveness of the KDE method, parametric probability distributions were also adjusted for the same data. The Kolmogorov-Smirnov test was applied to evaluate the goodness of fit of the parametric distributions. The distributions obtained through the two methods were compared graphically to identify if the nonparametric and parametric methods are equally efficient to obtain the underlying distribution, especially for bimodal and asymmetric distributions. The KDE method is an appropriate alternative for describing probability distributions in forest data, especially when bi- or multimodality occurs.