Classification of Forest Sites from Estimated Dominant Heights with Artificial Neural Network

Classification of Forest Sites from Estimated Dominant Heights with Artificial Neural Network

Abstract

This work has a aim to compare classification stability of productive capacity of forests with traditional methods to artificial neural networks (ANN). In this study, an mount of 30.893 pairs of data from 9.034 permanent portions of a continuos forest inventory (CFI) were used. Data was gathered from clones in Eucaliptus urophylla and Eucaliptus grandis plantations in Bahia south. Location index curves were drawn with guide curve method from dominant highs (Hdom) and estimated by traditional methods (Chapman and Richards model) and with ANN. Multilayer Perceptron network (MLP) were trained with backpropagation algorithm with 70% of data, while 30% of data was used for testing. Classification stability of sites were analysed with Chi-Square test. ANN presented greater results in Hdom estimation when compared to Chapman and Richards model. Also, use of categorical variable brought improvements to estitions. Moreover, ANN increased classification stability of sites when compared to traditional regression model.

Publication
In Proceedings of 3rd Brazilian Meeting of Forest Measurement (MensuFlor)
Date
Links
PDF

Text in Portuguese, Original Title: Classificação de sítios florestais a partir de alturas dominantes estimadas com redes neurais artificiais.