African Journal of
Agricultural Research

  • Abbreviation: Afr. J. Agric. Res.
  • Language: English
  • ISSN: 1991-637X
  • DOI: 10.5897/AJAR
  • Start Year: 2006
  • Published Articles: 6693

Full Length Research Paper

Selection of input vectors for estimation of aboveground biomass of Mimosa scabrella Benth. using an artificial neural network

Aline Bernarda Debastiani
  • Aline Bernarda Debastiani
  • Post Graduation in Forest Engineering, Federal University of Parana, City of Curitiba, State of Paraná, Brazil.
  • Google Scholar
Ana Paula Marques Martins
  • Ana Paula Marques Martins
  • Post Graduation in Forest Engineering, Federal University of Parana, City of Curitiba, State of Paraná, Brazil.
  • Google Scholar
Carlos Roberto Sanquetta
  • Carlos Roberto Sanquetta
  • Post Graduation in Forest Engineering, Federal University of Parana, City of Curitiba, State of Paraná, Brazil.
  • Google Scholar
Sebastiao do Amaral Machado
  • Sebastiao do Amaral Machado
  • Post Graduation in Forest Engineering, Federal University of Parana, City of Curitiba, State of Paraná, Brazil.
  • Google Scholar
Ana Paula Dalla Corte
  • Ana Paula Dalla Corte
  • Post Graduation in Forest Engineering, Federal University of Parana, City of Curitiba, State of Paraná, Brazil.
  • Google Scholar
Edilson Urbano
  • Edilson Urbano
  • Post Graduation in Forest Engineering, Federal University of Parana, City of Curitiba, State of Paraná, Brazil.
  • Google Scholar


  •  Received: 08 August 2016
  •  Accepted: 19 September 2016
  •  Published: 29 September 2016

Abstract

The objective of this study was to evaluate the effect of input vectors in an artificial neural network (ANN) and determine their best combination to estimate the individual dry biomass of native bracatinga. The dataset consisted of 178 trees of Mimosa scabrella Benth. (bracatinga) from the Metropolitan Region of Curitiba. The ANN used was a Multi-Layer Perceptron; the learning algorithm was the Levenberg-Marquardt, consisting of an occult layer where 50% of the data were used for training, 25% for cross-validation and the other 25% for the test. The input vectors were all the variables collected in the field, such as: diameter at breast height (dbh), total height (ht), crown height (hc), stem height (hf), crown diameter (dc) and age (i). The treatment 1 consisted of all the vectors; after the MLP trained, the Garson algorithm was executed for obtaining relative contribution of each vector; the less important vector was deleted and the MPL was retrained (treatment 2) and so on until only one vector was left. Based on the coefficient of determination and root mean square error, treatment 3 provided the best performance (i, hc, ht and dbh), followed by treatment 6 (dbh). The method of selecting attributes by the Garson algorithm was remarkable and provided the definition of essential vectors, allowing minimal costs and optimizing the performance of the MLP.

 

Key words: Bracatinga, multi-layer perceptron, Garson algorithm, relative contribution.