Redes neurais artificiais para a modelagem do volume de madeira e biomassa do cerradão com dados de satélite

Abstract

Resumo -O objetivo deste trabalho foi avaliar a eficácia da aplicação de modelos de análise de regressão e redes neurais artificiais (RNAs) na predição do volume de madeira e da biomassa acima do solo, da vegetação arbórea em área de cerradão. Volume de madeira e biomassa foram estimados com equações alométricas desenvolvidas para a área de estudo. Os índices de vegetação, como variáveis preditoras, foram estimados a partir de imagens do sensor LISS-III, e a área basal foi determinada por medições na floresta. A precisão das equações foi verificada pela correlação entre os valores estimados e observados (r), erro-padrão da estimativa (Syx) e gráfico residual. As equações de regressão para o volume de madeira total e do fuste (0,96 e 0,97 para r, e 11,92 e 9,72% para Syx, respectivamente) e para a biomassa (0,91 e 0,92 para r, e 22,73 e 16,80% para Syx, respectivamente) apresentaram bons ajustes. As redes neurais também apresentaram bom ajuste com o volume de madeira (0,99 e 0,99 para r, e 4,93 e 4,83% para Syx) e a biomassa (0,97 e 0,98 r, e 8,92 e 7,96% para Syx, respectivamente). A área basal e os índices de vegetação foram eficazes na estimativa do volume de madeira e biomassa para o cerradão. Os valores reais de volume de madeira e biomassa não diferiram estatisticamente dos valores estimados pelos modelos de regressão e redes neurais (χ² ns ); contudo, as RNAs são mais acuradas.Termos para indexação: índice de vegetação, inventário florestal, produção, regressão, sensoriamento remoto. Artificial neural networks for modeling wood volume and aboveground biomass of tall Cerrado using satellite dataAbstract -The objective of this work was to evaluate the effectiveness of regression models and artificial neural networks (ANNs) in predicting wood volume and aboveground biomass of arboreal vegetation in area of tall Cerrado (a forest, savanna-like vegetation). Wood volume and biomass were estimated with allometric equations developed for the studied area. The vegetation indices, as predictor variables, were estimated from LISS-III sensor imagery, and the basal area was determined from field measurements. Equation precision was verified by the correlation between estimated and observed values (r), standard error of estimate (Syx), and by the residual plot. The regression equations for total wood volume and bole volume (0.96 and 0.97 for r, and 11.92 and 9.72% for Syx, respectively), as well as for aboveground biomass (0.91 and 0.92 for r, and 22.73 and 16.80% for Syx, respectively) showed good adjustments. The neural networks also showed good adjustments for both wood volume (0.99 and 0,99 for r, and 4.93 and 4.83% for Syx) and biomass (0.97 and 0.98 for r, and 8.92 and 7.96% for Syx, respectively). Basal area and vegetation indices were effective in estimating wood volume and biomass for the tall cerrado vegetation. Measured wood volume and aboveground biomass did not differ statistically from the predicted values by both the regression models and neural networks (χ² ns ); however, the ANNs are more accurate.

Artificial neural networks in whole-stand level modeling of Eucalyptus plants

“…Miguel et al (2015) found similar results when testing ANNs, in which the aggregate difference presented positive values for total wood volume and stem volume (3.41 and 8.58%, respectively).…”

Morphometric indexes and dendrometric measures for classification of forest sites of Eucalyptus urophylla stands

“…According to Miguel et al (2015), variables which present Ei values close to zero demonstrate a better capacity to perform the desired estimate accurately. Such behavior occurs with the proposed variables, indicating their good suitability for the prognosis of volumetric production.…”

Wood Volume Estimation in a Semidecidual Seasonal Forest Using MSI and SRTM Data

“…In order to optimize forest activities, new techniques such as the use of satellite images have been employed in the prediction models to dendrometric variables as height, basal area and wood volume (Almeida et al, 2014), and they have been a support to get information from areas of difficult access in an easy way. Besides that, it is important to highlight the advantage of financial viability when using the method (Watzlawick et al, 2009;Miguel et al, 2015).…”