Uso de componentes de imagens de satélites na modelagem espacial do volume em povoamento de Eucalyptus sp.

Abstract

Forest inventory is an important tool used to estimate forest wood production. However, some methodologies used in forest inventory are based in Classical Statistics, which disregards any spatial continuity that may exist between sample unities. Some geostatistic interpolators such as ordinary kriging (OK) and external drift kriging (EDK) allow us to assess this spatial structure. Furthermore, besides spatial variability, interpolators as EDK use one or more auxiliary variables. Satellite images have different components that interrelate with dendrometric variables and that can be used as auxiliary variables in order to increase the degree of precision of estimates. The aim of this study was to assess EDK performance on the volume estimation of Eucalyptus sp. stands using satellite image components as secondary variables and to compare it with OK performance. With this purpose, a forest inventory of 210 circular plots of 500 m² was carried out in order to estimate the volume (m³ ha-1 ) in each plot. Images obtained of studied area had blue, green, red and near infrared band. From these bands, it were extracted: gray level in each band, the ratio between bands, vegetation index (NDVI, SAVI e ARVI), texture measures and index generated from textures related to plot area. Covariance model adjustement throughout Stepwise method and selection by AIC (Akaike Information Criterion) method were made to EDK geostatistic. EDK and OK semivariograms were adjusted by different theoretical models through Ordinary Least Squares (OLS) method and the choice of the best model was given by the lowest value of residual standard error. From statistic analysis of images and correlation matrix, it was observed a correlation of variables with volume and also autocorrelation between these variables. The best covariance model selected was composed by band 2, measure of COR texture of band 2, MULCOR texture index of band 1 and by age. In the two semivariograms, the best model adjusted was the exponential one. Analysing the results, volume estimates generated by EDK produced better results than OK estimates and had the lowest value of residual standard error and the best area under curve (AUC) in receiver operating characteristic (ROC) curve analysis.