Balanço de energia em ecossistema amazônico por modelo de regressão robusta com bootstrap e validação cruzada

Abstract

The energy balance closure is often analyzed as a means of assessing the quality of energy fluxes measurements obtained by the eddy covariance method, and indirectly as a way to assess the accuracy of carbon dioxide measurements. In the attempt to evaluate the efficiency of this method, several researchers used a simple linear regression model between the components of energy balance. However, these studies have not verified the appropriateness of the statistical assumptions of regression analysis, which can lead to erroneous conclusions if the model is not satisfactory. Thus, the aim of this study was to calibrate and validate simple linear and robust regression with bootstrap and cross-validation to verify the efficiency of the energy balance closure in the Amazon ecosystem. Measurements of net radiation, latent heat flux, sensible heat flux and soil heat flux were made from January to December 2008 in an area of tropical rain forest in southwest Amazonia, in an experimental site belongs to LBA Program’s towers. The results showed that the simple linear regression model is not suitable to analyze the energy balance closure and both the application of robust linear regression model with bootstrap and cross-validation improved the adjust.