Hybrid model of artificial neural networks and principal component decomposition for predicting greenhouse gas emissions in the brazilian region of MATOPIBA

Abstract

Greenhouse gas (GHG) emissions in agricultural production represent a global environmental challenge, and it is necessary to understand the factors that influence them to develop sustainable practices. The general objective of this research is to investigate some of the factors that probably influence GHG emissions and reductions in agricultural production in the MATOPIBA region of Brazil between 2006 and 2017. A hybrid methodology was used, and the first stage used linear models (decomposition into principal components) and non-linear models (artificial neural networks) to determine the relationships that should exist between the dependent variable (GHG emissions) and 11 variables. The data was obtained from the 2006 and 2017 Brazilian Agricultural Census, MapBiomas, SEEG, and NOAA. The results showed that of the 373 municipalities that make up MATOPIBA, only 100 did not see an increase in GHG emissions between 2006 and 2017. The principal component decomposition method reduced the 11 initial variables into 3 orthogonal and unobserved variables. In one of the unobserved variables, 4 of the five variables that are supposed to cause a reduction in GHG emissions were brought together. The 5 variables thought to have caused an increase in GHG emissions were condensed into 5.