Distribuição espaço-temporal do número e do valor de autos de infração ambiental em Minas Gerais de 2008 a 2016

Abstract

The aim of this thesis is to present statistical modelling for number and aggregate value of environmental infraction notices that happen in Minas Gerais from 2008 to 2016. Such important data at municipality basis lack integrated analysis. We implemented Bayesian space-time modelling with different likelihood functions: binomial for presence/absence of notices, poisson for the number of notices in the year and gamma for the aggregate value of the fines in the year. This and other candidate likelihood functions were initially fitted to one-dimensional data and joint likelihood was built with best fitting models. Fitting was carried out using INLA (R-INLA package). This allowed complex joint modelling was performed in about 10 hours of computing time using a 24 cores machine. Time trends of increasing number of notices and fine values are evidence of lack of efficiency in law enforcement on environmental infractions in MG. This model can be adapted to similar cases. For example, in the fields of science: ecological, economic and epidemiological.