The loss of tropical forests is responsible for 8% of global CO2 emissions. As an indirect result of the Peace Process, deforestation in Colombia has been growing alarmingly: in 2017 (220 thousand hectares) it was 23% higher than in 2016. Total forest coverage has decreased from 57% in 1990 to 52% in 2017. It is estimated that 27% (69Mton of CO2 equivalent) of national greenhouse gas emissions in 2012 were caused by deforestation. In this case, the activity, that is, the deforested area, is the controlling factor for the uncertainty in the emission estimate. The thematic accuracy assessment is a necessary input for the estimation of central values and confidence intervals for the activity data. This research has sought to improve the estimation of the confidence intervals and reduce the standard errors. For this, three different approaches to statistical inferences have been developed that are used as a generalization tool that from the data of a sample can produce estimates of the population parameters, which are: probability sampling (based on the design), mathematical models (model-based) or a mixture of the two above known as hybrid inference. Our approach includes the design-based approach, analyzing different scenarios of sampling design, response and estimation of statistical parameters. This study estimates the thematic error associated with deforestation in a region of the PNN (Natural National Park) Sierra de la Macarena. This article will present the methodological details, including new sampling strategies and their potential impact on thematic uncertainty. The results that are presented with the best performance were from stratified sampling, since it has a lower uncertainty and greater accuracy with respect to the other sampling designs.