Avaliação da percepção às secas usando técnicas de aprendizagem de máquina

Abstract

Drought management gains emphasis on the scenario of irrigated agriculture, an activity of considerable water vulnerability. Thus, the present study proposes the understanding of drought perception of irrigants and the identification of adaptation strategies, in order to support decision making in the water allocation process. The study was carried out in the irrigated perimeter Tabuleiro de Russas, in the interior of Ceará. For this, we sought to identify the characteristics, causes and consequences of droughts through the perception of irrigants. The profile of the interviewees was identified, as well as their perception of the definition of drought and its socioeconomic impacts. It was found the vision of drought for the irrigators based more on meteorological than hydrological characteristics. In addition, among the main impacts of the last drought on local farmers, it was identified that 97% of these had their crops broken, 93% reduced their profit and 20% had to relocate to other activities in the water crisis period. Using two machine learning techniques, the Random Forest and Pearson Correlation methods, applied to the R language, were selected the variables of most importance and correlation to the process of perception of users. The methods selected six of the nine variables tested (Number of active lots; Time spent working in agriculture; Drought reason; Age of respondent; Level of education; Participation in discussion groups about drought, in decreasing order of importance). Using this language, and by means of statistical computational simulation, a predictive model for linear perception of drought was developed to quantify and identify the level of perception of the perimeter, classifying from low to high level. The developed model was validated, with an error percentage of only 3.33% for the observed data. Perimeter perception was considered high. It was also evaluated the dissatisfaction of users and the need for improvement in the communication system between managers, administrators and irrigators. Therefore, research on users' perceptions should be encouraged, as their incorporation into the water allocation process can be beneficial in reducing conflicts and adopting proactive management practices, as well as developing a holistic and efficient early warning system for risk.