Localização de áreas susceptíveis à diversificação utilizando autômatos celulares em Gilbués-PI

Abstract

Desertification mainly affects hyperarid, arid, semi-arid and dry subhumid regions, resulting from a combination of anthropogenic actions and climate change. It is a global challenge, since it directly affects food security, implies mass migration and conflicts over water resources, fostering humanitarian crises. The need for a genuine methodology justifies the proposal of this work, which has as main objective to present a model formed by the combination between Cellular Automata, Geographic Information Systems and Remote Sensing, appropriate to model the advance of desertification in areas vulnerable to the phenomenon and in the environment of already desertified land. The municipality of Gilbués, in Piauí, was chosen as a study area to test the method, a decision that was justified in the fact that the site presents accelerated degradation dynamics and is the largest area in the process of desertification in Brazil. Cartographic data from the Brazilian Institute of Geography and Statistics (IBGE) were used for the mapping of the location. For the use and coverage of the land we used images of 2005 and 2010 of the satellite LANDSAT 5, sensor TM, and 2015 of LANDSAT 8, OLI sensor. Seven classes of land use/land cover were defined, which were classified in a supervised way using Random Forest algorithm. Land use changes between 2005 and 2010 were calculated using the Markov chain, assuming a stationary transition matrix. The Cellular Automata used the Markov chain as a reference to carry out a projection from 2010 to 2015, adding the geographic reference to the process, which was compared with the field truth represented by the real reference satellite image. The accuracy of both classification and projection was assessed using Kappa Index. The relationship of susceptibility of the classes of use and land cover to the desertification process was defined by means of the change to the dunes class. The projections for 2020 and 2025 were calculated using the transition matrix from 2005 to 2010 as a reference. According to the Kappa value,the thematic classification was excellent. The proposed model was vii able to generate a good prediction, showing that the dunes are expanding to the center-south region of the municipality, where the exposed soil is concentrated and the urban area is present. The analysis Markov chain’s results showed that the greatest probabilities of transition between land use and land cover classes are related to one class remaining unchanged. The future projection showed that the evolution of class changes is stabilizing. Information obtained on land use and land cover serves as a basis for decision-making on preventive action in the areas on alert