Assessing the Dynamics of a South Mediterranean Dryland-Type Forest Kind by Logistic Regression and Cellular Automata

Forest degradation is a severe problem that threatens forest resources around the world. The south Mediterranean forests are dryland types of forests prospering in arid and semi-arid climates, making them more vulnerable to degradation. This work aims to present a methodology that combines three approaches: geographical information systems (GIS), machine learning, and cellular automata, to assess the spatiotemporal behavior of south Mediterranean dryland-type forests (case study, the Ain Leuh Forest, Middle Atlas, Morocco). First, we used satellite images to establish forest state maps defining five classes: dense cedar cover (C1), low cedar cover (C2), dense Quercus cover (Q1), low Quercus cover (Q2), and bare soil (BS). Then, based on those historical maps, two scenarios were defined, P1: 1990–2000 and P2: 2000–2010, and the relative class transitions. Second, a logistic regression algorithm was used to create the probability transition maps. The model considered four factors (altitude, slope, aspect, and distance from the forest edge). Third, a cellular automata model was implemented to predict the forest's future state. The model takes into input the initial state of the forest, and the probability of transition drives it. Our model was first calibrated by predicting the 2019 forest state, evaluated by Kappa de Cohen and cosine distance coefficients, then projected to predict the 2030 forest state. Results showed the efficiency of combining logistic regression and cellular automata models to predict the form (density of area) and the rate of south Mediterranean dryland forest degradation.