Spatial modelling of deforestation in Romanian Carpathian Mountains using GIS and Logistic Regression

Deforestation process represents a wide concern mainly in the mountain environments due to its role in global warming, biodiversity loss, land degradation and natural hazards occurrence. Thus, the present study is focused on the largest afforested landform unit of Romania and, consequently, the most affected area by forest losses: Carpathian Mountains. The main goal of the paper is to examine and analyze the various explanatory variables associated with deforestation process and to model the probability of deforestation using GIS spatial analysis and logistic regression. The forest cover for 1990 and 2012, derived from CORINE Land Cover (CLC) database, were used to quantify historical forest cover change included in the modelling. To explain the biophysical and anthropogenic effects, this study considered several explanatory factors related to local topography, forest cover pattern, accessibility, urban growth and population density. Using ROC (Receiver Operating Characteristic) and 500 controlling sampling points, the statistical and spatial validations were assessed in order to evaluate the performance of the resulted data. The analysis showed that the area experienced a continuous forest cover change, leading to the loss of over 250,000 ha of forested area during the period 1990–2012. The most significant influence of the explanatory factors of deforestation were noticed in case of distance to forest edge (β=−4.215), forest fragmentation (β=2.231), slope declivity (β=−1.901), elevation (β=1.734) and distance to roads (β=−1.713). The statistical and spatial validation indicates a good accuracy of the model with reasonably AUC (0.736) and Kappa (0.739) values. The model’s results suggest an intensification of the deforestation process in the area, designing numerous new clusters with high probability in the Apuseni Mountains, northern and central part of the Eastern Carpathians, western part of the Southern Carpathians and northern part of the Banat Mountains. The study could represent a useful outcome to identify the forests more vulnerable to logging and to adopt appropriate policies and decisions in forest management and conservation. In addition, the resulted probability map could be used in other studies in order to investigate potential environmental implications (e.g. geomorphological hazards or impact on biodiversity and landscape diversity).