Multi-period monitoring and prediction of forest cover loss using logistic regression model in Arasbaran catchments

Knowledge and understanding of changes in forest cover in relation to environmental factors (topography) can be valuable in terms of conservational and protective guidances. The purpose of this study was to identify, quantify and predict deforestation in relation to topographic variables using logistic regression model. The Arasbaran catchments (Naposhtehchay, Ilginehchay and Mardanqumchay) in northwestern Iran were targeted because of their valuable forest ecosystems. In this regard, the reduction of forest cover (dependent variable) over three different time periods including 1984-1996, 1996-2006, 2006-2017, was detected using Landsat satellite imagery and post- classification method. On the other hand, 8 independent topographic variables were derived from digital elevation model (DEM) with 30 m resolution including altitude, slope, topographic position index (TPI), topographic wetness index (TWI), northness, eastness, plan curvature and profile curvature. The results of change detection over the three time periods was indicative of increasing trend of deforestation in Ilgineh and Mardanqum catchments, and in contrast, the declining trend of it in Naposhteh catchment. Also, logistic regression models were satisfactory and acceptable in terms of performance to explain and predicting the deforestation in relation to topographic variables. The ROC values ​​obtained in the three time periods were 0.76, 0.72 and 0.64, respectively. The modeling showed that the highest probability of deforestation was attributed to convex areas and ridges, high altitudes, valleys, steep slopes, south and east aspects, and wetlands.