Detecting and Mapping Forest Loss using Satellite Image Classification and Spatial Analysis Case Study: Tabarka and Ain Draham
Abstract
This study explores forest loss detection and mapping in the Tabarka and Ain Draham regions from 2016 to 2022 using satellite imagery and spatial analysis methods. A supervised classification technique employing the K-Nearest Neighbor algorithm $(K-N N)$ was used to classify forest and non forest areas. The classification process involved selecting representative training samples from both forest and non-forest zones to classify pixels based on their similarities to labeled data. Forest loss was detected by analyzing the transition of pixels from ‘forest’ to ‘nonforest’ between consecutive years. Any pixel classified as ‘forest’ in one year but ‘nonforest’ in the next was considered a forest loss. Classification results were validated based on Recall, User’s Accuracy (UAccuracy) and Kappa index ensuring reliable detection of forest cover changes over time. Beyond classification, spatial pattern mining techniques were applied to uncover underlying factors contributing to forest loss. Specifically, the Apriori algorithm was used to extract frequent spatial patterns associated with deforestation analyzing combinations of environmental and anthropogenic variables such as slope, proximity to water bodies, dams, roads and villages. The use of Geographic Information Systems (GIS) was crucial for visualizing and analyzing geospatial data.