Direction-dominated change vector analysis for forest change detection

Abstract

As forest is under increasing pressure, the rapid conversion or degradation of forest has attracted strong concern. Obtaining quantitative information of forest change based on satellite imagery becomes necessary and urgent, especially the detailed “from-to” information. In this study, a semi-automatic method called direction-dominated change vector analysis (DCVA) was proposed to detect “from-to” information of forest change. DCVA is composed of three steps: (1) determining candidate changed pixels, (2) determining direction ranges for different forest change types, and (3) determining final changed pixels for each forest change type. Like the classic change vector analysis (CVA), the magnitude and direction of change vector (CV) are used to detect the changed areas and types in DCVA, respectively. However, CVA is “magnitude-dominated” by setting only one magnitude threshold for different change types, while DCVA is “direction-dominated” by determining change types according to change direction at first, followed by setting different magnitude threshold for each change type. In this case, DCVA holds the advantage of accurately detecting changed areas for different change types by considering the specific characteristics of each change type. Experiments are performed with Sentinel-2A satellite images to demonstrate the advantages of DCVA for forest change detection. The changed areas with four types of forest change were successfully extracted by DCVA. The comparison of both geometric and thematic accuracies between DCVA and CVA further indicates the effectiveness of the proposed method for forest change detection.

Keywords

• Forest
• Land cover change
• Change detection
• Change vector analysis