Forest clearance monitoring with dual polarized SAR data in the peatlands of insular Southeast Asia

Abstract

© 2020 40th Asian Conference on Remote Sensing, ACRS 2019: "Progress of Remote Sensing Technology for Smart Future". All rights reserved. The natural forests in the peatland of insular Southeast Asia have experienced dramatic deforestation since 1990 by both accidental fire events and deliberate land conversion into cultivated areas. Up-to-date monitoring of the forest clearance is important for sustainable peatland management. Due to very cloudy conditions in the insular Southeast Asia, practical monitoring of forest clearance activities based on optical data are greatly hampered by severe scarcity of cloud free images. With its cloud penetrating capability and relatively high revisiting frequency (12 days repeat cycle), Sentinel-1 synthetic aperture radar (SAR) offers a tool for land cover change monitoring in the cloudy conditions of Southeast Asia. In this paper we present an approach to monitor forest clearance in peatland areas of insular Southeast Asia with Sentinel-1 dual polarized synthetic aperture radar (SAR) data taking advantage of the Google Earth Engine image database and processing capabilities. With this approach, change detection can be performed on a monthly basis. Sample data representing changes caused by deforestation and plantation harvesting in the VV and VH backscatter feature space of Sentinel-1 C-band SAR were collected to obtain the backscattering transition thresholds for forest clearance. These threshold values were used to flag out the potential forest clearance areas in two consecutive time steps. These potential forest clearance areas were then overlaid on a peatland land cover base map crated based on 2015 satellite imageries. All potential change areas intersecting with peat swamp forest or industrial plantations in the base map were marked as forest clearance area. The combination of up-to-date SAR data together with existing forest base map allows interpretation of the detected changes as either deforestation or plantation harvesting. The procedures were run for Sentinel-1 images in 2018 on monthly intervals. We also evaluated the accuracy of the monthly forest clearance in 2018 using Sentinel -2 optical images. The commission error is 22% and omission error 26% for the forest clearance detected by Sentinel-1 images from August 2018 to September 2019 by use of Sentinel-2 data as ground truth.