A Combination of ALOS-2, Sentinel-1 Imagery for Rapid Deforestation Detection in Vietnam

Abstract

Monitoring rapid forest loss using optical remote sensing presents significant challenges, particularly in tropical regions due to persistent cloud cover. Therefore, the use of Synthetic Aperture Radar (SAR) imagery offers a promising alternative. To provide timely deforestation information, improvements in monitoring frequency and the integration of various synthetic aperture radar (SAR) imagery have been implemented, enhancing effectiveness compared to single sensors. The wavelengths of SAR data influence their ability to penetrate canopies, leading to different scattering mechanisms and varied information about forest cover. Specifically, the C-band wavelength (~5 cm) captures signals from the canopy and small branches, while the L-band wavelength (~23 cm) captures signals from tree branches.

This paper focuses on combining two different types of SAR data (C-band and L-band) along with HV and HH polarizations to detect clear-cut and forest fires. The method employed involves comparing backscatter values before and after deforestation to identify forest loss. Sentinel-1 time series data has been analysed using the Radar Change Ratio (RCR) method, while three ALOS-2 image scenes have been processed using the RGB composite method. It has been demonstrated that C-band SAR data (Sentinel-1) can detect deforestation due to clear-cutting in Ha Long with monthly frequency, although its effectiveness in identifying forest fire areas has been limited. In contrast, L-band data (ALOS-2) has been shown to be capable of detecting various types of deforestation, including clear-cutting and forest fires. Within the C-band data, HV polarization yields better results than HH, whereas the L-band data provides similar outcomes irrespective of whether HV or HH polarization is applied. The integration of C-band and L-band SAR data provides more comprehensive information on deforested areas, improving the accuracy of backscatter-based detection methods.