Polarimetric characteristics of early-stage deforestation areas were examined with L-band synthetic aperture radar (SAR) to develop a forest early warning system. Time series of PALSAR-2/ScanSAR data and Landsat data were used to examine the differences in detection timing of deforestation in the most active deforestation sites in Peru and Brazil. The γ 0 H H value increased by 0.8 dB on average for areas undergoing early-stages of deforestation, in which felled trees were left on the ground. The detection timing was almost the same as that of using the optical sensor. On the other hand, the γ 0 H V value does not show significant γ 0 change at this early-stage of deforestation. The γ 0 H V value started to decrease 1-1.5 months after the deforestation was detected by Landsat. The γ 0 H V value decreased by 1.5-1.6 dB, 4-5 months after the deforestation. To understand the radar backscattering mechanism at the early-stage deforestation sites, field experiments were carried out 2-16 days after the PALSAR-2/fullypolarimetric observations. The early-stage deforestation sites revealed 1.1-2.5 dB and 2.9-4.0 dB increases for σ 0 H H and σ 0 su rface , respectively. This can be explained by the direct (single bounce) scattering from felled trees left on the ground. The sites in which felled trees were removed and the surfaces were flattened showed 5.2-5.3 dB and 5.3-5.5 dB decreases for σ 0 H V and σ 0 vo lu m e , respectively. This can be explained by the lower sensitivity of the HV polarization to both the branches remaining on the ground and the surface roughness, along with its increased sensitivity to the forest biomass. We conclude that an increase in L-band γ 0 H H is a good indicator for detecting early-stage deforestation sites, where felled trees are left on the ground, while γ 0 H V may be useful for detecting later-stage deforestation sites.

No abstract

No abstract

No abstract

Accessibility to cloud-free optical sensor images is essential for large-area monitoring of land and forest cover changes. In this study, the acquisition probabilities of cloud-free images were analyzed using MODIS cloud mask products from 2000 to 2008 in Southeast Asia. The daily cloud masks were summarized into monthly acquisition probabilities for cloud-free images over the period at a spatial resolution of 1km. The mean annual acquisition probability profiles were extracted averaging nine years' observation. Unsupervised clustering was conducted for zoning of the acquisition probabilities using the mean annual profiles. Annual variations in the acquisition probabilities were examined by the standard deviations calculated for each month and comparisons of the mean annual profiles of the whole period and individual years. The distributions of annual acquisition probabilities in forested areas were different in each country. These results suggested that selection of suitable methods and data allowing for the spatial and temporal differences in the acquisition probabilities is necessary for periodic large-area monitoring.