Analyzing Cloud-contaminated Remotely-sensed Images to Detect and Explain Deforestation in Two Mindanao Provinces

This paper presents an approach in extracting multi-temporal land-cover information (years 1976-2001) from cloud-contaminated remotely-sensed images acquired by the Landsat Multi-Spectral Scanner (MSS) and Enhanced Thematic Mapper plus (ETM+) sensors in order to detect and analyze deforestation and other types of land-cover change (LCC) in two forest resource-rich provinces of Agusan del Norte (ADN) and Agusan del Sur (ADS) in Mindanao, Philippines. The cloud contamination problem was addressed by developing a cloud-and-shadow masking algorithm comprising of image segmentation and Maximum Likelihood classification to remove clouds and shadows which account for 40-50% of the images. Forest-cover change from 1976-2001 were then detected using state-of-the-art Support Vector Machine classifier, and a post-classification change detection in portions of the land-cover maps un-contaminated by clouds and shadows. Using spatial analysis provided by a Geographic Information System (GIS) and logistic regression statistical analysis, deforestation in ADN and ADS were characterized with respect to bio-physical and socio-economic factors to determine the significance and magnitude of the relationship between the detected deforestation and the various factors. Major results showed that deforestation and other types of LCC can be detected, characterized and analyzed from cloud-contaminated Landsat images using the integrated image cloud masking, SVM classification, GIS and logistic regression approach. This study is a significant contribution to LCC research by providing a series of techniques to understand deforestation based on cloudcontaminated Landsat images and to associate it to bio-physical and socio-economic factors.