SATELLITE-DERIVED VEGETATION INDICES AND AN OBJECT-BASED ANALYSIS OF FOREST CONDITIONS IN SEMI-ARID TROPICS

摘要

Changes in forest conditions are influenced by climate and human activities. Variability in the quantity, quality and space of forest conditions is needed for management. The study aims to determine the forest conditions based on an integration of satellite-derived vegetation indices and object- based analysis using Landsat 8 OLI data. The NDVI and NDWI provide the biomass and vegetation water content when combining the landscape-vegetation pattern using the object-based analysis. We can differentiate the flourishing forest from various degree of degraded forest. Our study area, a part of the Phu Khieo-Nam Nao forest complex, covers an area of approximately 2,500 Km~2 and is located in the upper Thailand which has semi-arid climate. Landsat 8 data acquired on January 4, 2015 was geometrically and atmospherically corrected. The image was segmented into image objects, optimizing scale, shape and compactness using the eCognition Developer software package. The supervised classification was performed through the image segmentation using Nearest Neighbor decision algorithm. The Normalized Difference Vegetation Index (NDVI) and Normalized Difference Water Index (NDWI) were digitally performed and combined into one image to categorize the difference in greenness and vegetation wetness. The classified image obtained from the supervised classification was further analyzed to group the segments with respect to the relevant NDVI-NDWI images. As a result, the different forest conditions were classified with a total of 8 classes. The image output was checked against a number of ground truths for validation. The reliability with high Kappa coefficient was confirmed the result. The 8 classes are flourish forest, slightly degraded forest, moderately degraded forest, highly degraded forest, forest plantation, agriculture, resettlement and water body, accounting for 15.54%, 11.89%, 17.00%, 15.43%, 3.46%, 31.70%, 4.04%, and 0.94% respectively. The study indicates that an integration of the object-based analysis and satellite-derived indices yielded significant degree of the forest conditions and its distribution.