Please use this identifier to cite or link to this item:
|Title:||Synergistic use of airborne hyperspectral and LiDAR data for mapping Mediterranean forest in Portugal|
|Authors:||Pandey, Prem Chandra|
|Presented at:||University of Leicester|
|Abstract:||Forests are the major source of biodiversity and provide natural sources of wood, fodder, gums, resins, and medicines. Forests encounter damage by nature and human factors, which needs to be monitored for all tree species, whether invasion or intentional damage. This study focuses on the classification of an open tall stand coastal surrounding site for the mapping and classification of tree species and ground features using airborne imagery. So, improving the classification and mapping accuracy of forest in surrounding coastal regions is essential for the restoration and management decisions. The first objective of this thesis is to use segmented Principal Component (PC) images to classify the ground features including different tree species and to improve the classification results. More specific goals include (a) Use of hyperspectral images to map and classify the forest region using a segmented PC image, (b) Investigating the gain in mapping accuracy with segmented PC image as opposed to hyperspectral imagery alone. The second objective is to assess and investigate the fusion of airborne hyperspectral imagery and LiDAR derived Canopy Height Model for classification and assessing the results. These objectives aim at investigating the gain in mapping accuracy with fusion image as opposed to hyperspectral imagery alone. Thus, overall this study assesses the differences in classification outputs using a data fusion technique, segmented PC image and individual hyperspectral images, which differ in accuracy, in Mediterranean forest. MLC based supervised image classification method provided better accuracy (96.3%) with segmented PC images, (~92.9%) with the fusion of CHM and hyperspectral images than with hyperspectral image alone (89.6% with MLC and 67.5% with SAM). According to my results, CHM and HSI provide better classification and mapping results over extensive areas of forests. The overall accuracy of the classified maps ranged from 67.5 to 96.3% and k coefficient was found between 0.61 and 0.95. Segmented PC and PC fusion techniques provided a significant step to improve the distinction and classification results. Using the above methods, tree species and associated features could be classified and mapped, despite the problem of spectral mixing of different features. In future, more high spatial and spectral resolution images will provide a platform for the incorporation of enhanced characteristics for mapping and classification purposes.|
|Rights:||Copyright © the author. All rights reserved.|
|Appears in Collections:||Theses, Dept. of Geography|
Items in LRA are protected by copyright, with all rights reserved, unless otherwise indicated.