Object based wetland mapping in Newfoundland and Labrador using synthetic aperture RADAR (SAR) and optical data

• Main Author: Mahdavi, Sahel
• Format: Thesis
• Language: English
• Published: Memorial University of Newfoundland 2018
• Subjects: Newfoundland
• Online Access: https://research.library.mun.ca/13607/; https://research.library.mun.ca/13607/1/thesis.pdf

Wetlands are amongst the most valuable natural resources that provide many advantages to the ecosystem and humans. Therefore, their mapping and monitoring is crucial. In today’s dynamic world, where vast areas require observation with increasing frequency, remote sensing is an accessible, cost effective way of environmental monitoring. This thesis proposes novel remote sensing methods for mapping and monitoring wetlands and other complicated land covers and facilitates this by proposing alternative pre-processing or post-processing techniques. In Chapter 2, a comprehensive literature review was conducted that elaborates on different aspects of wetland studies. Various methods for wetland classification, along with the benefits and limitations of each, were provided, and areas which could be improved were highlighted. In Chapter 3, an innovative filter was proposed for reducing speckle in Synthetic Aperture RADAR (SAR) images, which is considered an important pre-processing step for land cover classification using SAR data. The proposed filter applies window sizes to each pixel based on the size of the object in which the pixel is placed. The filter was applied to two simulated and two real SAR images in both single-channel and full-polarimetric cases, and the filter results were comparable to several state-of-the- art filters. In Chapter 4, wetlands in four pilot sites within Newfoundland and Labrador were classified using multi-temporal RADARSAT-2 imagery by applying the proposed method for segmentation of SAR images. The covariance matrix was found to be a valuable feature, although textural and ratio features slightly increased the overall accuracy of wetland mapping. Furthermore, August was determined to be the best month for wetland classification. In Chapter 5, an innovative dynamic classification scheme was proposed for mapping complicated land covers. In this method, objects are not assigned labels simultaneously, but different classes are mapped using a separate feature selection and classification. The ...