Combination of optical and SAR remote sensing data for wetland mapping and monitoring
Wetlands provide many services to the environment and humans. They play a pivotal role in water quality, climate change, as well as carbon and hydrological cycles. Wetlands are environmental health indicators because of their contributions to plant and animal habitats. While a large portion of Newfo...
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
Memorial University of Newfoundland
2018
|
| Subjects: | |
| Online Access: | https://research.library.mun.ca/13604/ https://research.library.mun.ca/13604/1/thesis.pdf |
| Summary: | Wetlands provide many services to the environment and humans. They play a pivotal role in water quality, climate change, as well as carbon and hydrological cycles. Wetlands are environmental health indicators because of their contributions to plant and animal habitats. While a large portion of Newfoundland and Labrador (NL) is covered by wetlands, no significant efforts had been conducted to identify and monitor these valuable environments when I initiated this project. At that time, there were only two small areas in NL that had been classified using basic Remote Sensing (RS) methods with low accuracies. There was an immediate need to develop new methods for conserving and managing these vital resources using up-to-date maps of wetland distributions. In this thesis, object- and pixel-based classification methods were compared to show the high potential of the former method when medium or high spatial resolution imagery were used to classify wetlands. The maps produced using several classification algorithms were also compared to select the optimum classifier for future experiments. Moreover, a novel Multiple Classifier System (MCS), which combined several algorithms, was proposed to increase the classification accuracy of complex and similar land covers, such as wetlands. Landsat-8 images captured in different months were also investigated to select the time, for which wetlands had the highest separability using the Random Forest (RF) algorithm. Additionally, various spectral, polarimetric, texture, and ratio features extracted from multi-source optical and Synthetic Aperture Radar (SAR) data were assessed to select the most effective features for discriminating wetland classes. The methods developed during this dissertation were validated in five study areas to show their effectiveness. Finally, in collaboration with a team, a website (http://nlwetlands.ca/) and a software package were developed (named the Advanced Remote Sensing Lab (ARSeL)) to automatically preprocess optical/SAR data and classify wetlands ... |
|---|