Abstract
Recent advances in satellite observation have allowed for increasingly detailed monitoring of the Earth’s surface, providing the opportunity to capture the complex seasonal dynamics of many Land Cover (LC) types and to detect LC change more rapidly than ever before. In addition, capacity for large-scale LC monitoring applications is constantly increasing due to advances in computing capability. As a result there has been an increase in methods that monitor LC change on a per-acquisition rather than a yearly basis, such as Continuous Change Detection and Classification (CCDC). This thesis utilises the Landsat data record to investigate and apply a time series modelling method to the problem of mangroveforest change. Mangroves are of high biological, economic, and ecological importance globally and are particularly vulnerable to the effects of climate change in addition to being threatened by over-exploitation, pollution, and expansion of aquaculture practices. Long-term monitoring of global mangrove populations is therefore vital, and impossible to achieve without repeated satellite imagery. However, data availability from satellites for tropical and sub-tropical areas is often limited due to cloud cover, and the dynamic nature of mangrove ecosystems introduces uncertainty for traditional monitoring approaches. These problems can be mitigated using a season-trend modelling approach to utilise every available observation on a pixel-wise basis, accounting for ephemeral change and removing the need for whole cloud free images. A methodology for generating simulated optical time series is developed and used to objectively assess the ability of four algorithms to detect both the timing and nature of different LC change types. Based on this assessment, CCDC was chosen as being the most suitable method for monitoring of mangrove ecosystems due to its robustness to missing data
and low commission error. The CCDC algorithm was then implemented and applied within a High Performance Computing framework. Using CCDC, yearly class maps were generated for six study sites for the last 30 years, with the method achieving an overall classification accuracy >90% and providing the most comprehensive assessment yet of mangrove extent in the Sundarbans and Niger Delta regions. Results showed that while mangrove extent in the Sundarbans has remained stable, nearly a quarter of the forest shows evidence of degradation. In addition, a trend analysis found that 11% of the Sundarbans was affected by the impact of Cyclone Sidr in 2007, 48% of which had not recovered by mid-2018. For the Niger Delta, the method achieved high accuracy from the 2000s onwards despite extremely low data availability. Observation of extent over time suggests that CCDC was also able to capture changes in extent caused by the 2015
mangrove die-back event in the Gulf of Carpentaria, Northern Australia and highlighted a net loss of mangroves in the Matang Forest Reserve over the last two decades, despite ongoing management. CCDC is therefore a promising methodology for global, long-term monitoring of mangroves, allowing for broader changes in extent to be examined in addition to providing details of mangrove condition change. The thesis concludes with a discussion of each chapter within the broader context of Earth Observation science, and identifies areas for future research.
| Date of Award | 2021 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Pete Bunting (Supervisor) & Andy Hardy (Supervisor) |