Remote sensing of coastal turbidity with unmanned aerial vehicle-borne multispectral and hyperspectral sensors in the coastal region of Singapore

• Main Author: Pak, Hui Ying
• Other Authors: Law Wing-Keung, Adrian
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Earth and Environmental Sciences; Engineering; Remote sensing; Turbidity monitoring; Image processing
• Online Access: https://hdl.handle.net/10356/180742
• Institution: Nanyang Technological University

Singapore, a small city-state, has undergone rapid land expansion over the last several decades, facilitated by extensive land reclamation activities. Land reclamation activities involve dumping of sediments in the coastal water, which results in elevated coastal turbidity levels in the vicinity. The implementation of Environmental Management and Monitoring Plan (EMMP), including the monitoring of coastal turbidity concentration, is thus imperative to mitigate the impacts of land reclamation on the coastal environment. At present, coastal turbidity monitoring is commonly conducted via in situ sampling and/or fixed water quality monitoring stations, which can be labour intensive and costly. Additionally, remote sensing via satellite imagery serves as a more popular and convenient mode of coastal turbidity monitoring due to its ability to monitor a large spatial scale. However, challenges associated to satellite imagery such as extensive cloud cover, high atmospheric interference, and fixed overpass schedule limit the use of satellite imagery for monitoring dynamic changes in coastal turbidity. Recently, the possibility of monitoring coastal turbidity through remote sensing with Unmanned Aerial Vehicles (UAVs) and UAV-borne multispectral and hyperspectral sensors show great promise, with the key advantages of on-demand monitoring, higher spatial resolution and band resolutions. Despite the advantages of UAV imagery have over satellite imagery, there remain several challenges such as mosaicking of UAV imagery over featureless water body, correcting for image misalignment, and impact of sun glint on water quality that need to be addressed. This study aims to perform the research and development to address the challenges so as to enable the use of UAV-borne multispectral and hyperspectral sensors for monitoring and quantifying turbidity concentrations in the coastal environment in Singapore, where land reclamation activities are being carried out. A controlled laboratory set-up was first created to conduct hyperspectral imaging of simulated turbid waters to assess the influence of various sensor configurations on turbidity prediction. Subsequently, UAV flight surveys were conducted over terrestrial and coastal environments to assess the geospatial accuracy of UAV imagery. A Global Positioning System (GPS)-based orthomosaicking method facilitated by direct-georeferencing was established, and an image alignment solution was proposed to address the missynchronisation between the Global Navigation Satellite System (GNSS) module and the image capture. The proposed method was benchmarked against existing methods and was found to improve the spatial accuracy of UAV imagery compared to existing methods, and could overcome the challenges of mosaicking of UAV imagery over featureless water body. Throughout several multispectral and hyperspectral flight surveys that were conducted at land reclamation sites in the coastal region of Singapore, it was observed that sun glint artifacts in the UAV imagery pose a significant impediment to turbidity prediction. Existing sun glint correction algorithms for high-resolution imagery typically performs poorly at shallow/turbid environments and frequently results in overcorrection at high turbid regions. As such, a sun glint correction algorithm – Sun Glint Aware Restoration (SUGAR) was developed such that the sun glint correction effectively removes most of the sun glint artefacts and is invariant to various water body types, including highly turbid waters. The SUGAR algorithm was benchmarked against existing sun glint correction algorithms, and was found to be more effective at sun glint removal, and does not exhibit overcorrection at turbid regions compared to existing algorithms. The application of the aforementioned algorithms was applied on the UAV imagery obtained from these surveys to derive turbidity concentration. An open-source software – CoastalWQL was developed to facilitate the efficient processing of UAV imagery into a turbidity map, and the preprocessing steps such as radiometric correction, de-striping, sun glint correction, and object masking were evaluated at each step via turbidity retrieval. Overall, the results ascertain the effectiveness of the preprocessing procedure and contributed to the enhancement in turbidity prediction.