Testing Approaches and Sensors for Satellite-Derived Bathymetry in Nunavut

• Main Author: Holman, Kiyomi
• Format: Thesis
• Language: English
• Published: Université d'Ottawa / University of Ottawa 2020
• Subjects: Satellite-derived bathymetry; Remote sensing; Ocean optics; Canada; Arctic; Bathymetry; Hydrography; Nunavut; Cambridge Bay
• Online Access: https://dx.doi.org/10.20381/ruor-25626; http://ruor.uottawa.ca/handle/10393/41402

Nearshore bathymetry in the Canadian Arctic is poorly surveyed, but is vital knowledge for coastal communities that rely on marine transportation for resources and development. Nautical charts currently available are often outdated and surveying by traditional methods is both time consuming and expensive. Satellite-derived bathymetry (SDB) offers a significantly cheaper and faster option to provide information on nearshore bathymetry. The two most common approaches to SDB are empirical and physics-based. The empirical approach is simple and typically does well when calibrated with high-quality in-situ data, whereas the physics-based approach is more difficult to implement and requires precise atmospheric correction. This project tests the practical use of five methods within the empirical and physics-based approaches to SDB, using Landsat 8 and Sentinel-2 satellite imagery, at seven sites across Nunavut. Methods tested include: the Ratio-Transform, Multiband, and Random Forest Regression methods (empirical) and radiative transfer modeling (physics-based) using two atmospheric correction models: ACOLITE and Deep Water Correction. All methods typically use geolocated water depth data for validation, as well as calibration for the empirical methods. Spectral reflectance for model inputs were collected in Cambridge Bay, NU. Water depth data were acquired from the Canadian Hydrographic Service. All processing was conducted within the framework of plugins developed for the open-source GIS software, QGIS. Results from the empirical methods were typically poor due to poor calibration data, though Random Forest Regression performed well when good calibration data were available. Due to poor quality validation data, error for the physics-based results cannot be adequately quantified in most places. Additionally, atmospheric correction remains a challenge for the physics-based methods. Overall, results indicate that where large, high-quality calibration datasets are available, Random Forest Regression performs best of all methods tested, with little bias and low mean absolute error in water less than 10 m deep. As such datasets are rare in the Arctic, the physics-based method is often the only option for SDB and is an excellent qualitative tool for informing communities of shallow bathymetry features and assessing navigation risk.