Navigation in remote areas can pose a significant risk if shoals or the coastal topography need to be adequately mapped. Many coastal regions in northern Quebec and Canada remain unmapped for bathymetry. During the last decade, the advent of constellations of Earth Observation satellites yielded relatively high-resolution time series of optical imagery, and the democratization of AI-based algorithms opened the door of significant improvement of satellite-derived bathymetry (SDB) products. Nevertheless, SDB estimation remains difficult due to the ambiguity and complexity of the signal measured by satellites. During the last 3 years, Arctus has developed an innovative solution for SDB by leveraging the new techniques that emerged in AI. The solution encompasses an entire optical image processing chain, resulting in an analysis-ready SDB product. AI techniques are employed in the entire backend, starting from measured top-of-atmosphere radiance across the visible spectrum. These techniques include automatic selection of high-quality imagery (cloud-free, glint-free, etc), atmospheric correction, optically shallow water detection, and bathymetry predictions. When available, ground truth data of bathymetry can be used for deep learning model development. This process ensures that the imagery used is of high quality, with clear atmospheric and water conditions and accurate bathymetry predictions are made. This solution, referred to as AI4SDB, is being applied along the Quebec and Canada coastline to meet the needs of stakeholders. The presentation will first describe the AI4SDB toolbox, followed by numerous case studies demonstrating the developed solution quality. Finally, we’ll discuss the potential to improve our technical solutions and services to our clients.

SeafloorMapper is a GUI-based application designed to facilitate and partially automate identification of seafloor-reflected photons in ICESat-2 ATL03 data. It assists the user to ingest the raw ATL03 data files, manually or semi-automatically identify which photons represent seafloor locations, correct these locations for the effect of light refraction at the water surface, and output the corrected data in a user-specified format. Here we present the range of tools available within the software, as well as their user interface. To demonstrate its use, we employ SeafloorMapper in a case study in Cambridge Bay, Nunavut, to derive water depth point data from a single ICESat-2 ATL03 granule acquired on 21 August 2021 (Figure 1). We compare the resulting bathymetric data to two survey data sets provided by the Canadian Hydrographic Service – an airborne lidar (ALB) survey from 1985 (CATZOC B), and a multi-beam (MBES) survey from 2014 (CATZOC A1). The results are satisfactory, with mean water depth differences between the data sets of 0.65 m (ICESat-2 vs. ALB) and 0.64 m (ICESat-2 vs. MBES). SeafloorMapper is built with Python as free and open-source software, which facilitates not only free access but also customization and adaptation based on the specific needs and preferences of individual users. The software is available at https://github.com/ylin152/SeafloorMapper.git.

Humans are traditionally wary and distrustful of new technologies until they can be definitively proven. Emerging technologies employed in hydrographic and bathymetric surveys, especially when the collected data is to be used for safe navigation, follows this axiom. Engaging emerging technologies such as Satellite Derived Bathymetry (SDB) and Crowdsourced Bathymetry (CSB) for the creation of navigation quality products has historically been discouraged by hydrographic organizations globally on account of the lack of research, trials, validation or analysis necessary to assign appropriate levels of data uncertainty; which in turn defines Categories of Zone of Confidence (CATZOC) levels. Industry providers of SDB and CSB solutions, TCarta and Orange Force Marine (OFM), have leveraged their joint, overlapping data sets collected using their technologies, and have compared the SDB / CSB data sets against known survey grade data sets (Multibeam or bathymetric LiDAR) in an attempt to quantify the individual uncertainty levels of the two emerging technologies. Taking this validation process further, utilizing post-processed and corrected data sets from each of these emerging technologies, (alongside a comparison to trusted reference data sets), the authors will further demonstrate how the combination of these emerging technologies can be used collectively to enhance or augment each other and provide quantitative, multi-iteration “proof” of the value of depth data collected via SDB and/or CSB; in turn helping to provide greater confidence in the technologies. Appreciating the hesitation of using unproven technologies, and the limited means of depth confirmation when using only a single source of collected bathymetric data, this industry-driven initiative serves as a means to develop confidence in and grow acceptance of SDB / CSB by leveraging and quantifying data-driven, real-world uncertainty level examples to enlighten hydrographer audiences as to the individual and collective value of using SDB and CSB data in the creation of navigation quality hydrographic products.

SDB (satellite-derived bathymetry) is a cost-effective solution for large-area bathymetric mapping. Calibration of SDB models relies on in situ depth points traditionally obtained from ship-borne measurements, presenting challenges in collecting data in remote shallow areas. ICESat-2 (Ice, Cloud, Land, Elevation Satellite-2) is equipped with a laser altimeter that successfully records shallow water depths (<40m), providing an efficient solution for acquiring in situ depth data. This study compares manual and automated ICESat-2 bathymetric photon extraction techniques and determines an optimal bathymetric photon event sampling strategy. The objective is to assess how these techniques impact the accuracy of SDB maps. Integrated with Sentinel-2 imagery, the extracted ICESat-2 data train bathymetric inversion models. Various controlled ICESat-2 bathymetric photon resampling techniques are then employed to determine the optimal approach for improving the accuracy of SDB maps. NOAA lidar reference data are used to compare the automated and manual SDB extraction and evaluate the impact of resampling ICESat-2 bathymetric photon events. Anticipated results show that the depth distribution of ICESat-2 bathymetric photon events and the automated bathymetric photon extraction techniques have a considerable impact on SDB map accuracy. These findings hold significance for supporting initiatives in navigational safety, coastal management, marine ecosystem studies, and seafloor mapping.

The output of the combination of “complete coverage” remote sensing imagery and geographically sparser high-quality bathymetric data – usually LiDAR – to produce wall-to-wall bathymetric maps is known as “satellite-derived bathymetry” or simply SDB. Considerable effort continues to be made to improve SDB techniques and define where they produce the most accurate maps. Most efforts are focussed on producing SDB at a single time. Research on estimating “satellite-derived depth change” (SDDC) is much more limited. Where it has been studied, the general approach is to produce and apply separate models for t1 and t2 using time-appropriate imagery and LiDAR data. SDDC is then obtained by differencing the outputs of the two models. Uncertainty is often inferred from the accuracy of the SDB models for the two time periods and summarised using a metric such as root-mean-square error. In this study, pre- and post-hurricane Landsat imagery (30 m pixels) and airborne LiDAR data were obtained for an area centred on Key West, Florida (United States) that experienced Category 5 hurricane Irma in September 2017. SDDC was produced using the conventional approach described in the preceding paragraph. In addition, however, instead of fitting separate models for pre- and post-hurricane depths, the two Landsat images employed were “stacked” into a single multi-temporal image. That is, Bands 1 to 8 came from the pre-hurricane imagery and Bands 9 to 16 came from the post-hurricane imagery. “True” depth difference for each pixel was obtained using the pre- and post-hurricane LiDAR data and used to fit and apply an SDDC model. Accuracy of both approaches was evaluated in geographic and statistical/feature space. The accuracy of SDDC for both methods was poor. Subsequent uncertainty characterisation analysis indicated this was due to the episodic and localised nature of the depth change and the spatial resolution of the Landsat imagery.