Large-scale Data Assimilation Using Parallel and Cloud Computing
Understanding the evolving state of the nearshore zone, e.g., ocean surface waves and seabed elevation, is crucial to many tactical decisions for naval operations, coastal infrastructure design and management, protection of the hinterland against flooding, shoreline management, and recreational safety.
Among the many hydrodynamic conditions that determine nearshore wave transformation, seabed elevation, or bathymetry, is one of the most important variables for understanding and predicting nearshore processes such as waves, currents, and shoreline changes. The applications of bathymetric data for military and naval uses are varied and have far-reaching implications for training exercises and warfare. Examples include:
· accurate navigation
· route-survey planning
· identification of amphibious landing sites
· mine detection and placement
all of which requiring accurate knowledge of the nearshore depth.
Figure 1: Example snapshot images from Duck, NC with six cameras
However, direct bathymetry measurements over a large area are often not feasible due to the expensive sampling costs, logistic limitations, and inaccessibility of the shore (e.g., on enemy territory). As an alternative, more easily accessible ocean surface remote sensing data from satellites, aircrafts, drones (or fixed towers when possible), have been actively studied to infer the bathymetry map in combination with physics-based hydrodynamics model that relates ocean surface wave transformation to bathymetry.
Satellites, drones, etc., produce large amounts of detailed images and data. However, significant amounts of computing are required to extract the relevant information from this massive amount of data. This can become critical when conditions change rapidly and predictions need to be made in real-time. Consequently, remote-sensing data-based bathymetry estimation require high performance computing (HPC) resource utilization to store, process, and analyze these big datasets, and produce high-fidelity hydrodynamic models in real-time.
Novel mathematics is required to produce computationally more efficient algorithms that can account for errors in the measurements, e.g., cloud, sun glare, rain, sea foam, wave breaking, etc. Instead of making a single prediction, intervals of confidence need to be produced to assess how confident we are in the predictions, and what the margins of errors and uncertainties are.
To accurately identify nearshore seabed evolution in real time, the Stanford AHPCRC research team has developed a robust real-time bathymetry estimation algorithm that can scale up to very large nearshore zone (>> 1 mile2) estimation problems, and perform seamlessly in modern HPC environments. We are collaborating with the U.S. Army Corps of Engineers Research and Development Center, Coastal and Hydraulic Laboratory (USACE ERDC-CHL), and have applied our novel computational techniques to a testbed site near the USACE ERDC-CHL’s Field Research Facility in Duck, NC. At this site, high-resolution ocean surface images are being collected from a fixed tower and satellite. With our estimation algorithm, temporal changes in a bathymetry map of 1 mile2 were accurately estimated using historic ocean surface images with a two-dimensional hydrodynamics model. The results were successfully validated with in-situ depth measurements and proved the accuracy, robustness and applicability of our methods.
As a next step, we are currently testing accurate and robust real-time nearshore condition tracking. The high-resolution ocean surface images at our testbed site are taken at a high rate generating several terabytes worth of images in a day. Thus, real-time image processing (Figures 1-3) have become critical tasks for successful bathymetry identification. However, such computationally expensive tasks are impossible in traditional computing environments with limited hardware power. With AHPCRC’s computing resources, we will be able to access and analyze large-scale datasets, estimate the bathymetry, and eventually predict with great accuracy the nearshore processes.
Figure 3: Plain view ocean surface image processed from Figure 1.
Knowledge of bathymetry is critical for military operations in the nearshore. Unfortunately, estimating bathymetry in denied areas is still a major challenge and impedes the military’s ability to operate effectively and safely in both warfighting and humanitarian assistance and disaster relief.
Professor Darve’s group has worked closely with our ERDC team over the past year to bring their computational expertise to bear on the challenge of estimating bathymetry in denied areas through remote sensing and standoff assessment. With the help of their fast inversion algorithms, we are making great progress towards providing the Army with robust, high-fidelity estimation of bathymetry in real time. – Dr. Matthew Farthing, ERDC