An efficient Variational Inverse Modeling approach for estimating river bathymetry using small-UAS-based IR video imagery
Abstract
High fidelity bathymetry maps are necessary both for navigation and for the production of accurate forecasts in riverine environments. SRI International (SRI) has implemented new efficient tangent-linear forward-adjoint model (TLM-ADM) in its River Remote Survey (R2S) system, which has been used to estimate river characteristics given remote observations of the river. The R2S system, employing a variational inverse modeling framework, is a constrained optimization algorithm wherein the error between the observations and the model prediction of the observations is minimized, subject to the constraint that the governing equations are satisfied. The R2S system has been exercised on several reaches in multiple rivers and has been shown to be accurate in estimating bathymetry given surface velocity observations. The original algorithm is a numerical implementation of the analytic non-linear adjoint equations and carries a significant computational cost. The newly developed TLM-ADM system has led to significant reductions in computational resource requirements when applied to the traditional assimilation problem of estimating initial and boundary conditions. The algorithm has now been included in the R2S System.
SRI has extracted the surface velocity fields for riverine environments using image processing techniques (e.g., PIV) applied to Infrared (IR) video imagery of the water surface, collected by a small unmanned aerial system (UAS) developed by a team from Penn State University and Georgia Institute of Technology. The UAS has been equipped with an IR camera that can be used to observe and track thermal features on the water surface. The velocity extraction algorithm has been applied to data collected on the Pearl River, and the R2S system has been exercised on that data. Here we present a complementary set of bathymetry estimation results using the new TLM-ADM algorithms and solvers. The bathymetry fields are shown to be very similar, both quantitatively and qualitatively. The errors introduced with the linear assumptions are acceptable given the computational savings; the ultimate goal of the system is to be able to produce a bathymetry map in near-real time that is accurate enough for operational use. This effort has been funded by the Office of Naval Research under Contract Number N00014-17-C-7021.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFMIN51B..01A
- Keywords:
-
- 3360 Remote sensing;
- ATMOSPHERIC PROCESSES;
- 1964 Real-time and responsive information delivery;
- INFORMATICS;
- 4315 Monitoring;
- forecasting;
- prediction;
- NATURAL HAZARDS;
- 7924 Forecasting;
- SPACE WEATHER