EP046-0006
Deep learning application to fast estimation of riverine surface velocity

Monday, 14 December 2020
Poster
Mojtaba Forghani1, Yizhou Qian2, Jonghyun Harry Lee3, Matthew Farthing4, Ty Hesser5, Peter K Kitanidis6 and Eric F Darve1, (1)Stanford University, Mechanical Engineering, Stanford, CA, United States, (2)Stanford University, Institute for Computational and Mathematical Engineering, Stanford, CA, United States, (3)University of Hawai‘i at Mānoa, Civil and Environmental Engineering, Honolulu, HI, United States, (4)US Army Engineer Research and Development Center, Coastal and Hydraulics Laboratory, Vicksburg, MS, United States, (5)U.S. Army Engineer Research and Development Center, Coastal and Hydraulics Laboratory, Vicksburg, MS, United States, (6)Stanford University, Department of Civil and Environmental Engineering, Stanford, CA, United States
Abstract:
Estimation of the riverine flow velocity is important in practical applications, such as the safe

and efficient maritime transportation, prediction of potential beach erosion, land subsidence,

biological oceanography, and food risk management. The shallow water equations are typi-

cally used to predict flow velocity, provided with the boundary conditions (BCs), such as the

discharge and the free surface elevation, as well as the riverbed profile (bathymetry). How-

ever, other than a few simple cases such as the one-dimensional flow or idealized hyperbolic

riverbed profiles, these equations must be solved numerically and thus are computationally

expensive. Furthermore, these numerical methods typically require a fairly high resolution of

the riverbed profile and the BCs in order to predict accurately flow velocities. However, direct

high-resolution bathymetric surveys are time-consuming and costly for long study reaches in

watershed scales.

In this work, we propose multiple fast solvers for the shallow water equations that can

be used for the online prediction of riverine flow velocities with variable bathymetries and

BCs. Our approach consists of two major steps: first, using the principal component geo-

statistical approach (PCGA) we estimate the distribution of the bathymetry from the

velocity measurements; then we use several machine learning techniques in order to obtain a

fast solver of the shallow water equations, provided with the distribution of the bathymetry

(PCGA posterior distribution) and different BCs. Our method can incorporate bathymetry

information into the flow velocity prediction for improved accuracy; for example, in cases

where the bathymetry is available for a limited number of cross-sections. Furthermore, this

additional information of the sparse bathymetry measurement can be incorporated into the

prediction at no additional cost (i.e., does not require additional training). We

have validated our solvers on the Savannah River near Augusta, GA. Our results show that

the fast solvers are capable of predicting the flow velocities with variable bathymetry and BCs

with reasonable accuracy, at a very low computational cost.

<< Previous Abstract | Next Abstract >>