top of page

Stochastic Hydromorphodynamic Modeling

Advances in hydromorphodynamic models have greatly accelerated thanks to advances in parallel computing and an exponential increase in available computational power. Despite these advances, computational expense remains a significant challenge for models, especially for stochastic modeling frameworks which require hundreds, if not thousands of simulations. In addition, while hydromorphodynamic models are very useful for a detailed understanding of system dynamics, most coastal managers need information to answer one or zero dimensional questions, such as "how many times will I need to dredge this basin through 2100?"

My work has focused on developing techniques to significantly reduce the runtimes of hydromorphodynamic models through the use of fast surrogates (see Figure to the right), specifically response surface surrogate models. We develop these surrogates through simplification of the hydromorphodynamic equations to system variables of interest coupled with a response surface model. The response surface surrogate models are trained on a limited number of high fidelity hydromorphodynamic model runs to investigate how the system variables of interest respond to model forcing and parameters. Through my work, I've been able to design extremely fast surrogate models which achieve nearly equivalent accuracy to high fidelity models while achieving multiple orders of magnitude reduction in model runtimes. These models are then implemented in a stochastic modeling framework to better account for environmental uncertainties and help coastal managers better understand the full range of potential futures within their systems.

surrogate_model.png

Example of a response surface surrogate model: and Y represent model forcing and/or parameters, F(X,Y) is the variable of interest response, and the shaded curve is the surrogate model.

This work was applied to Newport Bay (see below), an urban estuary located in Southern California to improve sediment management under the SedRISE project (http://blumcenter.uci.edu/flood/sedrise/). The technical details of this work have been submitted for publication in the journal Advances in Water Resources, and is currently under review (full citation below).

Brand, M.W., Guo, L., Stein, E.D., & Sanders, B.F., (2020), Multi-decadal simulation of estuarine sedimentation under sea level rise with a response-surface surrogate model. Submitted to Advances in Water Resources (in review).

20180829_114426.jpg

Newport Bay, located in Southern California, where San Diego Creek (under bridge) meets the bay 

bottom of page