Improving Numerical Simulation of Streams and Shallow Groundwater in the Mississippi Alluvial Plain

Author(s): Leaf, A.; Breaker, B.; Adams, R.; Dietsch, B.

In humid regions such as the Mississippi Alluvial Plain (MAP), surface water typically exerts a fundamental control on both the water levels and flow directions of shallow groundwater. Groundwater pumping ultimately diverts water that would otherwise go to streams or lakes, and can have dramatically alter surface water features, even when only a small portion of the overall regional water budget is removed compared to recharge. The stream network in the MAP is an important source of water to wells thus serves as an important consideration for sustainable management of groundwater.

Representation of streams in groundwater models has historically been arduous and error-prone, requiring many GIS operations or even hand-digitizing of features. Software support for automation of stream network creation and visualization has been limited. Automated stream network generation in the MAP region is challenging in that it encompasses a large number of streams, many of which originate far from the area of interest, has a complex history landscape alteration, and highly variable surficial lithology. In the MAP area, three new approaches are being performed to improve the simulation of this important surface water network.

1) Automation, machine learning and additional field data collection are being used to improve the representation of streams in the Mississippi Embayment Regional Aquifer System (MERAS) model. Python code was developed to automatically translate information from the NHDPlus version 2 database into finite-difference stream networks. The revised networks include most streams that have base flow for at least part of the year, increasing the number of streams represented in the MERAS model from 43 to more than 5,000. The automated processes of generating the stream networks facilitates adaptation to different computational grids or inset areas within the larger Mississippi Embayment.

2) A random forest (RF) statistical model was developed to estimate streamflows originating outside of the MERAS study area as well as ungaged flows in both space and time within the model domain. The RF model considers drainage area, climate statistics, and other factors to estimate stream flows at monthly intervals, providing valuable information on stream inflows to the model because continuous flow observations that can be compared to output from the physics-based finite difference model.

3) Waterborne geophysical electrical resistivity surveys were conducted on more than 700 miles of streams, to identify areas of relatively high and low permeability in the streambed sediments, and inform the representation of streambed conductance in the model. Simulated streamflows will be compared to existing and newly collected measurements of base

Go back


Past Conference Archive