Informatics of Water-Level Measurements for the Mississippi River Valley Alluvial Aquifer, USA
Monday, December 4, 2017
The Mississippi River Valley alluvial aquifer (MRVA) underlies much of the Mississippi River alluvial plain (MAP). A large regional study by the U.S. Geological Survey (USGS) of the hydrogeology and numerical groundwater-flow models of the MAP is on going (c. 2016–2021). The poster describes some research into the MRVA water-level database (16,756 wells; 285,429 measurements). A complex history of MRVA data collection exists over the past century by the USGS and equally important, numerous other local, state, and federal agencies. Data anomalies potentially represent erroneous information and hinder scientific study—Objective large-scale identification of such data is the primary focus of groundwater informatics for the MRVA. Two types of anomalous data are sought: (1) time-series outliers and (2) spatial outliers. Time-series data permit well-hydrograph visualization and temporal trend analyses using generalized additive (GAM) and support vector machine (SVM) models. Measurements exceeding a residual magnitude of 20 feet are flagged as outliers and relayed to database administrators (DBAs) for further scrutiny. Spatial outlier detection is based on period-of-record minimums using a two-dimensional (2D) GAM. Wells with data "out-of-place" in a local region are flagged as outliers and relayed to DBAs for further scrutiny. Patterns in decadal means for 2000 and 2010 are depicted. Exploratory long-term monitoring network identification using a 2D-SVM is made. Wells "supporting" the 2D-SVM are more informative than those not supporting it.