| Piping beneath levees within the Middle Mississippi River, MMR, has been well documented for 78 years, when the Mississippi River Commission initiated geological investigations into underseepage following a substantial flood in 1937. The United States Army Corps of Engineers (USACE) defines a levee as an embankment designed to supply flood protection from seasonal high water. Piping is the "active erosion of sand or other soil from the top stratum as a result of substratum pressure and concentration of seepage in localized channels" (USACE, 1956a). The geological investigations beginning in 1937, and still continuing today, have consistently listed two conditions necessary for piping to occur: 1) a pervious substratum overlain (2) by a semi- to impervious top stratum (Fisk, 1945; Turnbull and Mansur, 1959). The phrase "conducive environment" is used for this type of environment. Where these factors are present during the time when a levee is subjected to water loading, the force exerted by the weight of water on the riverside of the levee can be transferred through the pervious substratum to the landside, resulting in a hydrostatic imbalance between strata and the surface landside of the levee (Turnbull and Mansur, 1959; USACE, 2000).;Innumerable miles of levee along the MMR and other rivers meet the "conducive environment" susceptible to piping and merit maintenance and piping prevention measures. Several secondary factors were identified in previous studies resulting in detailed geological investigations of all known levee districts meeting the "conducive environment". However, limited funding complicates data management and therefore, adequate piping prevention measures, along these levees. Using the Glynn and Kuszmaul (2004) database for PDR and FTC, several datasets were defined for regression analysis to develop a model that improves the efficiency of vulnerability assessments of the vast lengths of levees managed by the USACE. Single-variable regression analysis, to determine significance of each independent variable, and multi-variable regression analysis, to determine the final models for the datasets, were conducted during model building. Several possible models for each dataset were created using a modified forward stepwise regression procedure, also called a stepwise regression procedure, as suggested by Le (2010). Model selection was based on the chi-square statistic value and each models performance under thresholds discussed in subsection 4.3.1.;The model building process presented in this study proved to be a successful method for developing regression models meant to predict the potential for piping given the availability, or lack, of geologic and flood specific data. The final selected model, Limited Previous Model A, significantly predicted areas of high, medium, and low potential for piping along three levee districts; Prairie du Rocher Levee District (PDR), Fort Chartres Levee District (FTC), and East Cape Girardeau Levee District (ECG). The high significance of this model is largely attributed to the inclusion of previous piping events and interaction terms. |
