ALLUVIAL AQUIFER HETEROGENEITY : GENETIC VERSUS GEOSTATISTICAL MODELS

Alluvial aquifers are widely exploited and also very vulnerable, because the water is easily accessible with little natural protection. They are also extremely heterogeneous aquifers because of their formation processes: these are mainly fluvial processes acting over several thousands of years. Depending on climate and geological forcing functions, the river evolves and modifies by erosion and deposition its floodplain. It changes its pattern as well as its sedimentary functioning. It might incise or fill-in valleys, depositing various types of sediments from gravels to loam. The correlation length of alluvial facies is usually in the order of tens of meters. Changes can be sudden, with channel structures embedded in larger sedimentary structures. The shapes of these sedimentary structures is not well described by a geostatistical covariance function, as they are generally very anisotropic and not convex ellipsoids, but rather undulating ribbons.

We will describe a new method for generating the sediment heterogeneity of alluvial deposits, based on a genetic algorithm which mimics some of the hydraulic rules active on meandering or braided streams, to make them deposit or erode sediments. After describing this new model, and comparing it with other approaches, it will be applied to a 5.4x1.4 km² reach of the Aube River, upstream of Paris. The model is based on the data obtained from a single cross-section of the alluvial plain, with 44 auger-holes along the 1.4 km width of the plain. As the model describes the sediment facies, we assign to each facies an assumed permeability. Flow and transport simulations are then made using Modflow. In order to compare the results with simulations based on other descriptions of heterogeneous media, we will use a classical geostatistical tool (indicator simulations) to generate alternative images of the same medium, based on the same initial data, and with the same assumption on the permeability of each facies. Finally, an equivalent uniform medium (obtained through permeability upscaling) will also be also generated. Comparing these three approaches on the flow and transport results, we show the intrinsic capacity of the new genetic model to generate more realistic flow and transport processes in such heterogeneous alluvial media, where permeability barriers and buried channels dominate the spatial repartition of heterogeneities.