AbstractSaltwater intrusion (SWI) poses a significant threat to the livelihood of populations in coastal zones who are dependent on freshwater extracted from aquifers near to the sea. The sustainable management of coastal aquifers is crucial to prevent the degradation of freshwater resources by the landward intrusion of saltwater due to over-pumping and climate change. Difficulties arise when modelling the extent ofSWI given the inherent heterogeneity present in most coastal aquifers, which can significantly affect the flow and transport properties of the system. The objective of this study is to develop the understanding of the effects of heterogeneity on the hydrodynamics of SWI.To appreciate the effects of heterogeneity it is first essential to understand the hydrodynamics of SWI in homogeneous aquifers for comparison. An experimental investigation was undertaken to determine the hydrodynamics of key SWI parameters (toe length (ܶܮ ,(width of the mixing zone (ܹܯ (ܼand angle of intrusion (ܣܱܫ ((using image analysis of a 2D sandbox style laboratory-scale experiment. Investigation of the heterogeneous effects followed, involving analysis of SWI parameters in increasing levels of aquifer heterogeneity, from layered to block-wise structured and finally randomly distributed cases. The results were then compared to numerical models, including a stochastic model using randomly generated heterogeneous fields.In order to facilitate the goals of the experimental study, a novel methodology was developed to observe and quantify SWI parameters at high resolutions. Existing methods are mainly based on visual observations, which are subjective, labour intensive and limited in the temporal resolutions that can be analysed. The developed methodology promotes autonomy, minimises human input and achieves high resolution image to concentration conversion to allow quantification of SWI parameters under strong transient conditions.The developed methodology was applied to various homogeneous aquifers of different diameter porous media and the intrusion observed for advancing and receding transient conditions. The transient results showed good correlation between numerical and experimental intrusion rates. The results revealed that thereceding saltwater wedge ܶܮ reached a steady-state condition sooner than theadvancing case, indicating faster fluid velocities during retreat. The ܹܯ ܼexhibitedsimilar traits, where greater increases in ܹܯ ܼwere observed for the receding case,indicating higher dispersion due to the increased fluid velocity. Investigations ofܣܱܫ revealed the formation of a diluted saltwater volume at the toe during theinitial moments of the receding case. The repeatability of the experimentalprocedure was assessed in terms of ܶܮ and ܹܯ ,ܼwith an average coefficient ofvariation less than 18%. For heterogeneous cases, the results showed that ܶܮs generally decreased whileܹܯܼs generally increased when compared to the homogenous reference case. Thesaltwater-freshwater interface showed distinct gradient changes when transitioningbetween regions of high and low permeability due to flow refraction. The relativechange in ܣܱܫ across permeability zones was found to depend on hydraulicgradient, depth in aquifer, thickness of the layer and surrounding heterogeneousstructure. Steady-state ܶܮs and ܹܯܼs for blocked-wise and equivalent randomlydistributed aquifers compared well. However, transient observations showed thatthe saltwater-freshwater interface was heavily influenced by preferential flowchannelling governed by the heterogeneous structure. The results from a stochasticanalysis of random heterogeneous fields showed agreement with the trendsobserved experimentally.
The use of machine learning techniques (MLTs) to improve experimental test efficiency was assessed by applying the Random Forest method to the regression calibration required to convert image light intensity to concentration. The method saves significant preparation time by generating a calibration that is applicable to all heterogeneous configurations, negating the need to run individual calibrations for each case. The method showed promising results but was significantly affected bynon-uniform backlight distribution, causing deviations in key SWI parameters when compared to the high resolution pixel-wise method. For larger scale experiments,this method could prove beneficial with a concerted effort to improve the backlight uniformity.
|Date of Award||Dec 2017|
|Supervisor||Ashraf Ahmed (Supervisor) & Gerard Hamill (Supervisor)|