Cluster Analysis for Hydrogeological Units Identification and Characterisation in Glaciofluvial Sediments

In this work, a standard methodology for characterising heterogeneous unconsolidated sedimentary sequences and providing the initial input for a subsequent application of geostatistical methods is presented. The approach combines particle size distribution analysis with a machine learning clustering algorithm. The fluvioglacial sequence underlying the Sellafield nuclear legacy site in northwest Cumbria was selected as a suitable location to develop this approach. There, sedimentary heterogeneities are relevant due to the potential existence of preferential flow paths via high-permeability gravel layers, as well as variable sorption processes that occur. A total of 75 samples were analysed, and 5 clusters were defined using the K-Means algorithm. The classification considered the proportion of fines (< 0.0625 mm), sand (0.0625 - 2 mm), and gravel (> 2 mm) in each sample. Additionally, hydraulic conductivity distributions for each cluster were obtained using the Kozeny-Carman equation. Results suggest that the lithofacies can be grouped into 3 main families, which would simplify future groundwater models for the site. The workflow developed here is applicable to other contexts, such as oil, brine, gas, or geothermal reservoirs, due to its low cost and unbiased way of classifying geological units for the purpose of hydrogeological modelling.”