The k-Nearest Neighbor Evaluation of Groundwater Quality in the Distal Merti Aquifer, Modogashe Area

Abstract

Abstract— The trans-boundary Merti aquifer possesses various water quality categories, namely, saline, fresh and hard species of water. Some boreholes in the Modogashe area are found to be of fresh water quality, yet the vast majority of the wells are either saline or outright brackish. The Northern Water Works Development Agency wished to fund the sinking of a well ,and the company CEO wished to make an informed decision on whether or not to proceed with this project as any water of unacceptable quality would means being rejected by the community, way after millions have been sunk into the project. The objective of the present study is to help develop a groundwater exploration decision-making under uncertainty, so that the Board Technical Services Manager (TSM) and the CEO get the technical info backing the decision to proceed with the program. Consequently, much of the data used to develop the models were sourced from the Kenyan rather than the Somalia-side of the Merti Aquifer. To achieve this, a list of existing data of groundwater sources of the Merti aquifer were assembled and processed so that we now had longitudes, latitudes, depth, elevation, mean resistivity of the main aquifer, in the first five dataframe columns. The Total dissolved Solids (TDS) or Electrical conductivity (EC) or the respective rows of data were then analysed against the Kenya Bureau of Standards ( KEBS ) for water quality, so that a sixth column emerged, code-named as water category. This column expressed the status of the water: whether it was saline, fresh or hard. If fresh or hard, it was categorized as good. If saline, it was categorized as bad. To infer the water quality of the new field data points of a proposed drilling spot whose depths have been determined using Vertical Electrical soundings , a geoelectrical mapping tool, new rows of longitudes, latitudes, elevations, depth, and resistivity (abbreviated as rho) were brought in and predicted. This field dataset lacked the final column, as it is the one to be predicted. The predicted findings were hundred percent correct, for the expected water quality of the new spots. The kNN algorithm was used to generate a prediction algorithm with an accuracy of way over 90.9 percent. With this high level of accuracy, the model was deemed fit for use in predictions of new dataset class of water category, whether the new dataset would give rise to good or bad water species.