Groundwater Quality Prediction Using Logistic Regression Model for Garissa County

  • George Krhoda Department of Geography and Environmental Studies, University of Nairobi
  • Meshack Owira Amimo Department of Geography, Kenyatta University
Keywords: Groundwater, Water quality, Prediction, Logistic regression model

Abstract

Groundwater quality modeling can reduce the cost of exploration and siting of boreholes considerably. The present study applies Logistic Regression Model to predict the probability of siting boreholes of fresh or saline water based on geospatial data such as altitude (m), longitudes, latitudes and depths (m), and geophysical data such as electrical resistivity from 45 exploration sites. The geology of the study area is represented by permeable water-bearing Tertiary-Quaternary sediments located within the Anza Rift. The water bearing zones, or water struck levels, range in depth between 50 and 150 m and the average yield of about 1 - 5 m3 per hour, in the case of old wells done using percussion rigs in the period between 1960s to the 1990s. Recently, the discharge in the wells done using modern mud rotary equipment yields up to 30 m3 per hour, with depths ranging between 200 to 250m below ground level. The modeling results show strong correlation between the dependent variables; depth, mean resistivity, longitudes, and latitudes on one hand, and salinity status of aquifers. It is, therefore, possible to know the water quality of a location in the study area before actual drilling is undertaken. Of all the runs made, 93% were predicted accurately while only 7% of the cases deviated from the predicted quality. These findings prove the usefulness of the LRM in predicting and identifying sites of high groundwater accumulation and groundwater salinity in arid region.

Published
2019-02-28