Abstract

The issue of ground water contamination has become a growing environmental crisis in sub-Saharan Africa, and Nigeria has become acutely susceptible to this with its high rate of urbanization, industrial growth, and climate change. The conventional techniques used in assessing hydrogeology do not reflect the spatial heterogeneity and time variability of the process of pollution propagation. This paper combines the machine-based learning algorithms and the multispectral satellite imagery and on-site measurements to create some predictive models of the quality degradation of groundwater in the Niger Delta area in Nigeria. We used Representative Concentration Pathway with Landsat 8 OLI/TIRS, Sentinel-2 MSI, and SRTM digital elevation data and physicochemical parameters of 156 monitoring wells to predict contamination hotspots over Representative Concentration Pathway conditions using Random Forest, Support Vector Machine, Gradient Boosting, and Artificial Neural Network classifiers. The Random Forest model proved to be better because it achieved an overall accuracy of 89.3% and Kappa coefficient of 0.86, which is able to identify areas at high risk of nitrate, heavy-metals and total dissolved solids exceedances. The analysis of the feature importance showed that normalized difference vegetation index, land surface temperature, precipitation patterns, and distance to industrial facilities were the most significant predictors. Assessment of spatial autocorrelation based on Moran I showed that pollution events were significant with clustering. These results highlight the revolutionary nature of applying Earth observation systems with computational intelligence models in proactive management of groundwater under conditions of limited data in areas of growing anthropogenic pressures.

Keywords: Groundwater contamination, Remote sensing, Climate change adaptation, Niger Delta, Predictive modeling