Spatial Trend and Geostatistical Prediction of Water Quality Index in the Pra River Basin of Ghana

Abstract

Reliable spatial prediction of river water quality is essential in mining-affected river basins where pollution varies across space. This article examined the spatial behaviour of the Water Quality Index (WQI) in the Pra River Basin system of Ghana and compared the performance of inverse distance weighting (IDW), ordinary kriging (OK), and co-kriging (CK) for WQI interpolation. The study used 150 sampling points distributed across the Birim, Offin, and Pra basins, and WQI was derived from pH, total suspended solids (TSS), total dissolved solids (TDS), and electrical conductivity (EC). Spatial trend was assessed using Kendall’s tau, distributional skewness was reduced using the Yeo-Johnson transformation, and interpolation performance was evaluated using cross-validation metrics including RMSE, MAE, and R². Results showed a statistically significant positive spatial trend in WQI (Kendall’s τ = 0.2612, p < 0.001), confirming non-stationarity. The Yeo-Johnson transformation substantially reduced skewness from 0.4797 to 0.0276 at λ = -0.3611. Among the interpolation methods, ordinary kriging produced the best predictive performance (RMSE = 0.6782, MAE = 0.5103, R² = 0.5370), outperforming both the best IDW configuration and co-kriging. The findings show that spatially structured WQI data in the Pra Basin are better modelled with variogram-based geostatistical methods than with deterministic interpolation alone. Ordinary kriging therefore produced the best relative performance among the interpolation methods tested and is suitable for basin-scale WQI mapping in the study area, although further improvement may be achieved by incorporating additional spatial covariates and trend-explicit models.