Prediction of Saturated Hydraulic Conductivity in Irrigated Soil Using Multiple Linear Regression: A Data-Driven Approach

Abstract

Saturated hydraulic conductivity (Ksat) governs water movement in the soil profile and is critical for irrigation, drainage, and water management. Direct measurement of Ksat by constant-head permeameter is laborious and time-consuming, necessitating predictive models based on readily measurable properties. This study evaluated multiple linear regression (MLR) for estimating Ksat of Kadawa soils in the Kano River Irrigation Project (KRIP), Nigeria. Twelve composite samples (0 - 45 cm) were analyzed with clay content, silt content, bulk density (BD), and particle density (PD) as predictors. Pearson correlation, MLR, variance inflation factor (VIF), residual diagnostics, and leave-one-out cross-validation (LOOCV) were used to assess relationships and model performance. Ksat ranged from 165.81 to 552.30mm/day with a mean of 387.20±134.68mm/day. The fitted model explained 64.64% of Ksat variability (R2 = 0.6464; adjusted R2 = 0.4443) with an RMSE of 76.68mm/day and MAE of 58.88mm/day. Severe multicollinearity was detected between BD and PD (VIF = 10.74 and 9.56). Residual diagnostics showed no significant violation of normality (W = 0.972, p = 0.933) or homoscedasticity (BP = 8.134, p = 0.087). However, LOOCV revealed limited predictive generalizability (R2 = -0.3165, RMSE = 147.95mm/day, MAE = 108.11mm/day). The results indicate that clay, silt, BD, and PD provide useful exploratory information for Ksat variability in KRIP, but the model remains site-specific and should not replace direct measurements without independent validation. Larger datasets incorporating diverse soil conditions and structural properties are recommended for robust pedotransfer functions.