Diagnosis and removal of trend component in groundwater elevation data by using experimental semivariograms: An application to Mahdia shallow aquifer system of Tunisia

Groundwater is a precious source in arid and semi-arid regions of the world. Potentiometric level data show a non-stationary behavior due to the trend component. Consequently, it renders difficult to delineate groundwater-surface, i.e., spatial estimation. The staple objectives of this study are three-fold: (1) to diagnose trend component in potentiometric data by utilizing experimental semivariograms, (2) to model trend surface, (3) to remove trend component from groundwater elevation and demonstrate the effectiveness of the method by using experimental semivariograms of residuals. The study was conducted in the region of Mahdia, located in central-eastern Tunisia. Point-wisely scattered groundwater level data from 69 observation sites were used in the study. Data of the rainy period in 2008 and 2018 were utilized in order to elude negative anthropogenic impacts on potentiometric levels. Omnidirectional experimental semivariograms (OESs) were obtained from elevation data by using the minimum feasible lag interval. Experimental semivariograms helped us diagnose trend component, i.e., non-stationarity. Trend surface analysis was conducted through using the ANOVA technique. A very significant proportion of the variation in the observed data could be “explained” by a quadratic trend surface model. Comparing the partial cubic trend surface model with the quadratic one resulted in “statistically-not-significant” F value. Consequently, there was no real evidence that the partial cubic surface model was actually any better than the quadratic one. OESs of residuals indicated that the quadratic trend was effectively removed from the data.