Contribution of GIS tools and statistical approaches to optimize the DRASTIC model for groundwater vulnerability assessment in arid and semi-arid regions: the case of Sidi Bouzid shallow aquifer

The groundwater vulnerability evaluation against pollution becomes an effective tool for sensible resource management and land use planning. These resources are threatened by the emanating from anthropogenic activities and the intensive fertilizer applications in agricultural sector. Over time, groundwater abstraction exceeds aquifer recharge, leading to the groundwater levels decline and constitutes a continuous and serious problem. Protecting groundwater resources from pollution has become necessary especially by the use of the vulnerability models. The objective of this study is to establish a groundwater vulnerability map in the Sidi Bouzid region using geographical information system (GIS) by overlaying the available hydrogeological data. As well, a comparative study between two vulnerability models has been carried out to conduct more realistic management of groundwater quality. In fact, groundwater vulnerability map to pollution was determined by applying the DRASTIC model. This model was almost applied, without any changes, in the Sidi Bouzid aquifer. In this case, the DRASTIC method was modified to estimate vulnerability and pollution risk of shallow aquifers especially for nitrate component. The modified DRASTIC model is based on some of statistical and geostatistical approaches in odor to amend the rates and the weights of DRASTIC parameters via GIS tools. This method has proved to be an effective way to achieve better correlation between nitrate concentration ([NO3−]) and vulnerability index. The nitrate content was analyzed for thirty-eight samples. The obtained Pearson’s coefficients correlation showed that the vulnerability index and the [NO3−] relationship was increased from 44 to 82% (before and after modification, respectively). This requires the use of the modified DRASTIC method. Statistical approaches, taking into account parameter classes for the categorical parameters, show a R2 statistic equal to 75.4% for the total variance, explained by multiple linear regression (MLR) model, for the estimation of NO3−.