Minimization of Groundwater Observation Wells Using Geostatistics and Optimization Technique (Case study: Dezfoul-Andimeshk plain)

Groundwater is one of the most valuable water resources owing to its quantity and quality. Measuring water level is a basic and essential step in any groundwater study. Since the measured data are derived only from a limited number of observation points and, at the same time, they must be extended to the whole surface of the zone, it is essential to determine optimal location of measuring network. Accordingly, for reduction of observation wells, a number of them were eliminated in such a way that the remaining wells have an optimal combination. Dezfoul-Andimeshk plain, located in north of Khouzestan Province, was investigated as a case study. Kriging, as the best linear unbiased estimator, was used for interpolation of groundwater level. Based on the measured data of 76 studied observation wells, a theoretical variogram was fitted to empirical variogram points. To achieve the optimal combination, tabu search algorithm (a meta-heuristic algorithm) was used. Two computer programs including GSLIB and MATLAB were used for Kriging and tabu search, respectively. Linking the programs and preparing conditions for data exchange between them, a model called optimizer model was generated so that capability of optimizing the groundwater surface measuring network is made possible. The observation points are distributed such that variance estimation error limited to the plain extent is minimized. Optimization model is run in five different cases. For the first to third cases, simply considered for verification purposes, optimal selection of one to three observation wells were respectively presented, which authenticated the model performance against complete search method. In the fourth and fifth cases, the optimal selection of 50 and 60 out of 76 observation wells were presented for which the groundwater levels were compared and the results show that there exists a good match between the groundwater surface results in comparison with those of whole wells (76 wells). As the variogram fitted to the groundwater data is isotropic, the distribution of the observation wells along different directions will be the same. This is tangible for the first to third cases in which the number of wells is small. Optimization results of 50 and 60 observation wells were automatically achieved by the model without any interference. The comparisons of results indicate a good accuracy for the optimization model. Also, the results show that using the present model saves a lot of time.