Evaluation of sampling frequency from groundwater resources on pollution source identification characteristics

Groundwater is an important natural resource that is widely used for supplying domestic, industrial and agricultural demands. In the past decades, due to increasing detections of pollutants, monitoring and protection of groundwater resources as well as remediation of contaminants have been the main tasks in groundwater management. Results of quality monitoring systems and some outspread sampling on quality of groundwater resources showed the existence of different pollutant sources that due to lack of timely detection cause to severe pollution in groundwater resources. As an instance, Tehran refinery is reason of huge oil compounds leakage to Tehran aquifer. To solve this problem, it is necessary to design and perform a quality monitoring system for detecting pollution leakage from the probable point sources simultaneous with designing refinery facilities. Sampling frequency is one of the main parameters of quality monitoring in groundwater resources. With decreasing the sampling frequency the probability of pollution detection is increased while the related sampling costs are also increased. Increasing the sampling frequency causes to decrease the cost and probability of contamination detection. According to importance of this topic in problem of pollution monitoring, in this paper, sampling frequency intervals of groundwater resources in the contamination event were evaluated. Moreover, appropriate sampling frequency and their location with regards to predefined levels of decision-maker and stakeholders were also suggested. Proposed algorithm in this research was comprised of multi-objective optimization model namely Non-dominated Sorting Genetic Algorithm-II (NSGA-II) and MODFLOW and MT3D simulation models. Performance of the proposed approach was evaluated with available information of groundwater resources in the part of Tehran refinery.
In the proposed methodology, at first, groundwater quality and quantity simulation models calibrated and verified, using the available sampling data of the study area. Total petroleum carbon (TPH) was considered as water quality indicator. Monte Carlo analysis was used to represent the uncertainty in spatial and temporal distribution of water quality. 2000 number of simulations assumed in Monte Carlo analysis, where TPH concentration and leakage location were considered as random variables. NSGA-II was implemented with the aim of minimizing the number of monitoring wells and undetected polluted area (cells) for deriving the location of optimal monitoring wells in the groundwater system. Based on the optimal monitoring wells selected from potential ones, sensitivity analysis was done on sampling frequency. The best sampling frequency helps decision-makers to detect the contamination via desirable monitoring costs. Moreover, support vector machine (SVM) was used to estimate the location of unknown pollution source. In this study, 37 potential monitoring wells, which completely cover the study area, were designed for detecting the probable released contamination from eight large oil tanks.
In this study, groundwater model was developed based on one layer aquifer with 25 and 50 cells in horizontal and vertical dimension, respectively. Each cell has 20 meter length. To generate the different scenarios in the Monte Carlo analysis, 2000 evaluations of groundwater model (MODFLOW and MT3D) were executed. Main parameters in sensitivity analysis were leakage amount, TPH concentration and the number of leaking tanks. Normal distribution for leakage amount with 7 m3/day mean and standard deviation equal to 0.67 was considered. TPH concentration was also generated with normal distribution considering 850 kg/m3 and 0.05 for mean and standard deviation, respectively. Uniform distribution was assumed for generating the number of leaking tanks. Using these parameters, spatial and temporal distribution of groundwater quality indicator was modeled in the system. Results of Monte Carlo analysis are used to find the optimal location of monitoring wells. Water quality sampling in the optimal monitoring wells with frequency of 1, 2, 3, 5, 10 and 15 days was evaluated. Results indicated that the best pollution detection was achieved using 4 monitoring wells with frequency of 10 days. Sampling frequency with 10 and 15 days had the same results while in 15 days frequency, polluted areas may be increased and be detected with some delay. Therefore, sampling frequency of 10 days is recommended for the study area. SVM was trained and validated for estimating an unknown pollution source in the groundwater system. It was revealed that the location of pollution source can be estimated with accuracy of around 47, 62, 86 and 99 percent using 1, 2, 3 and 4 monitoring wells, respectively.