Improve the results of the DRASTIC model using artificial intelligence methods to assess groundwater vulnerability in Ramhormoz alluvial aquifer plain

Groundwater pollution is a complex and full of uncertainty process, on a regional scale. Development of an integrated method for assessing groundwater vulnerability, can be efficient in order to optimized management and protection of them. Because of fertile soil and sufficient water resources, Ramhormoz plain is suitable area for agriculture that by development of agriculture, use of chemical fertilizers and pesticide, this plain always is at risk of contamination. One of the suitable approach to prevent groundwater contamination, identify areas of potential contamination. The aim of this study is to produce vulnerability map of Ramhormoz plain alluvial aquifer using DRASTIC model, and then use artificial intelligence techniques to improve the results of the DRASTIC model. Due to the importance of groundwater resources in the study area that are used for various purposes including agriculture, Aquifer vulnerability study and protect these areas for development and management of water resources is essential.
In this study, first, vulnerability evaluation of Ramhormoz alluvial aquifer plain was performed using DRASTIC model and in the following, artificial intelligence methods was used to optimize the model. DRASTIC model includes the following parameters: depth to water table, net recharge, aquifer media, soil media, topography, impact of vadose zone and hydraulic conductivity that are effective in groundwater vulnerability assessment. This method, based on the standard weights of DRASTIC model and obtained layers for each of the seven parameters, calculates the amount of aquifer vulnerability. After preparation of the layers, vulnerability of Ramhormoz alluvial aquifer plain was determined using drastic model. Also the groundwater vulnerability map and DRASTIC index was calculated for the entire area. In order to evaluation of accuracy of the obtained results from the model, nitrate concentration data existing in groundwater have been used for verification. Following In order to improve results, DRASTIC model was integrated by artificial neural networks, fuzzy logic (Sugeno and Mamdani) and Adaptive Neuro-Fuzzy Inference System methods and four vulnerability maps was obtained using different models of artificial intelligence.
the groundwater vulnerability map toward the contamination was prepared by the division into three vulnerability ranges including low, medium and high and DRASTIC index was calculated for the entire area between 48 and 156. Correlation coefficient 0.97 between DRASTIC index and nitrate concentration reflects the relatively good accuracy of this method. Also, the results of the implementation of these models showed that the used artificial intelligence models have the ability to improve the primary DRASTIC model results. By comparing the results of the models can be concluded that the Adaptive Neuro-Fuzzy Inference System model has the best result.
The determination coefficient, R2, for the Adaptive Neuro-Fuzzy Inference System, neural networks and Mamdani fuzzy and Sugeno fuzzy models, is 0.99, 0.94, 0.98 and 0.87 respectively. According to the final model, South- Southeast regions have the highest potential for contamination.