Improvement of groundwater contamination vulnerability assessment using the adaptive neuro-fuzzy inference system with metaheuristic optimization algorithms

Abstract

This study aims to enhance the assessment of groundwater contamination-vulnerability using the adaptive neuro-fuzzy inference system (ANFIS) combined with metaheuristic optimization algorithms (MOAs) such as a genetic algorithm (GA), differential evolution (DE), and particle swarm optimization (PSO). The assessment of groundwater contamination vulnerability used two ways of the nitrate contamination and the contamination vulnerability index based on the DRASTIC assessment method. Miryang City of Gyeongsangnam-do Province in Korea was selected as a case study area since it had recently undergone groundwater nitrate pollution due to intensive agricultural and human activities. DRASTIC factors of Depth to water, net Recharge, Aquifer media, Soil media, Topographic slope, Impact to vadose zone, Hydraulic Conductivity and Land-use were transformed into numerical ratings and used as input variables of ANFIS-MOAs. Original DRASTIC vulnerability indices were calculated using 8 DRASTICL factors and their weights of hydrogeological data collected from 95 wells in the unconsolidated aquifer of the study area. The nitrate-nitrogen data (NO3-N) were also obtained from the monitoring wells, and natural log- transformed for the normal distribution. 95 DRASTIC ratings and NO3-N data were separated into 2 parts for ANFIS modeling: 70% training data and 30% testing data. The performance of ANFIS-MOAs was evaluated by several statistical evaluation criteria: mean absolute error (MAE), root mean square error (RMSE), correlation coefficient (R) and Receiver Operating Characteristic (ROC) under the curve (AUC). The testing results of the nitrate concentrations and the contamination vulnerability indices indicated that ANFIS-PSO (MAE= 0.323 ; RMSE= 0.436; R= 0.903; AUC= 0.95/ MAE = 20.752 ; RMSE = 37.670; R = 0.805 ; AUC= 0.921) implemented the more superior performance than ANFIS-DE (MAE= 0.338; RMSE= 0.464; R= 0.897; AUC= 0.912/ MAE= 23.149 ; RMSE=44.352; R=0.715; AUC=0.88), ANFIS-GA (MAE= 0.390; RMSE=0.561; R= 0.839; AUC= 0.88 / MAE=23.654 ; RMSE= 45.964; R=0.691 ; AUC= 0.865) and original ANFIS (MAE= 0.580; RMSE= 0.836; R= 0.60; AUC= 0.769/ MAE= 46.813; RMSE= 63.935 ; R= 0.468 ; AUC= 0.704) in predicting the nitrate contamination vulnerability and conditioned contamination vulnerability index, respectively. ANFIS-PSO model also exhibited very high susceptibility to the nitrate concentration and the vulnerability index maps at the area near the Nakdong River and the center of Miryang city, although other models of ANFIS showed relatively similar patterns for the spatial distribution of the highly vulnerable hotspot areas. ANFIS-GA and ANFIS-DE models were also useful for the prediction of nitrate data and the assessment of groundwater contamination vulnerability, compared to Original DRASTIC or Original ANFIS model. It is concluded that ANFIS-MOA models will also produce the excellent results in assessing groundwater contamination vulnerability in other areas of the world as well as the Miryang City of Korea.