Hydraulic head and groundwater Cd-111 content interpolations using empirical Bayesian kriging (EBK) and geo-adaptive neuro-fuzzy inference system (geo-ANFIS)

Abstract

In this study, hydraulic head and Cd-111 interpolations based on the geo-adaptive neuro-fuzzy inference system (Geo-ANFIS) and empirical Bayesian kriging (EBK) were performed for the alluvium unit of Karabaglar Polje in Mugla, Turkey. Hydraulic head measurements and 111Cd analyses were done for 42 water wells during a snapshot campaign in April 2013. The main objective of this study was to compare Geo-ANFIS and EBK to interpolate hydraulic head and Cd-111 content of groundwater. Both models were applied on the same case study: alluvium of Karabaglar Polje, which covers an area of 25 km(2) in Mugla basin, in the southwest of Turkey. The ANFIS method (called ANFISXY) uses two reduced centred pre-processed inputs, which are cartesian coordinates (XY). Geo-ANFIS is tested on a 100-random-data subset of 8 data among 42, with the remaining data used to train and validate the models. ANFISXY and EBK were then used to interpolate hydraulic head and heavy metal distribution, on a 50 m(2) grid covering the study area for ANFISXY, while a 100 m(2) grid was used for EBK. Both EBK- and ANFISXY-simulated hydraulic head and Cd-111 distributions exhibit realistic patterns, with RMSE < 9 m and RMSE < 8 mu g/L, respectively. In conclusion, EBK can be considered as a better interpolation method than ANFISXY for both parameters.