Hydraulic conductivity and uncertainty analysis of between-models and input data by using Bayesian model averaging of artificial intelligence model

The estimation of hydraulic conductivity is one of the most important part of hydrogeological studies which is important in groundwater management. But due to practical, time or cost constraints, direct measurement is difficult. Hence, the using artificial intelligence models with low cost and high performance can be an appropriate alternative for this purpose. Since input data and different training techniques in these models are the most important source of uncertainty, the effect of various sources of uncertainty in output should be considered. In this research a Bayesian Model Averaging (BMA) are developed which includes the model combination of artificial neural network, fuzzy logic and neuro-fuzzy to estimate hydraulic conductivity and uncertainty analysis. In the BMA model, the weight of the models is determined by the Bayesian information criterion (BIC), and the within-model variance, steam from the uncertainty of input data and the between-model variance steam from uncertainty associated with the nature of the artificial intelligence model are calculated. In this study, the developed method has been applied to estimate the hydraulic conductivity in the Urmia aquifer. The results show that although the determination coefficient of BMA is not higher than the determination coefficient of the best model, the output of the BMA is the result of assigning weights that take into account the uncertainty between the models and the input data. Also, the effect of groundwater level variation on estimated hydraulic conductivity from pumpage test up to 2015 was evaluated and the result indicated an insignificant changes in hydraulic conductivity.