Hydraulic head uncertainty estimations of a complex artificial intelligence model using multiple methodologies

Abstract

The purpose of this study is to examine the uncertainty of various aspects of a combined artificial neural network (ANN), kriging and fuzzy logic methodology, which can be used for the spatial and temporal simulation of hydraulic head in an aquifer. This simulation algorithm was applied in a study area in Miami - Dade County, USA. The percentile methodology was applied as a first approach in order to define the ANN uncertainty, resulting in wide prediction intervals (PIs) due to the coarse nature of the methodology. As a second approach, the uncertainty of the ANN training is tested through a Monte Carlo procedure. The model was executed 300 times using different training set and initial random values, and the training results constituted a sensitivity analysis of the ANN training to the kriging part of the algorithm. The training and testing error intervals for the ANNs and the kriging PIs calculated through this procedure can be considered narrow, taking into consideration the complexity of the study area. For the third and final approach used in this work, the uncertainty of kriging parameter was calculated through the Bayesian kriging methodology. The derived results prove that the simulation algorithm provides consistent and accurate results.