Spatial Estimation of Transmissivity Using Artificial Neural Network

Abstract

AbstractThe spatial interpolation of transmissivity and other hydrogeological parameters, based on data measured or observed at well locations, is necessary for the numerical simulation of various ground water flow and transport problems. The application of artificial neural network technology for this purpose has been investigated for the transmissivity of the Dammam Formation Aquifer in Kuwait. It has been found that reasonable estimates of this parameter can be obtained with the help of a network that uses the location coordinates, altitude of the top of the Dammam Formation, and logarithm of the total dissolved solids content of the ground water of the aquifer as the input neurons, and the logarithm of the transmissivity value as the output neuron. A better estimate is obtained with a model that takes into account the information on the occurrence of circulation loss in the Dammam Formation. This information, however, can only be obtained at a site where a well has penetrated the Dammam Formation, and unlike the altitude of a stratigraphic horizon and total dissolved solids content, cannot be extrapolated with any certainty away from a well site. This model, therefore, has limited value in estimating the transmissivity of the Dammam Formation. The artificial neural network models are found to give better estimates of transmissivity of the Dammam Formation in Kuwait when compared with the more commonly used kriging model.