An Artificial Neural Network Application for Predicting the Aspect Ratio and Permeability Images of a Heterogeneous Carbonate Aquifer in South Florida

Abstract Characterizing carbonate aquifers presents significant challenges, especially when it involves determining groundwater flow zones at the field scale. The combination of seismic data and well logs, along with a suitable processing approach, supports this effort. We processed high-resolution reflection seismic data obtained from the Port Mayaca aquifer, located in Martin County, South Florida. We applied a forward solution algorithm to invert this data for impedance. The resulting images displayed anomalies associated with dense, tight limestone layers. These anomalies, combined with Formation MicroImager (FMI) data and micro-resistivity logs, lead to the existence of fractures. To enhance our analysis, we applied an artificial neural network (ANN) using seismic attributes from the seismic waveforms and the impedance data. The initial image produced by the ANN algorithm illustrates the pore space based on the aspect ratio log, which serves as the target data for the neural network model. The Python library Keras provided the framework needed to develop these network models. The application of the Keras software allows us to create images of pore aspect ratio, Stoneley (ST) wave permeability, and intrinsic permeability. The interpretation and analysis of these images revealed the connectivity of the pore space between narrow and wide channels, as well as the flow paths linked to the permeable zones. The ST wave permeability image highlights the major flow zones present in the inter-well region. The results of the data analysis demonstrate that a flow zone structure, defined by the aspect ratio, is part of a groundwater channel. We found that a permeable zone identified in the ST permeability image represents the total flow created by the interconnected vugs and fractures. We generated ELAN intrinsic permeability images to examine the aquifer’s pore structure, and we found that the areas with low-intrinsic permeability correlate with locations of dense, fractured limestone layers.