William Duy and P.A.Ty Ferré
Department of Hydrology and Atmospheric Sciences
The University of Arizona
The purpose of this project was to determine the feasibility of using machine learning to predict the saturated hydraulic conductivity and which observations would be most important for the University of Arizona Tech Park site. This was done by analyzing samples from three boreholes drilled to a depth of 85 feet. The samples were collected in five-foot sections. Each section was split into samples based on variations in physical parameters. The samples were then analyzed to determine initial saturation, particle size distribution, dry bulk density, porosity, and saturated hydraulic conductivity. The data were then split into testing and training data and analyzed with Python coding to determine the R2 value of the predicted K values with all combinations of observations collected. From this analysis it was determined that the point of diminishing returns was reached when using four or five of the collected observations. The models generally show increasing R2 values as more were added, with smaller increases as more variables are added to the model with a maximum value of approximately 0.95. The models agree that while porosity and dry bulk density can be used to help improve the model, particle size distribution data tend to be more important for predicting saturated conductivity values.