78-1 Expanding karst vulnerability tools: exploring a machine learning approach to refine index-based vulnerability modeling of karst terrain

Session: New Frontiers in Cave and Karst Science

Presenting Author:

Authors:

Abstract:

Index-based vulnerability models have long been used to assist with land management decisions and planning. From mitigating groundwater contamination to identifying zones to avoid building, these models provide a basis for identifying where karst systems are most vulnerable to disturbance. While valuable, these models often rely on broad assumptions. As a result, researchers have consistently found it important to modify these models to suit local circumstances. Using different methodologies complicates comparing areas and providing guidance to land managers for landscape scale needs. Machine learning approaches may provide tools to create a consistent process across landscapes.  

Here we assess the use of MaxEnt, a machine learning approach to predicting feature distributions across a landscape to assist in refining the COP vulnerability tool in two areas: central Kentucky and Southeastern New Mexico. The COP model looks at three primary factors that influence groundwater vulnerability: concentration of flow (C), overburden (O), and precipitation (P). Using publicly available geospatial datasets coupled with cave entrance location data, we focused on modifying the C- factor through creating a predictive model of cave locations as a proxy for likely areas of increased infiltration. This approach resulted in an increased importance of faults as zones of infiltrations, more closely matching observations in the subsurface.  

Geological Society of America Abstracts with Program. Vol. 57, No. 6, 2025

doi: 10.1130/abs/2025AM-9880

© Copyright 2025 The Geological Society of America (GSA), all rights reserved.