USING A MACHINE-LEARNING APPROACH TO PREDICT CAVE LOCATIONS: IMPLICATIONS FOR SPELEOGENESIS

Caves are an important part of karst landscapes and protect fragile ecosystems, culturally significant artifacts, and a wide range of additional resources. Because cave environments are vulnerable to disturbance and degradation, knowing their locations can help minimize disturbance; therefore, understanding cave distribution on the landscape is critical. We used a machine-learning approach, MaxEnt, to predict the relationship between environmental variables and cave entrance locations.

Using data for 10 environmental variables (elevation, slope, aspect, regional waterways, lithology, vegetation, soil, faults, folds, and springs) in Grand Canyon National Park, we developed our model using 391 confirmed cave entrances. Our results showed high sensitivity (AUC=0.862) of the model and therefore good predictive capabilities; lithology (26.4), springs (19.8), slope (18.9), and faults (14.0) had the greatest percentage of contribution to the model.

This model shows that these environmental variables can be used to predict cave location. In addition, it suggests relationships between the selected variables and cave formation. It was no surprise that lithology was the most important variable, because of the role of carbonates in hosting cave formation. Springs occur where groundwater intersects the landscape surface along preferential flow paths and may be indicative of active speleogenesis. Slope’s importance is closely tied to the nature and origin of the Grand Canyon’s topography. Most observed caves in the canyon occur on steep canyon walls, which, along with waterways’ relatively low contribution (4.4 percent), could suggest that the characteristics of surface water are not as important to cave formation in the Grand Canyon. Rather, cave formation is isolated from the surface hydrology, indicating hypogene speleogenesis and the waterways that are responsible for forming the canyon are only responsible for exposing existing caves to observation. Faults can provide preferential vertical flow paths that allow groundwater to move between strata and through carbonate bedrock in the early stages of speleogenesis. This modeling effort shows promise in improving our understanding of cave distribution on the landscape.