Using Machine Learning to Predict Paleo Topography in the Northern Gulf of Mexico

Paleo topography provides information about a regions paleo climate and environment. The US Gulf Coasts low continental shelf elevation gradient results in dramatic change to the landscape with sea level fluctuation. For this reason, the region is highly valuable in recording topographical and biological change over time. Traditional techniques used to reconstruct submerged paleo topography include producing estimates derived from seismic reflection data. Making estimates the traditional way is a time-consuming process that requires expertise to yield a digitized continuous surface. Unlike traditional methods, machine learning produces models trained on a small sample of observed data in a fraction of the time. Machine learning also gives scientists the ability to add new data to existing models. Here, the Global Predictive Seabed Model (GPSM) is utilized to predict a machine-learned paleo topography model of the Northern Gulf of Mexico. Tradionally interpreted paleo topography served as training and observation data. The relationship between predictors including distance to interpreted paleo valleys and modern bathymetry were first established, and paleo topography was then predicted based on predictor-observation correlation. The result serves as a depiction of the Gulf Coast landscape that once stood above current sea level. It provides insight regarding climate change and the relationship between sea level and regional habitats.