Seamount age prediction machine learning model based on multiple geophysical observables: methods and applications in the Pacific Ocean

Seamount ages are important for understanding crust-mantle interactions and exploring sea bottom ore resources. Rock sampling and laboratory measurements are time-consuming and expensive, and are currently the main methods for dating seamounts. Thus, the ages of many seamounts in the Pacific Ocean are still unknown. To address this problem, gravity anomalies, magnetic anomalies, oceanic crustal ages, sediment thicknesses and seamount heights are chosen as the input parameters for seamount age prediction based on the potential relationship between geophysical observables and seamount ages. A Back-propagation (BP) neural network is constructed using the currently known seamount ages in the Pacific Ocean. Then, the weight and threshold of the BP network are optimized by a Genetic algorithm (GA); finally, the GA-BP model for seamount age production is derived. In addition, Convolutional neural network (CNN), BP model, and Support vector regression (SVR) methods are also used to predict seamount ages. The uncertainties in the prediction results decrease in the order of the GA-BP model, CNN, BP model, and SVR methods. The RMSE of the GA-BP prediction results is 10.22 Ma, and R2 is 0.90.