Prediction of nodule abundance in the northwestern Pacific using Gaussian process regression with multisource geological data

Polymetallic nodules, rich in cobalt (Co), nickel (Ni), copper (Cu), manganese (Mn), and rare earth elements (REEs), are important marine mineral resources with the utmost capacity for commercial employment in the future. Recently, the discovery of cobalt-rich nodules with high abundance in the western Pacific Ocean has attracted significant attention, but their distribution remains a vital geological problem to be solved in marine mineral resource exploration. Based on the multi-source geological data acquired from acoustic, optic and geological sampling in the northwestern Pacific Ocean, the regression model for nodule abundance is developed via Gaussian process regression (GPR). Compared to other machine learning approaches, such as stepwise linear regression (SLR), regression trees (DTR), ensembles of regression trees (EDTR) and support vector machine (SVM), intelligent prediction of nodule abundance by GPR is achieved with higher accuracy. Moreover, the spatial pattern of prediction error is obtained simultaneously by GPR, showing strong capability in uncertainty quantification. The predicted result reveals that the high-abundance (∼30.0 kg/m 2 ) nodules are mainly distributed in deep-sea basins around several seamounts, where Antartic Bottom Water (AABW) is developed and spatially coupled with the Co-rich crust distributed on seamounts in the targeted study area. The deep-sea basins between the Magallan and Marcus-Wake seamounts are considered potential areas for further exploration of Co-rich polymetallic nodules in the western Pacific Ocean.