Premium
Predicting Global Marine Sediment Density Using the Random Forest Regressor Machine Learning Algorithm
Author(s) -
Graw J. H.,
Wood W. T.,
Phrampus B. J.
Publication year - 2021
Publication title -
journal of geophysical research: solid earth
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.983
H-Index - 232
eISSN - 2169-9356
pISSN - 2169-9313
DOI - 10.1029/2020jb020135
Subject(s) - seafloor spreading , geology , drilling , sediment , scientific drilling , marine spatial planning , sampling (signal processing) , scale (ratio) , deep sea , oceanography , geomorphology , computer science , mechanical engineering , physics , filter (signal processing) , quantum mechanics , engineering , computer vision
Marine sediment density is of vital importance due to its influence on many geoacoustic and nongeoacoustic parameters, including but not limited to: acoustic impedance, sediment grain size, sediment attenuation, gravity, and isostatic and dynamic responses of oceanic crust. Additionally, it plays a fundamental role on geomechanical behavior which is important for pore pressure generation, seafloor slope stability, and seafloor infrastructure. Subsurface drilling from the Deep Sea Drilling Project, Ocean Drilling Program, International Ocean Discovery Program, a myriad of scientific research cruises, and a number of petroleum companies have amassed large quantities of invaluable seafloor geologic parameters. These point sampling location data yield accurate vertical constraints at a given location, but extrapolating that information away from the location is very difficult. To address this, we have taken a machine learning approach (the random forest regressor [RFR] in this instance) to predict global seafloor density and its associated uncertainty, both at a 5 × 5‐arc minute resolution. The RFR algorithm accepts a sparsely sampled observational data set with densely gridded relatable predictors and predicts statistically optimal estimates where no physical measurements have been made. The final prediction has median, mean, and root‐mean‐square errors of 0.058, 0.076, and 0.1009 g/cm 3 , respectively. Results show density predictions that coincide with expected lateral density variation (albeit at a very coarse spatial scale) with respect to ocean depth and uncertainties that are low for most of the seafloor. Finally, a parametric isolation indicates locations where additional samples are needed to improve the overall prediction.