Predicting Global Marine Sediment Density Using Machine Learning

Session: Ocean Bottom Seismology – New Data, New Sensors, New Methods [Poster]
Type: Poster
Date: 4/30/2020
Time: 08:00 AM
Room: Ballroom
Description: 

Marine sediment density is of vital importance due to its influence on many geoacoustic parameters (i.e., acoustic impedance, sediment attenuation, sediment grain size, etc.) and is also a key parameter when estimating seismic P- and S-wave speeds.; Many geophysical methodologies require an a priori knowledge of the subsurface – specifically the near-surface, as near-surface variability can often impact deeper structural models if not accounted for. Additionally, near-surface physical properties can amplify or attenuate seismic signals or create reverberations within the uppermost subsurface, hampering seismic analyses by masking important P-to-S conversions from deeper structures and increasing ambient noise levels. Subsurface drilling from DSDP (Deep Sea Drilling Project), ODP (Ocean Drilling Program), IODP (International Ocean Discovery Program), scientific research cruises and various petroleum companies have amassed large quantities of invaluable data regarding the geologic properties with depth beneath the seafloor. These well data yield great vertical constraints at a location, but extrapolating that information away from the well is very difficult. To address this, we have taken a machine learning approach (k-nearest neighbors (kNN) in this instance) to estimate global seabed sediment density at 5x5-arc minute resolution. The kNN algorithm accepts a sparsely sampled observational dataset with densely gridded relatable parameters and predicts statistically ideal estimates where no physical measurements have been made. Results currently show marine sediment density predictions at the 5x5-arc minute resolution that correlate with withheld observations. Moreover, the results indicate where additional in-situ samples are needed in order to improve upon the final prediction.

Presenting Author: Jordan H. Graw

Authors