Paper No. 230-6
Presentation Time: 2:50 PM
PREDICTING THE TIMING AND MAGNITUDE OF SDG IN THE SOUTH ATLANTIC BIGHT USING MACHINE LEARNING
Recent work describes pulses of SGD discharging 10-20 km offshore in the South Atlantic Bight and links these pulses to sustained prevailing winds that led to declines in sea level. These results represent a major shift in the way we understand SGD, but they were limited in time and did not have the resolution to detect porewater exchange or SGD velocities less than 1 cm/day. Over the past three years, new monitoring instrumentation has been implemented with greater thermal resolution on the upper 20cm of benthic sediments as well as sediment level sensors. Our study uses thermal time-series measurements to solve the inverse problem for SGD velocities and estimate hypokymatic fluxes. Using this new data, our model has the capacity to estimate the timing, depth, and magnitude of hypokymatic fluxes and the SGD velocity with low RMSEs. From 2018 through 2020 we observed 2-3 pulses of SGD occurring each summer ranging in velocities from 1cm/day to 8cm/day. These pulses occurred simultaneously with sustained wind events and a decline in sea level. The largest of the pulses took place during hurricanes whose path traveled over or near the field site. We then used our calculated velocities as our training and test data in a machine learning (ML) model with input data sourced from NOAA, USGS, and other local agencies. We were able to train our ML model on spring and summer months in 2019 to accurately predict two pulses of SGD during a hurricane that Fall. We can expand the model training set to find the main drivers of SGD at our field site, using correlations to identify the input variables that have the largest impact on predicting the timing and magnitude of future discharge events.