Name: Wenjing Liu
Date: 07/27/2026
Time (EST/EDT): 01:00 pm
Location: AREL Lecture Hall
Remote Access: email: mees@umd.edu

Committee Chair: Victoria Coles
Committee Members: Jian Zhao Kenny Rose Hongsheng Bi Weifeng (Gordon) Zhang
Dean’s Representative: James Carton

Title: Ocean Heat Content and Marine Heatwaves on the Bering Sea Shelf:
Processes and Predictability

Abstract: The ecosystems of the eastern Bering Sea shelf are highly sensitive
to ocean temperature and have been increasingly adversely affected by a regime
shift toward a warmer state with more frequent and prolonged marine heatwaves
(MHWs). Thus, reliable prediction of MHWs is essential to provide advance
warning for management action. This dissertation pursues this objective by
basing a prediction model in the physical mechanisms governing the shelf
thermal variability. The first study examines the cross-isobath exchange that
carries oceanic heat onto the shelf. Strong, persistent shelf-basin exchange is
concentrated at submarine canyons, with on-shelf transport along their northern
flanks and off-shelf transport along their southern flanks, arising from a
break in geostrophic balance where the along-slope Bering Slope Current (BSC)
encounters canyons. This exchange is most pronounced in winter, together with
eddy-induced exchange accounts for the full character of shelf-basin exchange.
The second study incorporates this oceanic heat supply within the full shelf
heat budget and examines the drivers of its interannual variability.
Year-to-year variability of ocean heat content is governed primarily by air-sea
heat flux as a damping term, while ocean advective heat flux modulates the heat
budget by delivering heat from the Pacific inflow. Both processes are organized
by the Aleutian Low through its influences on anomalous meridional winds, which
links local variability to large-scale climate patterns. The third study
addresses the local predictability of thermal extremes. Long Short-Term Memory
networks trained on physically coherent data predict sea surface temperature
and MHWs with the best performance at short-to-medium lead steps. Feature
attribution identifies the predictors that contribute to the model prediction,
which prove broadly consistent with the atmospheric and oceanic drivers
established in the preceding studies. Together, these studies trace a physical
system from mechanism to prediction, linked throughout by the BSC and the
Aleutian Low. This continuity allows physical understanding to inform
forecasting. Because the drivers of the heat budget also contribute to the
model, the predictions are interpretable. The framework developed here is
transferable to other high-latitude shelves where warming and marine heatwaves
threaten vulnerable ecosystems, offering a route toward the physically based
early warning that resource managers and coastal communities increasingly need.