1 of 9

Ocean reanalysis data-driven machine learning prediction of Loop Current eddy movements in the Gulf of Mexico

Anna B. Lowe, Tianning Wu, Ruoying (Roy) He

North Carolina State University

NASA Perpetual Ocean

2 of 9

My research background

Carmel Bay’s circulation

Rockfish larval transport

Lagrangian transport near fronts

ML to identify eddies

3 of 9

Loop Current sheds eddies in the Gulf of Mexico

Loop Current Eddies (LCE):

  • 150 km diameter
  • 1,000 m deep
  • Strong, circular flow (4 knots)

He et al (2003); He et al. (2004); He et al (2005), He and Wilkin (2006), He and McGillicuddy (2008), He et al. (2008); Li and He (2009); Chen and He (2010);

Hyun and He (2010), He et al. (2011 a, b); Yao et al (2012); Zhao and He (2012); He et al. (2018); Zeng and He (2018)

NRL

4 of 9

  • Over 4,000 oil platforms in the Gulf of Mexico affected by LCE eddies
  • Eddies affect: worker safety, drilling and production operations, structures, spill cleanup
  • $100Ms/year lost to oil rig downtime

Loop Current Eddies threaten safety of oil operations

Eddy

Deep Water Horizon 2010

TX

LA

MS

5 of 9

Purpose of this study

  • Develop accurate prediction process of Loop Current Eddy (LCE) shedding process and movement through the oil fields

  • Industry needs:
    • long-range forecast (~3 months)
    • Short-range forecast (days-hours)

6 of 9

CNAPS model output

  • Coupled Northwest Atlantic Prediction System (CNAPS)
  • EnKF data assimilated case
  • ~4 km horizontal resolution

NW Atlantic site: http://omgsrv1.meas.ncsu.edu:8080/ocean-circulation2

7 of 9

27 years of daily average fields (1993-2020)

    • Training data: 17 years (1993-2010)
    • Validation data: 5 years (2010-2015)
    • Testing data: 5 years (2015-2020)

Surface Variables:

    • SST
    • SSH
    • u- and v-velocity / speed
    • relative vorticity

UNET model setup

Weyn et al. (2020)

8 of 9

Work in Progress

  • Industry desires a 3-month forecast
    • When do the eddies form?
    • Where will they travel?
    • How strong when they reach the oil fields?

  • Create a prediction tool that industry will use by harnessing AI2ES partnerships

9 of 9

THANK YOU!

Advisor: Ruoying (Roy) He

NCSU OOMG Lab Members: Tianning Wu, Joe Zambon

AI2ES Project Collaborators: Ryan Lagerquist, Randy Chase, Christopher Wirz, Mariana Cains, Julie Demuth, Imme Ebert-Uphoff

Funding generously provided by the National Science Foundation