Impact of Geophysical Fields on Deep Learning-based Lagrangian Drift Simulations
Summary: arXiv:2604.03292v1 Announce Type: cross
This article assesses the influence of different Eulerian geophysical input fields on Lagrangian drift simulations using DriftNet, a learning-based method designed to simulate Lagrangian drift on the sea surface. The study encompasses two main experiments: a fully numerical experiment (Benchmark B1) and a real-world drifters-based experiment (Benchmark B2).
Study Overview
The experiments are conducted in two regions characterized by distinct ocean dynamics: the North East Pacific and the Gulf Stream regions. The primary goal is to evaluate the performance of DriftNet using three different metrics:
- Separation distance between simulated and ground-truth trajectories
- Normalized cumulative Lagrangian separation
- Autocorrelation of Lagrangian velocities
Results from Benchmark B1
In the first benchmark, results indicate that the combination of assimilated sea surface currents (SSC) with fully observed sea surface height (SSH) leads to the most significant improvements in trajectory simulation. Specifically, this configuration:
- Reduces separation distance by over 50%
- Significantly decreases normalized cumulative Lagrangian separation
- Improves metrics related to velocities autocorrelation functions compared to the baseline using SSC alone
Conversely, the inclusion of sea surface temperature (SST), either alone or in conjunction with SSC, generally results in degraded performance.
Insights from Benchmark B2
The second benchmark explores the application of satellite-derived data. By utilizing satellite-derived SSH, Ekman, and wind velocities, the simulation of surface drifters’ trajectories shows marked improvement, particularly in the North East Pacific. Additionally, in the Gulf Stream region, the combination of satellite-derived SST with reanalysis-based SSC configurations facilitates better trajectory simulations.
Conclusion
Overall, this study highlights the added value of integrating multiple geophysical fields to enhance Lagrangian drift simulations. The findings emphasize that while certain geophysical inputs can significantly improve simulation accuracy, others may hinder performance. The insights gained from both numerical and real-world experiments pave the way for future advancements in oceanographic research and applications.