Search and rescue at sea aided by hidden flow structures

The study addresses the urgent need to improve the efficiency of search and rescue (SAR) operations at sea, where rapid, accurate prediction of the short-term movement and accumulation of floating objects can save lives. Traditional SAR relies on probabilistic, trajectory-based modeling combined with Bayesian updates and grid-based search patterns, but these approaches suffer from large uncertainties in initial conditions, environmental forcing, and model errors, leading to error accumulation and slow convergence. The core research question is how to rigorously and quickly assess short-term, material transport variability in highly unsteady and uncertain ocean surface flows to identify where people and objects in the water are likely to converge. The authors propose using objective Eulerian coherent structures (OECSs) to reveal transient attracting profiles (TRAPs), which can be computed from a single snapshot of the velocity field and are hypothesized to act as short-term attractors guiding floating objects. The approach aims to provide actionable, interpretable, and rapidly updatable guidance for SAR asset allocation.

The paper builds on prior work in transport and mixing in geophysical flows, where Lagrangian coherent structures (LCSs) have been effective for predicting tracer behavior over longer timescales in approximately two-dimensional flows. However, operational SAR typically has access to Eulerian (instantaneous) velocity fields from models and remote sensing. Classical Eulerian diagnostics (e.g., streamlines, vorticity, velocity magnitude) are not objective and can be misleading for predicting material transport, especially across different frames of reference. OECSs were introduced as objective, instantaneous limits of LCSs and capture short-term attracting and repelling structures governing immediate material deformation in unsteady flows. Existing SAR practice employs Bayesian methods and ensemble trajectory simulations to handle uncertainty, but these can be time-consuming and yield probability maps that are not easily actionable. The authors position TRAPs—attracting OECSs—as a remedy to these shortcomings, offering an objective, fast, and interpretable alternative that avoids trajectory integration pitfalls and remains robust over short times.

Conceptual and computational framework: TRAPs are defined as attracting objective Eulerian coherent structures (OECSs) in two-dimensional unsteady flows. They are computed from a single snapshot of the velocity field v(x,t) without integrating trajectories. The method uses the rate-of-strain tensor S(x,t) = [∇v + (∇v)ᵀ]/2. Its smallest eigenvalue field s1(x,t) and corresponding eigenvectors (e2 for the largest eigenvalue s2) determine the local instantaneous deformation. TRAP cores are objective saddle points located at negative local minima of s1, and TRAP curves are integral curves everywhere tangent to e2 that emanate from these cores and mark the strongest short-term attracting material lines.

Algorithm (per Methods):

• Input: 2D velocity field v(x,t) on a spatial grid at time t.

1. Compute the velocity gradient ∇v and the rate-of-strain tensor S(x,t) on the grid.
2. Compute eigenvalues s1 ≤ s2 of S and the unit eigenvector field e2 associated with s2.
3. Identify the set Sm(t) of negative local minima of s1(x,t) (TRAP cores).
4. Integrate the ODE dr/dσ = sign(e2(r(σ))·ṙ(prev)) e2(r(σ)), with arclength parameter σ, starting from r(0) ∈ Sm, enforcing local smoothness via the sign term. Terminate when s1(r(σ)) > 0.35 s1(r(0)) or when s1 ≥ 0, ensuring at least 30% of the core’s attraction rate along retained segments.

• Output: TRAP curves and their normal attraction rate s1(x,t) at time t. This construction is objective, robust to small perturbations, and provides complete domain coverage without trajectory integration.

Field experiments:

1. 2014 experiment (HFR-driven nowcast):

• Domain: South of Martha’s Vineyard, near Muskeget Channel; HFR coverage within a hatched polygon; analysis focus area shown in figures.
• Data: High-frequency radar (HFR) surface velocities on a uniform 800 m × 800 m grid spanning longitudes [-70.7979°, -70.4354°] and latitudes [41.0864°, 41.3386°], updated every 30 min. Streamlines used for visualization only.
• Drifters: 68 CODE surface drifters with GPS positions recorded every 5 min. Analysis start: 4 Aug 2014, 17:00 EDT.
• Procedure: Compute TRAPs every 30 min from the updated HFR velocity; quantify attraction by the average distance d of each drifter to the nearest TRAP and its standard deviation sd.

2. 2017 experiment (model forecast-guided targeted release, HFR verification):

• Model: MIT Multidisciplinary Simulation, Estimation, and Assimilation Systems (MIT-MSEAS) 24 h forecast issued 16 Aug, 19:00 (local).
• Targeting: Identify strong TRAPs near trenches of s1(x,t) in the forecast field for 17 Aug from 11:00 onwards; plan release north of a predicted s1 trench to account for model uncertainty (two parallel trenches seen).
• Deployment: Release four CODE drifters at targeted locations around 11:00 on 17 Aug 2017; verify with HFR-derived TRAPs and streamlines where available. Compare drifter evolution to TRAPs computed from HFR fields at later times (e.g., 13:50, 14:20, 14:50) and compute drifter-to-TRAP distances.

3. 2018 experiment (model forecast-guided targeted releases of drifters and manikins):

• Objects: 8 CODE drifters and 4 OSCAR Water Rescue Training manikins (Emerald Marine Products) with GPS tracking.
• Model: MIT-MSEAS 24 h center forecast issued 8 Aug, 20:00. HFR unavailable in the domain due to tower relocation.
• Deployment: Two vessels deployed objects at four designated locations (A–D) plus additional drifters; initial positions and trajectories recorded. Compute model-based TRAPs at 10:15, 11:15, and 12:30 on 9 Aug 2018 and quantify object-to-TRAP distances.

Analyses and diagnostics:

• Compare TRAP locations to instantaneous streamlines and to the horizontal divergence ∇·v; assess cases where divergence suggests non-accumulation but TRAPs attract.
• Evaluate robustness: TRAP attraction demonstrated without explicit inclusion of windage, leeway, or inertial effects, reflecting practical SAR uncertainties. Supplementary analyses compare TRAP-based predictions to ensemble trajectory forecasts, showing convergence of diverse trajectories to nearby TRAPs.
• Time scales: Emphasis on short-term horizons of 2–3 h relevant for SAR nowcasting.

• TRAPs, computed from single snapshots of modeled or measured surface velocities, act as strong short-term attractors for floating objects and reveal hidden one-dimensional attracting structures not evident in streamlines or divergence fields.
• 2014 HFR-driven experiment (68 drifters): Emergence of strong TRAPs around 19:00 EDT organized initially scattered drifters into filamentary structures aligned with TRAPs within ~2 h. Average drifter-to-TRAP distance decreased by about 30% from 18:00 to 19:00 (1.7 km to 0.86 km) and by about 60% within 2 h (to 0.55 km). The standard deviation decreased from 1.5 km to 0.4 km to 0.2 km, indicating increasing clustering along TRAPs. A prominent TRAP (TRAP A) strongly attracted drifters despite being located in a region of positive horizontal divergence, demonstrating divergence can yield false negatives while s1 < 0 along TRAP correctly signals attraction.
• 2017 model-guided targeted release with HFR verification: TRAPs identified from the MIT-MSEAS 24 h forecast guided drifter releases. Subsequent HFR-derived TRAPs confirmed the presence and locations of strong attractors; drifters converged and aligned with TRAPs within ~3 h, even though these structures were hidden in streamline visualizations. Average drifter-to-TRAP distances decreased markedly over the observation period (see Table 2 in the paper).
• 2018 model-guided releases of drifters and manikins: Both object types converged to and aligned with strongest nearby model-based TRAPs within ~2 h, despite differing individual trajectories due to inertia/windage/leeway. Mean object-to-TRAP distance dropped from 1.9 km at 10:15 to 0.8 km at 11:15 and 0.4 km at 12:30, with standard deviations decreasing from 1.0 km to 0.8 km to 0.5 km.
• Robustness and practicality: TRAP predictions were effective without explicit parameterization of windage or inertia and provided complete spatial coverage within available velocity domains, unlike Lagrangian methods that can exit data-supported regions. TRAPs can be computed and updated rapidly, providing actionable, localized guidance for SAR asset allocation on critical short time scales (≤ 2–3 h).

The findings directly address the central challenge of rapid, reliable short-term prediction of floating object convergence in uncertain, unsteady coastal flows. By leveraging objective Eulerian coherent structures, TRAPs provide observer-independent, instantaneous attractors that organize material transport over the next few hours. Across three ocean experiments using both HFR-measured and model-assimilated velocities, TRAPs consistently predicted where drifters and manikins would accumulate and align, even when classical diagnostics (streamlines, horizontal divergence) provided no clear guidance or were misleading. This demonstrates that TRAPs capture the dynamically relevant skeleton of short-term transport and can substantially narrow search areas quickly. For SAR, this translates to more focused, interpretable, and rapidly updatable search paths, potentially improving rescue outcomes within the time window where survival probability is highest. TRAPs complement existing probabilistic SAR frameworks by providing deterministic, localized targets for resource allocation without costly trajectory ensembles, and by remaining robust to common operational uncertainties (release time/location, windage, inertia, and model errors) over the short horizons of interest.

This work introduces and validates Transient Attracting Profiles (TRAPs) as objective, easily computable, and highly informative short-term attractors in ocean surface flows, suitable for guiding search and rescue and hazard response operations. TRAPs, derived from the instantaneous rate-of-strain tensor, were shown in three field experiments to attract and align floating drifters and manikins within 2–3 hours, while remaining hidden to conventional Eulerian diagnostics. The approach offers fast, interpretable, and localized guidance for search-asset allocation without reliance on trajectory integration. Potential future directions include: integrating TRAP computation into operational SAR toolchains; fusing TRAPs with probabilistic frameworks and observational updates; extending methodology to incorporate windage/inertial object models when available; assessing performance across diverse coastal and open-ocean regimes and seasons; exploring three-dimensional or surface-wave-influenced extensions; and leveraging expanded real-time sensing (e.g., HFR networks, satellite-derived velocities) for broader coverage and timelier updates.

• Temporal scope: TRAPs are instantaneous structures with predictive power over short times (typically less than about 6 hours in this context). Their reliability diminishes over longer horizons as the velocity field evolves.
• Data dependence: Accuracy depends on the quality and availability of velocity fields. HFR coverage can be limited or unavailable; model fields carry assimilation and forecast uncertainties.
• Object physics: While experiments show robustness without explicit windage/leeway/inertia modeling, not accounting for these effects may limit accuracy for some object types and environmental conditions.
• Spatial coverage in Lagrangian comparisons: Trajectory-based methods can leave the domain of reliable velocity data; while TRAPs avoid this by construction, comparisons across methods may be constrained by differing data support.
• Operational integration: Real-time deployment and decision support require streamlined data pipelines and visualization tools not evaluated here.