Conditions for Reliable Divergence Estimates from Drifter Triplets

• Published in: Journal of atmospheric and oceanic technology Vol. 39; no. 10; pp. 1499 - 1523
• Main Authors: Huntley, Helga S., Berta, Maristella, Esposito, Giovanni, Griffa, Annalisa, Mourre, Baptiste, Centurioni, Luca
• Format: Journal Article
• Language: English
• Published: Boston American Meteorological Society 01.10.2022
• Subjects: Algorithms; Aspect ratio; Bias; Computation; Convergence; Coordinate transformations; Criteria; Datasets; Divergence; Drift; Drifters; Drifting buoys; Errors; Estimates; Kinematics; Ocean models; Oceans; Reliability; Sampling; Shape; Strain rate; Thresholds; Triangles; Uncertainty; Velocity; Velocity gradient; Velocity gradients; Vorticity
• ISSN: 0739-0572; 1520-0426
• DOI: 10.1175/JTECH-D-21-0161.1

Horizontal velocity gradients of a flow field and the related kinematic properties (KPs) of divergence, vorticity, and strain rate can be estimated from dense drifter deployments, e.g., the spatiotemporal average divergence (and other KPs) over a triangular area defined by three drifters and over a given time interval can be computed from the initial and final areas of said triangle. Unfortunately, this computation can be subject to large errors, especially when the triangle shape is far from equilateral. Therefore, samples with small aspect ratios are generally discarded. Here we derive the thresholds on two shape metrics that optimize the balance between retention of good and removal of bad divergence estimates. The primary tool is a high-resolution regional ocean model simulation, where a baseline for the average divergence can be established, so that actual errors are available. A value of 0.2 for the scaled aspect ratio Λ and a value of 0.86 π for the largest interior angle θ are found to be equally effective thresholds, especially at scales of 5 km and below. While discarding samples with low Λ or high θ values necessarily biases the distribution of divergence estimates slightly toward positive values, this bias is small compared to (and in the opposite direction of) the Lagrangian sampling bias due to drifters preferably sampling convergence regions. Errors due to position uncertainty are suppressed by the shape-based subsampling. The subsampling also improves the identification of the areas of extreme divergence or convergence. An application to an observational dataset demonstrates that these model-derived thresholds can be effectively used on actual drifter data.