An adaptive unstructured grid for HF radar current mapping based on constrained k-means clustering

Abstract. High-Frequency (HF) radar systems provide high-resolution observations of surface currents in coastal oceans, but their effective representation strongly depends on the spatial discretization used to combine radial measurements. Standard regular grids impose a uniform resolution across the domain, despite the highly heterogeneous sampling geometry of HF radar systems, leading to inefficient use of observations and limited effective resolution in key coastal regions.

We propose a fully automated, non-uniform adaptive gridding framework specifically designed for HF radar applications. The method defines grid nodes directly from the spatial distribution of radar observations, redistributing resolution according to local data availability while preserving a consistent discretization of the domain. The resulting unstructured grid naturally refines resolution in densely sampled near‑coastal areas and adopts coarser spacing offshore. Grid nodes can be associated with either Voronoi polygons or equivalent‑area circular supports, providing flexibility in the geometrical representation of spatial support without altering node placement.

The proposed discretization yields a more realistic spatial representation of HF radar surface currents, reducing artefacts linked to uniform gridding and enhancing the visibility of coastal circulation structures. Comparison with surface drifter observations indicates that this improved spatial coherence is achieved without compromising overall reconstruction skill. The framework provides a radar‑aware basis for current mapping, interpolation, and data assimilation applications where geometrical consistency and adaptive resolution are critical.