Dual-polarimetric SAR is widespread, and recent studies show that scattering components can be derived without full-pol data (e.g., forest power models; dihedral/surface index factorization). Yet many methods are scene-targeted or rely on intensity-space indices and Euclidean fitting. We propose TD4C—a compact, parameter-free decomposition that treats the normalized C2 as a target density, partitions residual power with a Bures-Gibbs gate, and resolves the ground split on a logit-geodesic line. This avoids negative components, respects polarization purity, and remains strictly dual-pol and SPAN-conservative. A scene-tied phase gauge establishes a common reference emphasizingin-phase cross-channel information. Each pixel’s 2 × 2 sub-covariance is recast as a unit-trace, positive-semidefinite “target density,” enabling information-geometric distances. Helical scattering is isolated first by a linear projection of the quadrature cross term, while the remaining power is partitioned between ground-like and volume-like mechanisms by comparing the target density to simple dual-pol templates through the Bures angle; the distance difference is mapped to allocations with a purity-conditioned logistic, yielding decisive assignments for near-deterministic pixels and conservative behaviour for mixed states. The ground share is then split into oriented-dihedral and trihedral using a logit-geodesic Dihedral Index (DBLI) that blends a scene-normalized Pauli cue with the gate’s ground confidence; a directional-entropy term limits context influence to ambiguous yet ground-like cases. An NESZ mask and explicit metadata complete the workflow. TD4C attains a mean RMSE of 0.167 over helix, volume, oriented-dihedral, and trihedral fractions, with strict SPAN closure and realizability (unit trace, PSD). The result is a reproducible, information- geometric dual-pol decomposition suited to urban, forestry, and hazard-monitoring applications.
Roy et al. (2026) studied this question.