Residual Viewing-Geometry Dependence in Provisional NISAR L-Band GCOV Backscatter over a Steep Mediterranean Pine Forest

Radiometric terrain correction normalises an illuminated-area contribution to NISAR GCOV backscatter, but intrinsic angular scattering can remain. We evaluated terrain-associated variation in seven provisional granules acquired during June–July 2026 over Paphos Forest, Cyprus, using 422,996 stable Pinus brutia cells on a 20 m analysis grid and an independent 1 m ALS terrain survey. Generalised additive mixed models were assessed with buffered spatial cross-validation and a spatial cluster bootstrap. With the supplied-look log-bias adjustment, conditional-path spans are approximately 1.3 dB in HH and 2.2 dB in HV; alternative ENL multipliers yield 1.280–1.785 dB and 1.598–2.228 dB, respectively. Ascending-minus-descending differences reverse sign with viewing geometry, and the pattern recurs in three acquisition pairs despite a change in the sign of their humidity contrast. Operational-DEM geometry predicts the paired difference better than the tested ALS geometry, although sub-pixel registration narrows the gap. DEM-disagreement and polarisation-specific geometry terms retain small predictive gains across alternative spatial partitions. These results describe associations at one site and product version. They do not separate intrinsic scattering, normalisation error, and acquisition effects; orbit direction, track, and overpass time remain confounded.