Quantifying angular sensitivity and synergistic effects of SIF and vegetation indices for robust winter wheat yield estimation

• The angular sensitivity of SIF and VIs was quantified across winter wheat phenology. • Stronger angular variability of SIF than VIs was identified (CV up to 75.3%). • Angular effects on yield estimation were assessed using a multi-head attention LSTM. • The combined VIs + SIF achieved the highest accuracy and angular robustness. Accurate crop yield estimation is crucial for ensuring global food security. The integration of solar-induced chlorophyll fluorescence (SIF) with traditional vegetation indices (VIs) has shown great potential in yield prediction, yet how data-layer uncertainties propagate and impact model performance remains unclear. Using winter wheat as the target crop, we first quantified the angular sensitivity of SIF and several key VIs within the two-dimensional solar zenith angle (SZA)-viewing zenith angle (VZA) domain and examined its evolution across phenological stages. Subsequently, three models (VIs_Only, SIF_Only, and VIs + SIF) were developed for the Guanzhong Plain using a multi-head self-attention Long Short-Term Memory (LSTM) framework, and their performance dependency on observation geometry was systematically evaluated. The results show that: (1) SIF exhibited markedly stronger angular sensitivity than the VIs (CV up to 75.3%), with the highest instability under dual-oblique geometries (high SZA and VZA) and increasing with canopy development; (2) this angular variability propagated into the models, causing a significant accuracy decline for the SIF_Only model under dual-oblique conditions; (3) SIF provided key phenological complementarity: when VIs saturated (mid-to-late stage), SIF inclusion substantially increased model R 2 ; and (4) the VIs + SIF fusion was the optimal strategy for accuracy and robustness (R 2 = 0.40, RMSE = 568.35 kg/ha). The combined model achieved the highest precision and used VIs as a stabilizer to compensate for SIF ’ s angular uncertainty, maintaining robustness across all geometries. This study quantitatively demonstrates the impact of angular variability on yield estimation and highlights that synergizing SIF and VIs can substantially enhance accuracy and robustness, providing important guidance for future satellite strategies, sensor design, and agricultural applications.