Accurate field-scale soil moisture is essential for hydrological processes such as infiltration, land–atmosphere exchange, and agricultural water management. UAV-borne L-band radiometry offers a promising intermediate scale between in situ measurements and satellite observations, but retrieval remains ill-posed due to uncertainties in vegetation attenuation, surface temperature, and sub-footprint heterogeneity. This study develops an uncertainty-aware Bayesian retrieval framework that integrates dual-polarized UAV L-band brightness temperature with RGB and thermal infrared information through footprint-consistent priors. Optical fraction cover, thermal state, and texture descriptors are used to constrain vegetation optical depth and its uncertainty at the scale of the radiometric footprint. The method was evaluated over heterogeneous cropland in Pengzhou, China, using independent calibration (4 scenes, ∼1.3 ha) and validation datasets (6 scenes, ∼3.3 ha). The proposed approach reduced RMSE from ∼0.07 to ∼0.04 m³ m^-3 and largely eliminated the systematic dry bias of the conventional τ–ω inversion. Analysis further shows that sub-footprint heterogeneity primarily increases uncertainty in vegetation attenuation, leading to representation error in soil moisture retrieval. These findings highlight that retrieval performance is fundamentally constrained by observation scale and surface heterogeneity. Overall, the study demonstrates that physically informed multi-source priors can improve both accuracy and interpretability, providing a pathway toward more reliable field-scale soil moisture estimation for hydrological applications.