A BPDF database retrieved using multi-angle, multispectral and polarimetric observations from POLDER-3 and DPC-2

Abstract. The bidirectional polarization distribution function (BPDF) characterizes the dependence of land-surface polarized reflectance on illumination and viewing geometry and provides a key surface boundary condition for polarimetric retrievals of aerosol properties. In recent years, the increasing number of polarimeters in orbit has stimulated growing interest in polarized surface reflectance. Reliable multisensor BPDF databases are therefore urgently needed. However, deriving surface polarized reflectance that is comparable across spectral bands and sensors remains challenging. Earlier databases generally corrected only for molecular scattering and adopted the 865 nm results as the final surface polarized reflectance, assuming that aerosol effects were negligible under low aerosol loading and that a single band adequately represented surface polarization. This approach neither explicitly accounted for aerosol effects nor exploited the weak spectral dependence of land-surface polarized reflectance. In fact, even under low-aerosol-loading conditions, across the POLDER-3 and DPC-2 observations used in this study, the relative differences between surface polarized reflectance obtained using molecular-only correction and that obtained by accounting for both molecular and aerosol effects exceeded 10 % in approximately 60 %–75 % of the observations at 490, 670 and 865 nm. Moreover, global BPDF studies have relied primarily on POLDER observations. Because most existing studies use observations from 2005–2013, their results may not adequately represent current land-surface conditions. Newly operational polarimeters therefore require an atmospheric-correction method applicable across-sensors to update existing BPDF records and support future BPDF studies. In addition, the effects of mixing among different surface types on BPDF remain poorly understood. Here, we developed a unified atmospheric-correction framework that uses Aerosol Robotic Network (AERONET) aerosol optical depth and vector radiative transfer calculations together with vector radiative-transfer calculations to account for the effects of both atmospheric molecules and aerosols. After correcting the multiband observations, the spectral consistency of the multiband surface polarized reflectance was substantially improved. On this basis, we used weighted fusion to derive a common estimate of surface polarized reflectance from the three corrected bands. The framework was applied to POLDER-3 observations from 2005–2013 and DPC-2 observations from 2022–2023, producing two quality-flagged BPDF databases. The database is publicly available and registered at doi.org/10.57760/sciencedb.015ek. For the cross-sensor evaluation, directional observations acquired under similar surface and observation conditions were matched between the two sensors. The matched directional observations yielded correlation coefficients of 0.60 to 0.87 across 12 International Geosphere–Biosphere Programme (IGBP) land-cover classes, showing clear positive correspondence under similar surface and observation conditions despite the nearly decade-long separation between the two sensor records. The subpixel analysis further showed that the higher-resolution DPC-2 observation sequences were generally more homogeneous, with a mean dominant land-cover fraction 5.8 percentage points higher than that of POLDER-3. After accounting for observation geometry, a lower dominant land-cover fraction was associated with lower surface polarized reflectance, with a greater decrease for POLDER-3 than for DPC-2, indicating a stronger association between surface polarized reflectance and subpixel composition at coarser spatial resolution. The direction of this relationship varied among land-cover classes, showing that subpixel mixing does not produce a uniform monotonic shift but also depends on the polarization contrast between dominant and secondary surface components. By developing a unified atmospheric-correction framework, this study extends BPDF database construction from single-band and single-sensor processing to multiband and multisensor applications. The resulting framework provides a foundation for updating historical BPDF records, developing long-term multisensor records of land-surface polarized reflectance, and improving the surface boundary conditions used in atmospheric polarimetric retrievals.