Long-term records of daily mean photosynthetically available radiation (PAR) are essential for quantifying variability and trends in marine primary production, upper-ocean heating, and ocean biogeochemistry. However, existing PAR products derived from MODIS, VIIRS, OLCI, SGLI, and other satellite missions differ in calibration, retrieval algorithms, temporal sampling, spatial coverage, resolution, and uncertainty. Their periods of operation also overlap only partially, potentially introducing discontinuities into multidecadal records. We present a multiresolution Bayesian framework for harmonizing these products into a continuous daily mean PAR record on a common 4 km grid, a resolution sufficiently fine to represent regional and mesoscale variability while remaining compatible with long-term global processing. The method first standardizes PAR definitions, units, daily averaging conventions, and spatial support. Sensor-specific additive and multiplicative biases, including possible temporal drift and processing-version changes, are characterized using mission-overlap periods and daily mean PAR measurements from a limited number of long-term in situ stations. Partial pooling across stations and broad atmospheric regimes prevents overfitting while allowing physically meaningful variations in product performance. Bias-corrected satellite products are then combined simultaneously rather than through sequential pairwise merging, thereby avoiding repeated use of common information. Fine-resolution satellite products directly determine PAR at 4 km, whereas coarser products, including EPIC and reanalysis-driven estimates, constrain spatial averages through explicit footprint operators. Reanalysis provides a complete large-scale prior during satellite gaps but is not interpreted as independent evidence at 4 km. The fusion uses a covariance matrix that incorporates reported random and bias uncertainties, footprint-representativeness errors, and correlations arising from shared algorithms and ancillary data. The framework produces daily PAR together with random, systematic, and total uncertainties, source-contribution metrics, effective spatial resolution, and provenance flags. Performance and temporal stability are evaluated using withheld in situ records, sensor-withholding experiments, and tests that reproduce historical patterns of limited satellite availability. The resulting approach is designed to preserve genuine 4 km satellite variability while providing a consistent, uncertainty-characterized multidecadal PAR record across changing satellite constellations.

暂无