Snow Parameter Retrieval Algorithm Enhanced by Optimal Estimation (SPR-OE) and Its Validation Over Greenland

We present SPR-OE, a snow parameter retrieval (SPR) algorithm enhanced by optimal estimation (OE) for optical remote sensing of snow properties. The SPR-OE algorithm combines: 1) a first guess from the machine-learning (ML)based SPR algorithm and 2) a Levenberg–Marquardt (LM) OE step. Applied to Global Change Observation Mission–Climate (GCOM-C)/Second-Generation Global Imager (SGLI) satellite observations over the Greenland Ice Sheet (GrIS) during 2018–2021, SPR-OE improves the accuracy of the retrieved snow parameters relative to the SPR algorithm, reducing the mean residual by more than 80%. At the SIGMA-A site, blue-sky albedo retrievals agree within a root mean square error (RMSE) = 0.048 and mean absolute percentage error (MAPE) = 3.3% compared to in situ measurements. Comparisons with albedo measured by the Greenland Climate Network (GC-NET) at Summit have an RMSE and MAPE of 0.035 and 3.6%. Beyond improved accuracy, SPR-OE yields pixel-level uncertainties for validation and data assimilation. Further improvements are possible using polarimetric measurements to discriminate absorption and attenuation by aerosols from absorption by snow impurities and by modeling wet snow properties in melt-affected areas.