Search for a command to run...
Abstract. The near-infrared (NIR) albedo of snow is controlled by optical snow grain size (ropt). Therefore, characterizing the spatial and temporal variability in ropt at the snow surface is critical for understanding melt timing and magnitude for water availability and Earth's energy budget toward future climates. While numerous studies have demonstrated estimates of ropt by means of optical instruments that span scales from in situ to satellites, they leverage different retrieval techniques, radiative transfer models, and modeled snow grain shapes. Variation in these factors causes tremendous uncertainty in ropt retrievals, yet a thorough evaluation has yet to be conducted. To address this knowledge gap we conducted a laboratory bidirectional reflectance study, using NIR hyperspectral imaging (NIR-HSI) to retrieve grain size metrics for a wide variety of snow microstructures and evaluate them against micro-CT benchmarks. Toward enhanced ropt retrieval accuracy, we sought to determine (1) the optimal modeled snow grain shape; (2) the best-performing radiative transfer model; and (3) variability associated with retrieval techniques, spanning broadband, narrowband, multispectral, and hyperspectral approaches. Our results for optimizing grain shape parameters align with existing studies for the TARTES model, and we offer first recommendations for the SNICAR model. The retrieval technique also displayed considerable variation, with the hyperspectral residual method performing best. Multispectral and single-band techniques were comparable to their hyperspectral counterparts at times, but this was attributed to the idealized laboratory conditions and high instrument signal-to-noise ratio. Following shape optimization, the SNICAR and TARTES models produced the best results (median absolute error of 15.6 %–17.4 %, depending on technique), outperforming the AART model and the Random Mixture model. Toward a more direct comparison with albedo estimate error, we also evaluated the square root of ropt retrievals; median absolute error values ranged from 7.9 %–26.2 %, depending on model and technique, with most pairings resulting in values <15 %. Our results demonstrate that the accuracy of ropt retrievals is highly sensitive to the choice of retrieval technique, radiative transfer model, and grain shape parameters. To minimize error, each of these factors should be carefully selected in the context of the specific measurement. As NIR-HSI instruments and other NIR detectors become increasingly affordable and their resolution improves, the findings presented here provide guidance for improved ropt and snow albedo mapping across ground-based, aerial, and satellite platforms.