Sensitivity of active-layer freezing process to snow cover in Arctic Alaska

The contribution of cold-season soil respiration to the Arctic–boreal carbon cycle and its potential feedback to the global climate remain poorly quantified, partly due to a poor understanding of changes in the soil thermal regime and liquid water content during the soil-freezing process. Here, we characterized the processes controlling active-layer freezing in Arctic Alaska using an integrated approach combining in situ soil measurements, local-scale (∼50 m) longwave radar retrievals from NASA airborne P-band polarimetric SAR (PolSAR) and a remote-sensing-driven permafrost model. To better capture landscape variability in snow cover and its influence on the soil thermal regime, we downscaled global coarse-resolution (∼0.5∘) MERRA-2 reanalysis snow depth data using finer-scale (500 m) MODIS snow cover extent (SCE) observations. The downscaled 1 km snow depth data were used as key inputs to the permafrost model, capturing finer-scale variability associated with local topography and with favorable accuracy relative to the SNOTEL site measurements in Arctic Alaska (mean RMSE=0.16 m, bias=-0.01 m). In situ tundra soil dielectric constant (ε) profile measurements were used for model parameterization of the soil organic layer and unfrozen-water content curve. The resulting model-simulated mean zero-curtain period was generally consistent with in situ observations spanning a 2∘ latitudinal transect along the Alaska North Slope (R: 0.6±0.2; RMSE: 19±6 days), with an estimated mean zero-curtain period ranging from 61±11 to 73±15 days at 0.25 to 0.45 m depths. Along the same transect, both the observed and model-simulated zero-curtain periods were positively correlated (R>0.55, p) with a MODIS-derived snow cover fraction (SCF) from September to October. We also examined the airborne P-band radar-retrieved ε profile along this transect in 2014 and 2015, which is sensitive to near-surface soil liquid water content and freeze–thaw status. The ε difference in radar retrievals for the surface (∼<0.1 m) soil between late August and early October was negatively correlated with SCF in September (R=-0.77, p); areas with lower SCF generally showed larger ε reductions, indicating earlier surface soil freezing. On regional scales, the simulated zero curtain in the upper ( m) soils showed large variability and was closely associated with variations in early cold-season snow cover. Areas with earlier snow onset generally showed a longer zero-curtain period; however, the soil freeze onset and zero-curtain period in deeper (>0.5 m) soils were more closely linked to maximum thaw depth. Our findings indicate that a deepening active layer associated with climate warming will lead to persistent unfrozen conditions in deeper soils, promoting greater cold-season soil carbon loss.