Investigating the sensitivity of soil heterotrophic respiration to recent snow cover changes in Alaska using a satellite-based permafrost carbon model

The contribution of soil heterotrophic respiration to the boreal–Arctic carbon (CO2) cycle and its potential feedback to climate change remains poorly quantified. We developed a remote-sensing-driven permafrost carbon model at intermediate scale (∼ 1 km) to investigate how environmental factors affect the magnitude and seasonality of soil heterotrophic respiration in Alaska. The permafrost carbon model simulates snow and soil thermal dynamics and accounts for vertical soil carbon transport and decomposition at depths up to 3 m below the surface. Model outputs include soil temperature profiles and carbon fluxes at 1 km resolution spanning the recent satellite era (2001–2017) across Alaska. Comparisons with eddy covariance tower measurements show that the model captures the seasonality of carbon fluxes, with favorable accuracy in simulating net ecosystem CO2 exchange (NEE) for both tundra (R > 0.8, root mean square error (RMSE – 0.34 g C m−2 d−1), and boreal forest (R > 0.73; RMSE – 0.51 g C m−2 d−1). Benchmark assessments using two regional in situ data sets indicate that the model captures the complex influence of snow insulation on soil temperature and the temperature sensitivity of coldseason soil heterotrophic respiration. Across Alaska, we find that seasonal snow cover imposes strong controls on the contribution from different soil depths to total soil heterotrophic respiration. Earlier snowmelt in spring promotes deeper soil warming and enhances the contribution of deeper soils to total soil heterotrophic respiration during the later growing season, thereby reducing net ecosystem carbon uptake. Early cold-season soil heterotrophic respiration is closely linked to the number of snow-free days after the land surface freezes (R =−0.48, p < 0.1), i.e., the delay in snow onset relative to surface freeze onset. Recent trends toward earlier autumn snow onset in northern Alaska promote a longer zero-curtain period and enhanced cold-season respiration. In contrast, southwestern Alaska shows a strong reduction in the number of snow-free days after land surface freeze onset, leading to earlier soil freezing and a large reduction in cold-season soil heterotrophic respiration. Our results also show nonnegligible influences of subgrid variability in surface conditions on the model-simulated CO2 seasonal cycle, especially during the early cold season at 10 km scale. Our results demonstrate the critical role of snow cover affecting the seasonality of Published by Copernicus Publications on behalf of the European Geosciences Union. 5862 Y. Yi et al.: Investigating soil heterotrophic respiration sensitivity to snow cover changes in Alaska soil temperature and respiration and highlight the challenges of incorporating these complex processes into future projections of the boreal–Arctic carbon cycle.

where ( , ) is the temperature (°C) at a specific soil depth ( ) and time step ( ), is the latent heat of fusion of water (J m -3 ),  is the total soil water content (m 3 m -3 ), and is the unfrozen liquid water fraction (%). and λ are the volumetric heat capacity (J m -3 K -1 ) and thermal conductivity (W m -1 K -1 ) of soil, respectively, varying with depth, soil moisture and F/T state. The upper boundary condition is set as the surface temperature at the snow/ground surface ( ), while a heat flux characterizing the geothermal gradient is applied at the lower boundary ( = 60 ). The soil thermal properties, including the soil heat capacity and thermal conductivity, are a function of the thermal properties of mineral and organic soil solid and liquid water, and ice components, weighted by their volumetric fraction.
The thermal conductivity λ is estimated as a normalized thermal conductivity of the dry ( ) and saturated ( ) soil thermal conductivity, weighted by soil saturation: where the Kersten number ( ) is a function of the soil saturation degree, using a logarithm form for unfrozen soils and linear form for frozen soils (Farouki, 1981;Lawrence and Slate, 2008). is estimated from the soil bulk density; is estimated as a geometric mean of the thermal conductivity of different soil components (Farouki 1981), including mineral and organic soil solid, liquid water and ice, which can vary several-fold from pure organic soil (~0.5 W m -1 K -1 ) to mineral soils (1.5 ~ 3 W m -1 K -1 ).
Soil water usually freezes at a sub-zero temperature depending on solute concentration and other factors, and the model uses the following empirical function to estimate the unfrozen liquid water fraction ( ): The constant * T represents the freezing point depression, with values generally above -1°C (Woo, 2012). b is a dimensionless parameter determined by fitting the unfrozen water curve, which can vary significantly depending on soil type (Schaefer and Jafarov, 2016).   Jackson et al. (1996): = 1 − , where is the cumulative root fraction from soil surface to depth z (cm), and is the extinction coefficient parameter.

Fig. S1
Comparison between effective snow depth (a) derived from in-situ observations at Snotel sites and downscaled MERRA2 data, and observed and model simulated monthly soil temperature at 20 cm depth (b). Note that the sites compared for snow depth and soil temperature may be inconsistent due to inconsistency in the snow depth and soil temperature measurements at the Snotel sites. Generally, there are more snow depth measurements than soil temperature measurements.

Fig. S2
Model simulated temperature sensitivity of ecosystem respiration at the US-Atq tundra site. "Reco1 obs" and "Reco2 obs" represent ecosystem respiration estimates derived using tower-based NEE measurements and different partitioning methods provided by the tower PI.         (2001). The standard deviation of GPP and Rh flux across the study area was approximately 50% of the regional mean, and was not shown. The timing of snow disappearance (i.e. snow offset) in the spring was defined as the center of the 8-day composite period being snow free, and with mean snow depth less than 5 cm depth within a 24-day moving window.