Snow-climate interaction in NCAR CCM3

Although detailed process-oriented snow models exist, general circulation models (GCMs) use relatively simple schemes for computational considerations. Therefore it is critical to assess their simulations for two related reasons. First, GCM results have been used to study climate change and water resource issues. Second, GCMs are known to be highly sensitive to snow processes. Two widely-used land-surface models are selected in this study, namely, the Biosphere-Atmosphere Transfer Scheme (BATS) and the NCAR Land Surface Model (LSM), each including a snow submodel. Both BATS and LSM have been coupled to the NCAR CCM3, and long-term integrations (more than 10 years) are available for analyses. For both North America and Eurasia, LSM-CCM3 produces less snow mass than BATS-CCM3 during the snow season, which generally coincides with more absorbed solar radiation, higher 2-m air temperature, less snowfall, and smaller latent heat flux in LSM-CCM3. In terms of continental averages, the snow mass is more accurately simulated in BATS than in LSM during the accumulation period. Because LSM underestimates peak snow mass, its ablation rate appears to agree with observations, while the BATS melting rate is slower than observed. Both models predict snow depth greater than the estimates from the Nimbus-7 Scanning Multichannel Microwave Radiometer (SMMR). Some of the differences in the snow simulations between BATS and LSM can be attributed to differences in precipitation, surface air temperature, and the surface energy balance terms, resulting from the fully-coupled land surface-atmosphere interactions. To filter out this feedback effect, an intercomparison of both land models is made in off-line mode (i.e. uncoupled to the CCM3) using the long-term snow cover and meteorological data from a grassland catchment at the Valdai water-balance research site in Russia. I will first discuss results from the models in their original structure, and then show the impacts from using different methods of layering, albedo and areal snow cover extent. The off-line runs confirm that LSM predicts less snow mass than BATS, mainly due to two reasons. First, the iterative surface energy balance in LSM tends to predict higher surface temperature and greater snow melt. Second, the albedo parameterization in LSM generally leads to lower albedo values than in BATS. Guided by our recent survey of the existing snow models (http://www.atmo.arizona.edu/~zly/snow.html), a versatile multi-layer snow model is being developed to address the level of complexity required in GCM snow modeling.