Modeling Snow and Ice Microwave Emissions in the Arctic for a Multi‐Parameter Retrieval of Surface and Atmospheric Variables From Microwave Radiometer Satellite Data

Abstract

Monitoring surface and atmospheric parameters—like water vapor—is challenging in the Arctic, despite the daily Arctic‐wide coverage of spaceborne microwave radiometer data. This is mainly due to the difficulties in characterizing the sea ice surface emission: sea ice and snow microwave emission is high and highly variable. There are very few data sets combining relevant in situ measurements with co‐located remote sensing data, which further complicates the development of accurate retrieval algorithms. Here, we present a multi‐parameter retrieval based on the inversion of a forward model for both, atmosphere and surface, for non‐melting conditions. The model consists of a layered microwave emission model of snow and ice. Since snow scattering and emission effects, as well as temperature gradients, are taken into account, a high variability in brightness temperatures can be simulated. For ocean regions and the atmosphere existing parameterized forward models are used. By using optimal estimation, the forward model can be inverted allowing for the simultaneous and consistent retrieval of nine variables: integrated water vapor, liquid water path, sea ice concentration, multi‐year ice fraction, snow depth, snow‐ice interface temperature and snow‐air interface temperature as well as sea‐surface temperature and wind speed (over open ocean). In addition, the method provides retrieval uncertainty estimates for each retrieved parameter. To evaluate the forward model as well as the retrieval, we use the extensive data sets acquired during the year‐long Arctic expedition Multidisciplinary drifting Observatory for the Study of Arctic Climate (2019–2020) as a reference.

Simultaneous Inference of Sea Ice State and Surface Emissivity Model Using Machine Learning and Data Assimilation

“…One aim is to extend the modeling to higher microwave and sub-mm frequencies using additional sensors. Another is to move beyond the initially crude description of the surface by an effective emissivity and a skin temperature, and instead to use a model which describes the known physics of radiative transfer within the snow and sea ice (e.g., Kang et al, 2023;Rückert et al, 2023). In this approach, empirical state variables would still be required to describe the microphysical properties of the sea ice and snow, but the empirical model would have the more targeted responsibility of generating the optical properties that are required as input to such a model.…”

“…This would further complicate the model and potentially make it under-constrained. The ideal solution would be to use a physical model representing radiative transfer within multiple layers of sea ice and snow, which avoids needing to use either effective emissivity or effective temperature (e.g., Picard et al, 2018;Rückert et al, 2023), though possibly this brings even more issues in needing to constrain the required vertical profiles of temperature and snow and ice microphysical properties. Further developments in this direction are intended in future work.…”

“…For physical retrieval techniques including data assimilation in weather forecasting, the interaction of the sea ice and its snow cover with electromagnetic radiation needs to be represented with an observation model. Typically this is done in terms of surface emissivity (e.g., Tonboe et al, 2013) or by using a multi-layer model for radiative transfer in the sea ice and snow (e.g., Kang et al, 2023;Rückert et al, 2023;Scarlat et al, 2020). However, physical modeling of snow and ice radiative transfer at microwave frequencies is difficult, in particular due to the importance of centimeter to micron scale details of the snow and ice, including air inclusions in ice or grain shapes and their layout in snow, which strongly affect the optical properties of the snow and ice.…”

“…Physical models of sea ice are typically better at fitting observations at low microwave frequencies than above due to increasing sensitivity to the poorly known microstructural details of the snow cover at higher frequencies (e.g., 6.9 vs. 89 GHz results in Rückert et al (2023)). Similar issues are seen with physically modeling snow radiative transfer over land surfaces (e.g., Hirahara et al, 2020).…”

“…To deal with this is necessary to make empirical adjustments to fit the observations at each location -for example, modifying the snow and sea ice optical properties (Kang et al, 2023) or similarly the surface emissivity (Tonboe et al, 2013). Alternatively, very detailed assumptions of snow and sea ice microphysical structure need to be made a-priori, though so far this approach has been validated only for a limited set of conditions (e.g., Arctic winter, Rückert et al, 2023). For global assimilation of snow and ice surfaces, observation models will need to be reliable across all seasons and hemispheres, and will need to handle many different types of ice and snow, including wet and refrozen snow in the warmer seasons, so it is likely to be challenging to scale up physical modeling efforts to provide all-season, global capabilities.…”