Estimating the landscape and soil freeze-thaw (FT) dynamics in the Northern Hemisphere (NH) is crucial for understanding permafrost response to global warming and changes in regional and global carbon budgets. A new framework for surface FT-cycle retrievals using L-band microwave radiometry based on a deep convolutional autoencoder neural network is presented. This framework defines the landscape FT-cycle retrieval as a time-series anomaly detection problem, considering the frozen states as normal and the thawed states as anomalies. The autoencoder retrieves the FT-cycle probabilistically through supervised reconstruction of the brightness temperature (TB) time series using a contrastive loss function that minimizes (maximizes) the reconstruction error for the peak winter (summer). Using the data provided by the Soil Moisture Active Passive (SMAP) satellite, it is demonstrated that the framework learns to isolate the landscape FT states over different land surface types with varying complexities related to the radiometric characteristics of snow cover, lake-ice phenology, and vegetation canopy. The consistency of the retrievals is assessed over Alaska using in situ observations, demonstrating an 11% improvement in accuracy and reduced uncertainties compared to traditional methods that rely on thresholding the normalized polarization ratio (NPR).

The soil moisture active passive (SMAP) satellite mission distributes a product of CO2 flux estimates (SPL4CMDL) derived from a terrestrial carbon flux model, in which SMAP brightness temperatures are assimilated to update soil moisture (SM) and constrain the carbon cyclemodeling. While the SPL4CMDL product has demonstrated promising performance across the continental USA and Australia, a detailed assessment over the arctic and subarctic zones (ASZ) is still missing. In this study, SPL4CMDL net ecosystem exchange (NEE), gross primary production (GPP), and ecosystem respiration (R-E) are evaluated against measurements from 37 eddy covariance towers deployed over the ASZ, spanning from 2015 to 2022. The assessment indicates that the NEE unbiased root-mean-square error falls within the targeted accuracy of 1.6 gC.m(-2).d(-1), as defined for the SPL4CMDL product. However, modeled GPP and R-E are overestimated at the beginning of the growing season over evergreen needleleaf forests and shrublands, while being underestimated over grasslands. Discrepancies are also found in the annual net CO2 budgets. SM appears to have a minimal influence on the GPP and R-E modeling, suggesting that ASZ vegetation is rarely subjected to hydric stress, which contradicts some recent studies. These results highlight the need for further carbon cycle process understanding and model refinements to improve the SPL4CMDL CO2 flux estimatesover the ASZ.