The freeze-thaw (FT) status of soil regulates ecological and hydrologic processes and is, therefore, a vital component of land surface models. This study utilizes a hidden Markov model (HMM) to retrieve surface FT status from L-band [Soil Moisture Active Passive (SMAP)] and Ka-band [Advanced Microwave Scanning Radiometer (AMSR)] satellite microwave brightness temperatures in cold-constrained lands north of 45 °N. The HMM, parameterized on a per-gridcell basis such that there are two possible states and emissions probabilities are assumed to be a two-component Gaussian Mixture, produces the posterior probability (a continuous variable between 0 and 1) that the surface is frozen given the radiometer input. HMM classification accuracy, averaged over five core validation networks, is acceptable (Ap > 80%) when judged against in situ air and soil temperature measurements from individual validation sites and is comparable to that of current FT products that produce a discrete state. Patterns in performance across variable land class, open water fraction, and elevation are assessed from 91 sparse network weather stations within 87 gridcells in the domain. The resulting satellite data record provides a continuous variable estimate of the daily probability of frozen conditions over northern land areas experiencing widespread thawing of permafrost and a shrinking frozen season due to global warming.
|IEEE Transactions on Geoscience and Remote Sensing
|Published - 2022
- Hidden Markov model (HMM)
- surface freeze-thaw (FT)