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).

This paper presents a convolutional autoencoder deep learning framework for probabilistic characterization of the ground freeze-thaw (FT) dynamics in the Northern Hemisphere to enhance our understanding of permafrost response to global warming and shifts in the high-latitude carbon cycle, using Soil Moisture Active Passive (SMAP) satellite brightness temperatures (TB) observations. The autoencoder recasts the FT-cycle retrieval as an anomaly detection problem in which the peak winter (summer) represents the normal (anomaly) segments of the TB time series. The results demonstrate that the new framework outperforms the widely used fixed-thresholding of the Normalized Polarization Ratio (NPR) by learning the land surface structural and radiometric complexities that might arise in TB times series due to snow cover and vegetation. Validation against ground-based measurements over Alaska shows that the accuracy of the FT-cycle retrievals can be improved by 12%, primarily due to a marked reduction in false detection of short snowmelt episodes as ground thawing by the NPR thresholding approach.

To account for the impacts of snow on surface upwelling emission at the L-band microwave, this study employs a soil-snow-vegetation emission model to examine the retrieval errors associated with the simultaneous estimation of vegetation optical depth (VOD) and ground permittivity. The findings demonstrate that neglecting snow cover in retrievals can lead to an overestimation of VOD by approximately 30% and substantial error in ground permittivity, largely depending on the ground freeze-thaw status, snow density, and VOD. We utilize the Soil Moisture Active Passive (SMAP) satellite observations to retrieve VOD and compare it to the vegetation proxies such as AGB and tree height on a global scale. The results show consistent spatial patterns with the land cover types and dependencies with above-ground biomass (AGB) values and tree heights. The preliminary results offer promising possibilities to obtain global estimates of VOD and soil permittivity over snow-covered areas, where current SMAP observations are underutilized.

This article presents a new algorithmic framework for multilayer inversion of the dense media radiative transfer (DMRT) equations of snowpack emission, with particular emphasis on the role of high-frequency microwave channels above 60 GHz. The approach relies on dictionary matching and locally constrained least squares. The results demonstrate that the algorithm can invert the DMRT model and retrieve depth, density, and grain size of a single-layer snowpack when dependencies of density and grain size on depth are properly accounted for. However, as the number of layers increases, the sensitivity of the inversion to observation noise grows markedly. Using observations, over the Great Plains in the United States, from the microwave imager onboard the global precipitation measurement (GPM, 10–166 GHz) core satellite, the initial results demonstrate that under a clear-sky condition and no vegetation canopy, the algorithm is capable to retrieve the snow depth and water equivalent of seasonal snow with a mean absolute error (MAE) of less than 0.15 m—when compared to the high-resolution analysis data from the SNOw Data Assimilation System (SNODAS).

Soil and its water content can remain unfrozen below an insulative snow cover and modulate snowmelt infiltration and runoff. In this paper, an emission model is proposed to account for L-band microwave emission of wet soils below a dry snowpack covered with an emerging moderately dense vegetation canopy. The model links the well-known tau-omega emission model with the snowpack dense media radiative transfer theory as well as a multi-layer composite reflection model to account for the impacts of a snow layer on the upwelling soil and the downwelling vegetation emission, respectively. It is demonstrated that even though a dry snow is a low-loss medium at L-band, omission of its presence leads to underestimation of soil moisture (SM), especially when soil (snow) becomes wetter (denser). Constrained inversion of the proposed emission model, using brightness temperatures from the Soil Moisture Active and Passive (SMAP) satellite, shows that the retrievals of SM and vegetation optical depth (VOD) are achievable with unbiased root mean squared errors of 0.060 m3.m-3and 0.124 [-], when compared within situdata from the International Soil Moisture Network (ISMN) as well as VOD-derived values from the Normalized Difference Vegetation Index (NDVI) obtained from the Moderate Resolution Imaging Spectroradiometer (MODIS) observations.

This article investigates the use of deep knockoff, a modern statistical variable selection methodology, to uncover the spectral signatures of marine debris. This method uses a generative model by leveraging deep neural networks to learn the high dimensional distribution of reflectance in visible to near infrared wavelengths. To that end, a public dataset obtained from ground-based labeling of the observations by the Multispectral Instrument (MSI) on-board Sentinel-2 satellite is used. Through controlling the false discovery rate, consistent with the known physical causalities, the results indicate that the near infrared (band 8, 833 nm) and red (band 4, 665 nm) are the most important bands, respectively for discrimination of marine (plastic) debris from the background water. In the presence of denseSargassummacroalgae, the deep knockoff isolates the green (band 3, 560 nm) and the narrow NIR (band 8a, 864.7 nm) as another important band.

The uncertainty of passive microwave retrievals of snowfall is notoriously high where the high-frequency surface emissivity is significantly reduced and varies markedly in response to changes of snowpack physical properties. Using the dense media radiative transfer theory, this article studies the potential effects of terrestrial snow-cover depth, density, and grain size on high-frequency channels 89 and 166 GHz of the radiometer onboard the Global Precipitation Measurement (GPM) core satellite, which are commonly used to capture the snowfall scattering signals. Integrating the inference across all feasible grain sizes, ranges of snowpack density and depth are identified over which the snowfall scattering signatures can be time varying and potentially obscured. Using 10 years of reanalysis data, the seasonal vulnerability of snowfall retrievals to changes of snowpack emissivity in the Northern Hemisphere is mapped in a probabilistic sense and connections are made with uncertainties of the GPM passive microwave snowfall retrievals. It is found that among different snow classes, relatively light Arctic tundra snow in fall, with a density below 260 kg m-3, and shallow prairie snow during the winter, with a depth of less than 40 cm, can reduce the surface emissivity and obscure the snowfall passive microwave signatures. It is demonstrated that, during the winter, the highly vulnerable areas are over Kazakhstan, and Mongolia with taiga and prairie snow. In the fall, these areas are largely over tundra and taiga snow in north of Russia and the Arctic Archipelagos as well as prairies in Canada and the Great Plains in the United States.

This article studies changes in microwave signals of oceanic snowfall in response to the formation of snow-covered sea ice using active and passive coincident data from the radar and radiometer onboard the CloudSat and the global precipitation measurement satellites. Using reanalysis data of liquid and ice water path as well as satellite retrievals of sea ice snow-cover depth, spectral regions are determined over which the snowfall signatures are likely to be obscured or falsely detected. Relying on ana prioridatabase populated with the active–passive coincidences, a Bayesian snowfall retrieval algorithm is presented that links a$k$-nearest neighbor matching with the inverse Gaussian estimator used in the Goddard profiling algorithm. Without relying on any ancillary data of air temperature, the results demonstrate that over open oceans (sea ice), we can passively retrieve the CloudSat active snowfalls with a true positive rate of 92 (85%) and the root mean squared error of 0.24 (0.15) mmh−1.

Improved microwave retrieval of land and atmospheric state variables requires proper weighting of the information content of radiometric channels through their error covariance matrix. Inspired by recent advances in metric learning techniques, a new framework is proposed for a formal approximation of the channel error covariance. The idea is tested for the detection of precipitation and its phase over oceans, using coincidences of passive/active data from the Global Precipitation Measurement (GPM) and CloudSat satellites. The initial results demonstrate that the presented approach cannot only capture the known laws of radiative transfer equations, but also the surrogate signatures that can arise due to the co-occurrence of precipitation and other radiometrically active land-atmospheric state variables. In particular, the results demonstrate high precision (low error) for the low-frequency channels of 10-37 GHz in the detection of both rain and snowfall over oceans. Using the optimal estimate of the channel error covariance through the multi-frequency k-nearest neighbor (kNN) classification approach, without any ancillary data, it is demonstrated that the probability of passive microwave detection of snowfall (0.97) can be higher than that of the rainfall (0.88), when drizzle and light snowfall are the dominant form of precipitation. This improvement is hypothesized to be largely related to the information content of the low-frequency channels of 10-37 GHz that can capture the co-occurrence of snowfall with an increased cloud liquid water content, sea ice, and wind-induced changes of surface emissivity.

Understanding and reducing the uncertainties in the inversion of the first-order radiative transfer models at the L-band are important for the improved spaceborne retrievals of soil moisture (SM) and vegetation optical depth (VOD) over dense canopy. This article quantifies and compares the sensitivity of dual-channel inversion of the two-stream (2S) and τ-ω models and proposes a new inversion approach for simultaneous retrievals of SM, VOD, and vegetation-scattering albedo (ω) from a single satellite overpass. In particular, the inversion algorithm incorporates the information of the nearby spatial observations, assuming that the values of VOD and ω remain locally invariant, and constrains its solutions to high-resolution a priori physical/climatological knowledge of the retrieval variables. The results demonstrate that the uncertainty in the inversion of 2S model is slightly higher than the τ-ω model under noisy observations and remains homoscedastic for SM and ω, while grows heteroscedastically for higher VOD values due to the shape of the cost function. The results are validated using the SMAP data, the dense Mesonet SM network, the in situ measurements from the International SM Network (ISMN), and the derived VOD from the Moderate Resolution Imaging Spectroradiometer (MODIS)-normalized difference vegetation index (NDVI) over the state of Oklahoma in the United States. It is shown that the new approach can recover simultaneously high-resolution features of SM, VOD, and ω only from a single Soil Moisture Active Passive (SMAP) overpass, where the unbiased root-mean-squared error (ubRMSE) of SM and VOD is reduced by 30% and 70%, respectively, when compared with an unconstrained time-windowed inversion approach.

This paper introduces a new Bayesian approach to the inverse problem of passive microwave rainfall retrieval. The proposed methodology [called the shrunken locally linear embedding algorithm for retrieval of precipitation (ShARP)] relies on a regularization technique and makes use of two joint dictionaries of coincident rainfall profiles and their corresponding upwelling spectral radiative fluxes. A sequential detection-estimation strategy is adopted, which basically assumes that similar rainfall intensity values and their spectral radiances live close to some sufficiently smooth manifolds with analogous local geometry. The detection step employs a nearest neighbor classification rule, whereas the estimation scheme is equipped with a constrained shrinkage estimator to ensure the stability of retrieval and some physical consistency. The algorithm is examined using coincident observations of the active precipitation radar and the passive microwave imager onboard the TRMM satellite. We present promising results of instantaneous rainfall retrieval for some tropical storms and mesoscale convective systems over ocean, land, and coastal zones. We provide evidence that the algorithm is capable of properly capturing different storm morphologies including high-intensity rain cells and trailing light rainfall, particularly over land and coastal areas. The algorithm is also validated at an annual scale for calendar year 2013 versus the standard (version 7) radar (2A25) and radiometer (2A12) rainfall products of the TRMM satellite.