Global warming accelerates permafrost degradation, compromising the reliability of critical infrastructure relied upon by over five million people daily. Additionally, permafrost thaw releases substantial methane emissions due to the thawing of swamps, further amplifying global warming and climate change and thus posing a significant threat to more than eight billion people worldwide. To mitigate this growing risk, policymakers and stakeholders need accurate predictions of permafrost thaw progression. Comprehensive physics-based permafrost models often require complex, location-specific fine-tuning, making them impractical for widespread use. Although simpler models with fewer input parameters offer convenience, they generally lack accuracy. Purely data-driven models also face limitations due to the spatial and temporal sparsity of observational data. This work develops a physics-informed machine learning framework to predict permafrost thaw rates. By integrating a physics-based model into machine learning, the framework significantly enhances the feature set, enabling models to train on higher-quality data. This approach improves permafrost thaw rate predictions, supporting more reliable decision-making for construction and infrastructure maintenance in permafrost-vulnerable regions, with a forecast horizon spanning several decades.

Soil moisture detection research, which influences crop growth, land use, and soil erosion, is receiving significant attention. This study proposes a nondestructive, integrated ultrawideband (UWB)-based framework for soil moisture measurement and prediction. The method utilizes a UWB-loaded unmanned aerial vehicle (UAV) to gather radar echo data, circumventing soil damage issues inherent in current research and equipment. We first employ time-frequency analysis methods to convert the echo signals into 2-D spectrograms, constructing datasets labeled with soil moisture. Then, a trained neural network is used to predict the soil moisture at single point. Additionally, a novel interpolation method is proposed to enhance prediction accuracy (ACC) for the ridge-furrow structure of farmland. The experimental results demonstrate that the proposed algorithm achieves a soil moisture measurement ACC of 98% in both vegetated and nonvegetated conditions, indicating strong robustness. In terms of moisture distribution prediction, the mean squared error (mse) of soil moisture spatial distribution prediction is reduced by 42% compared to traditional methods. Therefore, this system provides technical support for efficient, large-scale, and nondestructive soil information collection.

The freezing front depth (z(ff)) of annual freeze-thaw cycles is critical for monitoring the dynamics of the cryosphere under climate change because z(ff) is a sensitive indicator of the heat balance over the atmosphere-cryosphere interface. Meanwhile, although it is very promising for acquiring global soil moisture distribution, the L-band microwave remote sensing products over seasonally frozen grounds and permafrost is much less than in wet soil. This study develops an algorithm, i.e., the brightness temperature inferred freezing front (BT-FF) model, for retrieving the interannual z(ff) with the diurnal amplitude variation of L-band brightness temperature (?T-B) during the freezing period. The new algorithm assumes first, the daily-scale solar radiation heating/cooling effect causes the daily surface thawing depth (z(tf)) variation, which leads further to ?T-B; second, ?T-B can be captured by an L-band radiometer; third, z(tf) and z(ff) are negatively linear correlated and their relation can be quantified using the Stefan equation. In this study, the modeled soil temperature profiles from the land surface model (STEMMUS-FT, i.e., simultaneous transfer of energy, mass, and momentum in unsaturated soil with freeze and thaw) and T-B observations from a tower-based L-band radiometer (ELBARA-III) at Maqu are used to validate the BT-FF model. It shows that, first, ?T-B can be precisely estimated from z(tf) during the daytime; second, the decreasing of z(tf) is linearly related to the increase of z(ff) with the Stefan equation; third, the accuracy of retrieved z(ff) is about 5-25 cm; fourth, the proposed model is applicable during the freezing period. The study is expected to extend the application of L-band T-B data in cryosphere/meteorology and construct global freezing depth dataset in the future.