Abstract
<title>Abstract</title> <p>Current climate models assume that soil freezes at 0°C, neglecting freezing-point depression caused by dissolved salts and matrix effects—a long-recognized but poorly quantified mechanism at large scales. Here, we integrate freeze-thaw states and soil temperature records from Soil Moisture Active Passive (SMAP) satellite observations with soil property datasets to characterize sub-zero unfrozen soil (SUS) events and quantify the freezing-point depression effect across the Northern Hemisphere. Using 137,554 SMAP pixels spanning 2015–2022, we estimate a median soil freezing point of 271.50 K for nocturnal freezing events, approximately 1.65°C below that of pure water. SUS events have a wide range of occurrence temperatures (interquartile range: 269.04–272.74 K), with 48.79% occurring below the median freezing point. Machine learning analysis reveals that these low-temperature events are primarily driven by soil salinity and the spontaneous thermodynamic character of thawing. Validation against 19 collocated FLUXNET eddy-covariance sites shows that SMAP-derived (271.65 K) and in-situ (272.01 K) median freezing points differ by only 0.36 K, with no statistically significant difference between their distributions. These findings indicate that the 0°C assumption in Earth system models systematically underestimates permafrost thaw onset and degradation risks, with critical implications for carbon-climate feedbacks from permafrost regions.</p>