Frost heave — the upward displacement of soil when ice lenses form within it — has been studied in laboratories since Taber's experiments in 1929. The standard test configuration freezes soil from the bottom: a cold plate below, warmer conditions above, temperature gradient driving heat extraction downward. This configuration is convenient. The sample sits on the cold surface. Instrumentation goes on top. The setup is mechanically stable.
Nature freezes soil from the top.
Niggemann, Ziegler, and Fuentes, publishing in Acta Geotechnica in 2024, ran 62 frost heave experiments comparing top-down and bottom-up freezing under otherwise identical conditions — same soils, same temperature gradients, same durations up to 10 days. The result: bottom-up freezing produced systematically larger heave than top-down freezing. The direction of freezing changed the answer.
The mechanism is gravity. When soil freezes from below, water migrating toward the freezing front moves downward — assisted by gravity. The hydraulic supply to the growing ice lens is enhanced. When soil freezes from the top, water must move upward against gravity to reach the front. The supply is restricted. Additionally, bottom-up freezing induces vertical cracking from suction stresses that creates preferential flow paths, further amplifying water transport. The asymmetry is structural: the freezing process interacts with gravity differently depending on its orientation, and the interaction changes the physical outcome.
The implication is uncomfortable. Classical frost heave theory — from Taber through the modern segregation potential models — treats the temperature gradient as the controlling variable and assumes direction is irrelevant. Engineering design standards built on laboratory data have internalized bottom-up measurements as representative of field conditions. But natural ground freezing is always top-down. The laboratory configuration introduced a systematic overestimate that has been embedded in geotechnical practice for decades.
The error is not in the measurement. The measurements were precise. The error is in the assumption that the test configuration was equivalent to the field condition. The soil didn't know it was being tested upside down. The engineers didn't know they needed to ask.