Sorption hypotheses and models are required for the prediction of 137 Cs migration in soils contaminated after nuclear reactor accidents and nuclear weapons tests. In assessment models, the K d (distribution coefficient) hypothesis for sorption, which assumes that sorption is instantaneous, linear and reversible, has often been coupled with the convection-diffusion equation (CDE) to model 137 Cs migration. However, it fails to describe 137 Cs migration velocities which often decrease with time. Alternative equilibrium-kinetic (EK) hypotheses of 137 Cs sorption/desorption have been suggested by laboratory experiments but have not been fully validated in field conditions. This work addressed the influence and magnitude of non-equilibrium 137 Cs sorption in field conditions by reinterpreting, with an inverse approach, series of 137 Cs profiles measured in mineral soils of forest plots located in Fukushima Prefecture (2013-2018). Our results show that the inclusion of non-equilibrium sorption significantly improves, compared to the equilibrium hypothesis, the realism of simulated 137 Cs profiles. Fitted sorption parameters suggest a fast sorption kinetic (half-time of 1-7 ) and a pseudo-irreversible desorption rate (half-time of 3.2 × 10 0-3.4 × 10 6 years), whereas equilibrium sorption (4.0 × 10 − 3 L kg − 1 on average) only affects a negligible portion of 137 Cs inventory. By June 2011, such EK parameters fitted on our plots realistically reproduced profiles measured in the same forest study site (Takahashi et al., 2015). Predictive modeling of 137 Cs profiles in soil suggests a strong persistence of the surface 137 Cs contamination by 2030, with exponential profiles consistent with those reported after the Chernobyl accident. This study demonstrates that hypotheses and parameters of 137 Cs sorption can be partially inferred from in situ measurements. However, further experiments in controlled conditions are required to better estimate the sorption parameters and to identify the processes behind non-equilibrium sorption.