Monitoring Constraints and Simulation-Based Distinguishability of Thermal-History Models for a Crushed-Rock Ventilated Embankment

Field temperature and elevation records from a G219 crushed-rock ventilated embankment show transverse differences, but the available spring observations do not cover a complete thaw-peak-to-recovery sequence. We used these records to constrain the scale and observation structure of controlled simulation experiments, rather than to estimate a field recovery mechanism. Candidate models represented current thaw exposure, recovery deficit, cumulative thaw exposure, and finite lag. Models were fitted on the first two prescribed freeze–thaw cycles and evaluated using withheld recovery responses from the third cycle. Under the observation structure corresponding to the existing records, instantaneous ground truths produced false-positive rates of 0.149 and 0.224 for historical explanations. Under the reference simulated monitoring design, the recovery-deficit model met the prespecified E3a decision criterion in all 1000 noise repetitions and reduced holdout RMSE by 0.491 cm on average relative to the best instantaneous model. A separate 25-setting simulation sensitivity grid met the complete decision rule in 13 settings, indicating dependence on thermal history and holdout timing. Other historical models nevertheless achieved similar predictive errors. These findings establish a simulation-based predictive result under the specified scenarios and errors; the available G219 records do not validate a slow-recovery mechanism in the field.