First-passage-based boundary estimation in functional data

We study a functional threshold-crossing setting in which a population-level boundary evolves smoothly over time and crossing events are observed through subject-specific functional trajectories. Motivated by biomedical and environmental applications where brief excursions above a critical level may trigger risk activation, we formulate boundary recovery as a first-passage identification problem and propose a practical estimation strategy based on smoothing observed crossing pairs. The boundary is represented using penalized cubic B-splines with roughness control, and tuning parameters are selected by generalized cross-validation. We further emphasize the distinction between (i) a time-varying benchmark used to define first-passage times and (ii) a reconstructed conditional crossing-level curve summarizing typical values at the moment of transition. A Monte Carlo study demonstrates that the proposed estimator accurately recovers a variety of smooth threshold shapes under different sample sizes and noise levels, with improved performance as crossing information increases. An illustrative application to country-level age-standardized colon cancer incidence trajectories demonstrates how the framework yields stable empirical boundary reconstructions for male and female series, while explicitly accounting for left-censoring induced by the limited observation window.