When does seasonal refinement improve stineman interpolation for missing air-quality time series? A similarity-weighted seasonal anchor approach

Environ Monit Assess. 2026 Aug 6;198(9):922. doi: 10.1007/s10661-026-15770-8.

ABSTRACT

Missing values are common in air-quality monitoring records and can distort downstream analyses when reconstruction methods do not match the temporal structure of the data. This study examines whether a lightweight seasonal refinement can improve shape-preserving Stineman interpolation for hourly air-quality series and when that refinement is reliable in practice. To address this question, we propose similarity-weighted seasonal anchor Stineman interpolation (SW-SAST), which preserves a local Stineman anchor while selectively borrowing cross-day information when nearby daily analogues are compatible with the local estimate. The method was evaluated on the UCI Air Quality and Beijing PM 2.5 datasets under MCAR, MAR, and MNAR mechanisms, with missing rates from 10% to 50% across 30 repeated runs per condition. The evaluation includes a univariate LSTM reference baseline, RMSE, MAPE, Pearson’s R, confidence intervals, and contiguous gap-length experiments. SW-SAST slightly improved performance on the UCI dataset, especially under MCAR and MAR, but it did not outperform the original Stineman interpolator on the Beijing dataset. The contribution is therefore not a universal replacement for local interpolation, but a clearer account of when similarity-weighted seasonal borrowing can strengthen or weaken imputation in air-quality monitoring records.

PMID:42560544 | DOI:10.1007/s10661-026-15770-8