Periodicity-aware similarity-weighted linear interpolation for missing-sample reconstruction in environmental sensor time series

Missing samples degrade time-series analysis, forecasting, and decision-making. Linear interpolation is a simple, interpretable, training-free baseline, but it ignores periodic structure commonly present in hourly sensor signals. This paper proposes Similarity-Weighted Seasonal Anchor Linear Interpolation (SW-SALI), a periodicity-aware method for univariate missing-sample reconstruction. SW-SALI augments the linear interpolation path with same-hour seasonal anchors from neighboring days and assigns each anchor a composite weight based on temporal proximity and contextual similarity to the interpolation boundary. This retains the model-free nature of linear interpolation while downweighting contextually inconsistent seasonal anchors. SW-SALI is evaluated on two benchmark environmental sensor datasets comprising 24 variables under MCAR, MAR, and MNAR masking mechanisms, five missing rates, and 30 replications per condition. Across the benchmark, SW-SALI improves over standard linear interpolation on signals with measurable same-hour recurrence, particularly under MCAR and MAR, while slope-preserving interpolation remains preferable for trend-dominant or weakly periodic signals. Column-level analysis shows when seasonal anchoring is effective and when direct or slope-based interpolation is more appropriate. We further quantify applicability using lag-24 autocorrelation and daily-frequency spectral power; the daily-frequency power ratio is positively associated with representative SW-SALI gain over linear interpolation. Runtime results show that SW-SALI is slower than direct interpolation baselines but remains sub-second per univariate series and substantially faster than Kalman smoothing. Overall, SW-SALI is a lightweight extension of linear interpolation for recurrence-aware reconstruction, not a universal replacement for slope-based interpolation.