Evaluation of Pipeline Leakage Phenomena in Actual Accident using Sentinel-2 Time Series Imagery Data

Underground pipelines are difficult to inspect because they are buried and extend over large areas. Satellite imagery provides repeated observations along pipeline routes and may support preliminary leakage screening. However, the applicability to actual facilities has not been sufficiently examined. In this study, the feasibility of detecting pipeline leakage at an actual leakage site is investigated. The target facility is an irrigation pipeline system. The leakage was identified after the road surface above the pipeline collapsed, and the pipeline was later repaired. Sentinel-2 Level-2A imagery from January 2020 to December 2025 is used. Clear-sky screening with the Scene Classification Layer (SCL) retains 124 images from 262 scenes. An image is accepted when at least 50% of a 2 × 2 pixel window (20 × 20 m) is free of cloud, cloud shadow and cirrus. Reflectance is averaged over this window. Band 12 (SWIR2), natively 20 m, is resampled to 10 m. A Prophet model is fitted to pre-leakage observations. Prediction residuals are standardized using residuals from a validation period that is excluded from fitting. SWIR2 reflectance in the leakage section is lower than the predicted value during the leakage period. A negative shift is observed in the prediction residuals, and the residuals tend to recover after repair. Standardized anomalies are also computed for 1,042 sections along the route. The median anomaly is −0.52 for the 14 sections within 60 m of the leakage point, against −0.07 for the 992 sections beyond 200 m. The difference is significant (Mann–Whitney U test, p = 1.7 × 10⁻⁸). The decrease in SWIR2 reflectance is consistent with increased surface moisture caused by subsurface leakage. Sentinel-2 time series imagery therefore provides basic information on temporal and spatial changes associated with pipeline leakage. In future work, further leakage cases and surface conditions will be examined.