Leakage Detection in Levees Via Statistical Analysis of Distributed Temperature Sensing Data

Modern levee monitoring is progressively shifting toward systems capable of collecting large volumes of spatially distributed and temporally continuous data, with the goal of providing early warnings before structural failure occurs. Among these technologies, Distributed Temperature Sensing (DTS) offers high-resolution thermal measurements along fibre-optic cables, enabling continuous assessment of seepage-related processes potentially over kilometre-scale levee sections. However, the large amount of data generated requires automated methods to quickly identify meaningful thermal anomalies. This study presents a methodology that integrates DTS with multivariate statistical analysis to detect early thermal signatures indicative of leakage. Principal Component Analysis (PCA) is employed to reduce data dimensionality and highlight dominant temperature patterns, while Independent Component Analysis (ICA) isolates statistically independent components associated with advective heat transfer caused by seepage. The approach was implemented on a levee section along the Adige River in the province of Bolzano (Italy), where DTS fibres and conventional sensors were installed in multiple vertical boreholes. During a flood event, the PCA–ICA framework successfully detected transient thermal anomalies consistent with observed seepage on the landside, demonstrating its diagnostic capability. The results show that the proposed PCA–ICA framework enhances the interpretation of DTS data, enabling early identification of anomalous hydraulic responses within the structure and supporting data-driven decisions for levee safety management.