Study of spatiotemporal changes within the abandoned flotation plant site of Kirki (NE Greece) via clustering on Sentinel-2 data

The abandoned flotation plant at Kirki, northeastern Greece, represents one of the most prominent unmanaged mining legacies in the eastern Mediterranean. Since operations ceased in 1997, seven tailing ponds covering several hectares have remained exposed to atmospheric weathering, generating persistent acid mine drainage and mobilizing heavy metals — primarily Pb, Zn, and Cd — into the surrounding drainage network and the Eirini river basin. Despite an established body of geochemical and hydrogeological research at the site, the spatial extent and temporal dynamics of surface mineral assemblages have never been examined through satellite-based spectral analysis. This thesis applies multi-temporal remote sensing to characterize contamination patterns across the Kirki tailings complex between 2015 and 2022. Thirty-one cloud-free Sentinel-2 acquisitions were processed to extract per-pixel reflectance time series from 458 active pixels distributed across seven tailings ponds and three near-pond zones. Each pixel's spectral signature was compared against a library of 13 mineral endmembers (from the USGS Spectral Library) — including jarosite species (K-, Na-, H₃O-jarosite), goethite, hematite, gypsum, butlerite, and lead-bearing phases (cerussite, anglesite, hydrocerussite, pb-jarosite, beaverite) — using Pearson correlation as the primary similarity metric. The temporal evolution of mineral assemblages was further analyzed through K-means clustering (k=4), hierarchical clustering with Ward's linkage, and the Hungarian algorithm for cross-date cluster matching. Cumulative 20-day precipitation records preceding each acquisition date were integrated to assess the influence of rainfall on spectral variability. Four geochemically distinct clusters emerge consistently across the observation period: a dynamically transitional zone dominated by butlerite and hydrated sulfates (Cluster 1), a peripheral system characterized by hydrated sulfate assemblages (Cluster 2), a stable contamination regime associated with iron oxides and jarosite (Cluster 3), and a persistent jarosite-dominated zone concentrated in the northern ponds (Cluster 4). Central ponds (1–4) exhibit high temporal stability while peripheral zones undergo seasonal transitions. Contamination patterns, as revealed by spectral clustering, extend beyond the physical limits of individual ponds, suggesting that pond boundaries are poor proxies for actual geochemical footprints. The near-pond areas immediately south and east of the main pond complex express a coherent multi-mineral assemblage of comparable or greater spatial extent than any individual pond, indicating that current management perimeters appear to substantially underestimate the true footprint of active geochemical alteration. Precipitation emerges as the dominant environmental control: drought conditions reinforce spectral stability and jarosite preservation, while cumulative rainfall exceeding 60 mm triggers mineral dissolution, butlerite mobilization, and cluster reorganization across the site. More broadly, results demonstrate that applying clustering techniques to spectral data can resolve sub-pond-scale contamination heterogeneity and track its seasonal evolution without field access.

Authors dc:creator

• Κωστάκης Αλέξιος
• Kostakis Alexios

• Θετικές Επιστήμες
• Science

Language dc:language

English

Identifier

uoadl:5374650

OAI identifier oai:identifier

oai:lib.uoa.gr:uoadl:5374650