Thawing permafrost poses environmental threat to thousands of sites with legacy industrial contamination

The Arctic permafrost region is warming at least twice as fast as the global average, with some analyses indicating nearly four-fold faster warming. This rapid change undermines assumptions that permafrost provides a stable foundation and hydrological barrier for industrial infrastructure and waste containment. Historic industrial practices in the Arctic often relied on permafrost to confine wastes (e.g., drilling muds, fuels, heavy metals, radioactive materials) through methods such as buried dumps, drilling sumps, and disposal in closed basins. Notable incidents like the 2020 Norilsk diesel spill highlight risks from infrastructure destabilization on thawing ground. Despite these risks, a pan-Arctic assessment of industrial sites and associated contamination in permafrost regions has been lacking. This study aims to compile the spatial distribution of industrial sites within permafrost-dominated regions, estimate associated contaminated sites, assess types of toxic substances involved, and evaluate how present and future permafrost thaw under different climate scenarios may increase risks of contaminant release. The work addresses the need for science-based risk assessments and long-term management strategies under ongoing and projected climate change.

Prior work documents accelerated Arctic warming, permafrost degradation impacts on hydrology and infrastructure, and emerging biogeochemical risks as thaw mobilizes contaminants. Reviews have suggested potential releases (e.g., mercury) and highlighted permafrost’s reduced bearing capacity near 0 °C, increasing infrastructure failure risk. Case studies, including the Norilsk oil spill, demonstrate real-world consequences. However, a comprehensive pan-Arctic quantification of industrial sites, their contamination legacy, and projections of risk under future permafrost thaw has been missing. This study builds on regional contaminated site inventories (Alaska CSP, Canada FCSI) and permafrost modeling literature to provide a first-order circumpolar assessment.

• Study domain delineation: The permafrost model domain was defined using the Northern Hemisphere Permafrost Map (NHPM), retaining grid cells with permafrost occurrence probability >50%.
• Industrial site database: Compiled from OpenStreetMap (OSM) features (landuse/building=industrial) north of 55°N and APSEA (Atlas of Population, Society and Economy in the Arctic) infrastructure datasets (airports, ports, mines, pipelines, roads, petroleum fields). Overlaps were resolved with 1 km buffers. After intersecting with the NHPM permafrost domain, 5234 entries remained; a subset of about 4494 were counted as industrial sites within the model domain. Sites were categorized per IPCC sectors (AFOLU, Energy, IPPU, Waste), with many unlabeled/uncertain due to missing tags. Completeness was compared to the SACHI dataset, showing >85% spatial consistency but indicating OSM-APSEA undercounts absolute site numbers (40 ± 20% missing), implying conservative estimates.
• Contaminated site datasets: Used Alaska’s Contaminated Sites Program (CSP) and Canada’s Federal Contaminated Sites Inventory (FCSI). Both provide geolocations and cleanup status; CSP additionally allows classification by industrial origin through semantic analysis of site names. Chemical substances at CSP sites were extracted semi-automatically via keyword matching (Levenshtein distances) and matched to CAS registry entries; aquatic toxicity (LC50-fish, 96 h) was compiled from GESTIS and literature.
• Spatial relationship modeling: Conducted spatial cross-analysis between industrial and contaminated sites in Alaska and Canada. Fitted two inhomogeneous Poisson point process models (PPM1, PPM2) to CSP/FCSI data using Gaussian kernel density (bandwidth 50 km) to quantify how contaminated site density scales with industrial site density. Applied these models to the pan-Arctic industrial density map to estimate contaminated site intensities and totals by country.
• Validation for Russia: Compiled 102 media-reported contamination incidents (2000–2022) with locations, of which 44 were within the permafrost domain. Compared observed incident distribution across modeled intensity classes to random draws from modeled totals; observed counts fell within plausible modeled ranges, supporting applicability to Russia despite data sparseness.
• Permafrost stability projections: Used CryoGridLite (1D transient permafrost model) forced by ERA-Interim for historical climate and CMIP5 anomalies (CCSM4 and HADGEM2-ES) under RCP 2.6 and RCP 8.5 for 2020–2100. Model spin-up from 500 AD to 1979; one-degree grid resolution over the northern circumpolar permafrost region. Defined permafrost degradation by persistent talik formation (unfrozen layer ≥0.10 m above permafrost). Compared modeled talik presence to NHPM probabilities for consistency. Note: rapid thaw processes (thermokarst, thermal erosion), infrastructure feedbacks, and some infrastructure types (e.g., pipelines) were not included.

• Industrial site inventory: 4494 industrial sites were mapped within the Arctic permafrost model domain (OSM-APSEA). Among clearly labeled entries, Energy and AFOLU each exceed 10%, but >65% of sites are unlabeled/uncertain by sector.
• North American contaminated sites: CSP (Alaska) and FCSI (Canada) list ~8000 (AK) and ~22,000 (CA) contaminated sites; ~18% (≈3600) lie in permafrost-dominated regions. About 30% of contaminated sites in these permafrost regions are active (not cleaned up/uncertain status). In Alaska’s permafrost domain, >60% (≈850) of contaminated sites are linked to industrial or military activity. Registrations of new contaminated sites peaked in the 1990s, then declined (from ~90 in 1992 to 38 in 2019), echoing North Slope oil production trends.
• Substances: Fuels (diesel, kerosene, gasoline) and associated chemicals constitute about half of listed contaminants at Alaskan permafrost sites; high-toxicity metals (mercury, lead, arsenic) are among the top twenty substances.
• Pan-Arctic upscaling: Point process models estimate a total of 13,047–19,933 contaminated sites across permafrost-dominated Arctic regions. Country-level ranges: Russia 8,538–15,149; Canada 2,308–2,720; Alaska 1,186–1,493; Greenland 296–301; Svalbard 306–681. Most contaminated sites (70.5 ± 5.5%) are in Russia, reflecting ~85% of industrial sites; Canada+Alaska account for 23 ± 5% of contaminated sites and ~18% of industrial sites.
• Present exposure to thaw: As of 2020, simulations indicate ~22% (~1000) of industrial sites and 20 ± 4% (≈2200–4800) of contaminated sites are in regions where permafrost degradation is possible (persistent talik). About 15% of these sites are in grid cells that transitioned from stable to unstable permafrost between 1960 and 2000.
• Future exposure under warming scenarios: ~77% (~3500) of industrial sites and 60 ± 15% (≈5800–15,100) of contaminated sites are currently in regions simulated stable (talik-free). The number of sites in stable regions is projected to decrease by 3 ± 2% by 2050 under RCP 2.6 and by 46 ± 13% under RCP 8.5. By 2100 under RCP 2.6, the number of industrial sites in degrading permafrost regions increases by ~1100 (from 2020); contaminated sites affected increase by ~3400–5200. Under RCP 8.5, almost all industrial and contaminated sites fall within degrading permafrost regions by 2100.
• Policy-relevant estimate: Approximately 1100 industrial sites and 3500–5200 contaminated sites currently in stable permafrost are projected to begin thawing before 2100, elevating environmental risk.

The study demonstrates a widespread co-location of industrial infrastructure and contaminated sites across Arctic permafrost regions and shows that ongoing and future permafrost thaw will increasingly compromise infrastructure integrity and containment of hazardous substances. Even near 0 °C, permafrost loses bearing capacity, so failure can occur before modeled talik formation. Thaw will open new hydrological pathways, increasing mobility and dispersion of contaminants while also reducing site accessibility for mitigation and cleanup. The strong dependence of contamination density on industrial site density enables first-order risk mapping, but regional differences in industrial sectors and regulatory enforcement affect contamination occurrence. Results emphasize the necessity of extending risk assessments beyond mid-century, as permafrost processes and legacies will drive impacts into the latter 21st century. The pronounced divergence between RCP 2.6 and RCP 8.5 highlights the benefits of limiting warming to meet the 2 °C target. Transparent documentation (as in CSP and FCSI) is pivotal for science-based assessments; current gaps, especially in Russia and for legacy sites, hinder comprehensive risk and cost analyses. Monitoring, early-warning systems, and proactive adaptation at active sites could reduce abrupt-release risks, but systematic remediation and long-term monitoring of numerous legacy sites remain challenging. Economic implications from infrastructure destabilization and contamination management are likely substantial and underappreciated in cost-benefit considerations for Arctic development.

This work provides the first pan-Arctic synthesis linking industrial site distributions to thousands of legacy-contaminated sites within permafrost regions and projects how climate-driven permafrost thaw will escalate risks of contaminant release. Key contributions include: (i) a geospatial inventory of ~4494 industrial sites in permafrost-dominated areas; (ii) a data-driven estimate of 13,047–19,933 contaminated sites pan-Arctic, dominated by Russia; (iii) identification of fuels and toxic metals as prevalent contaminants; and (iv) projections showing a substantial increase in sites exposed to degrading permafrost, with nearly all sites affected under high-emissions scenarios by 2100. To mitigate future hazards, the study calls for long-term planning and management strategies that explicitly account for permafrost dynamics beyond 2050, improved and transparent contaminant inventories across the Arctic, development of monitoring and early-warning systems for at-risk infrastructure, and robust, long-term remediation plans for legacy sites. Future research should prioritize comprehensive circumpolar contaminated site databases (including Russia and smaller Arctic regions), quantification of contaminant stocks and leachability, integration of rapid thaw processes and infrastructure feedbacks into risk models, and evaluation of cost-effective monitoring and remediation strategies at scale.

• Data incompleteness and labeling: >65% of industrial sites are unlabeled by sector in OSM-APSEA; OSM-APSEA likely undercounts absolute site numbers (40 ± 20% missing vs SACHI). Historical/hidden infrastructure (e.g., revegetated drill pads, covered sumps, landfills) and extensive pipeline networks were not fully captured.
• Geographic data gaps: Lack of comprehensive, standardized contaminated site inventories outside Alaska and Canada (notably Russia) introduces large uncertainty in upscaling; Russian estimates differ between models by a factor of ~1.7.
• Modeling assumptions: The spatial relationship was modeled as an inhomogeneous Poisson process using North American data assumed representative of the Arctic; this may not capture region-specific regulatory/industrial differences.
• Permafrost model limitations: Coarse 1° spatial resolution and forcing uncertainty; simulations did not include rapid thaw processes (thermokarst, thermal erosion), infrastructure-induced ground warming, or effects of contaminants (e.g., freezing point depression), likely leading to conservative risk estimates. Talik-based stability is a proxy and may not reflect local engineering conditions.
• Site-level uncertainty: Local ground ice content, stratigraphy, and design choices (e.g., pre-excavation of permafrost) can strongly influence stability and contaminant mobilization but were not resolvable at pan-Arctic scale.