Permafrost degradation increases risk and large future costs of infrastructure on the Third Pole

Warming and thawing of near-surface permafrost reduce substrate strength, increase mass movements, and produce thermokarst activity, making permafrost degradation a major threat to infrastructure in cold regions. Damage shortens infrastructure lifespan and increases maintenance costs, creating financial risks. In the Northern Hemisphere, nearly 70% of current infrastructure may face high potential for degradation by mid-century and substantial investments will be needed to maintain service function in the circumpolar Arctic. The Qinghai-Tibet Plateau (QTP), a key part of the Third Pole with extensive permafrost, hosts most mid- and low-latitude permafrost and a large share is warm and unstable. The region has critical infrastructure for transportation, energy, and buildings that support more than 10 million inhabitants and many more indirectly. The QTP has warmed at roughly twice the global average, increasing the vulnerability of warm permafrost and threatening infrastructure stability, though various engineering adaptations have been deployed (e.g., shading boards, ventilation ducts, thermosyphons, air-cooled and crushed rock embankments, special foundations for power lines). Previous work has focused on biophysical changes and less on vulnerability and especially economic impacts, leaving uncertainty about how much infrastructure will be at risk and at what cost, and which adaptation strategies are most economical at regional scale. Research question and aim: quantify future risk to current infrastructure from permafrost degradation on the QTP and estimate the additional costs required to maintain service function, with and without adaptation, under multiple climate scenarios (four SSPs and two Paris Agreement warming targets). The study integrates high-resolution projections of permafrost thermal state with multi-hazard indices and a lifespan replacement cost model.

Prior studies identified widespread risk to infrastructure from permafrost degradation across the Northern Hemisphere, with nearly 70% of infrastructure at risk by mid-century and multibillion-dollar additional investments needed in the Arctic. Economic damage estimates exist for Alaska (e.g., ~$1.6 billion under RCP4.5 for public infrastructure by century’s end) and Russia (tens of billions for roads, buildings, and maintenance by mid-century), but central estimates for the QTP were lacking despite its extensive and relatively warm permafrost. Engineering adaptations have been developed and applied on the QTP’s highways, railway, and power transmission lines. However, regional-scale assessments integrating permafrost hazard, infrastructure exposure, and economic costs, including adaptation benefits, have been missing, and previous work often separated biophysical drivers (temperature, precipitation, freeze-thaw, flooding) that are tightly coupled to permafrost change.

The assessment combines: (1) an ensemble statistical/machine learning projection of permafrost thermal state, (2) a composite multihazard index to classify permafrost degradation hazard levels for infrastructure, and (3) a lifespan replacement model to compute additional costs (net present value). Projections: Mean annual ground temperature (MAGT, depth of zero annual amplitude 10–25 m) and active layer thickness (ALT) were projected at 1 km resolution for a reference period centered on 2008 (2000–2016) and future periods centered on 2050 (2041–2060) and 2090 (2081–2100), under four SSPs (SSP126, SSP245, SSP370, SSP585) and two Paris Agreement warming targets (1.5 °C and 2.0 °C above preindustrial by end-century). MAGT used an ensemble mean of five models: GLM, GAM, SVR, RF, and GWR, trained with 253 MAGT boreholes using climate, soil, terrain, and radiation predictors (freezing/thawing degree-days, solid/liquid precipitation, soil bulk density, soil organic content, solar radiation, elevation). Uncertainties were quantified using 200 distance-blocked resampling runs (97.5th–2.5th percentile range); 10-fold cross-validation gave RMSE ~0.93 °C, bias −0.02 °C. ALT was estimated via a simplified Stefan approach using thawing degree-days and an E-factor (capturing vegetation, snow, soil texture) modeled from 157 ALT sites; the E-factor was held constant for future projections given sparse vegetation, low snow, and arid conditions on the QTP; cross-validation RMSE for E-factor ~18.7. Climate inputs were from downscaled and bias-corrected WorldClim datasets (CMIP6 for SSPs and CMIP5 for warming targets), adjusted to 1 km resolution. Soil and terrain inputs (SoilGrids, SRTM) were considered static. Hazard indices: Five indices were computed and combined by majority vote to classify each 1 km cell into high, medium, or low hazard: (i) Thermal index based on transition among permafrost temperature classes (cold, cool, warm) and permafrost disappearance; (ii) Settlement index combining relative ALT increase with volumetric excess ground ice (ground ice content derived from boreholes and Quaternary sediments, class-specific values of 5, 15, 35%); (iii) Bearing capacity index based on MAGT change and soil texture group (coarse/fine) using empirical linear relations; (iv) Risk zone index integrating bedrock exposure, soil grain size, ground-ice threshold (>20% vs ≤20%), and simulated thaw potential; (v) Expert-based multicriteria index weighting changes in MAGT and ALT, ground ice content, fine-grained sediment, and slope (weights from literature via analytic hierarchy process). Ties defaulted to medium hazard. Under SSP245 by 2090, this yielded 63.3% high, 34.7% medium, and 2.0% low hazard areas; under SSP126 by 2090, 16.9% high, 71.0% medium, 12.1% low. Cost model: An equivalent lifespan replacement model (after Larsen et al. 2008) computed additional costs as the difference in present value (2008 USD, 2.85% real discount rate) between permafrost degradation and normal conditions over 2008–2050 and 2008–2090. The model assumes permafrost-induced degradation shortens the useful life of infrastructure, increasing the frequency and cost of repair/replacement. Adjusted useful life reductions were assigned by hazard level and adaptation status (without adaptation: high 40%, medium 25%, low 15%; with adaptation: high 20%, medium 10%, low 5%), informed by long-term QTP highway observations and expert judgment. Infrastructure database: Current inventories included roads (five classes), railways, power lines, and buildings within permafrost regions: roads 9,389 km, railways 580 km, power lines 2,631 km, buildings 1,064,590 m². Each category had a useful life and replacement cost per unit; adaptation cost was parameterized as a percentage of replacement cost (roads 10–50% by grade, railway 20%, buildings 5%, power lines 5%). Adaptation measures (e.g., thermosyphons, air-cooled and crushed rock embankments, shading boards, specialized foundations) were assumed to extend useful life with associated incremental costs, already deducted in the “with adaptation” case. Infrastructure spatial data were sourced from national GIS catalogues and OpenStreetMap; power lines used a predictive mapping dataset. Uncertainties propagated from MAGT/ALT projections to hazard levels and exposed infrastructure fractions.

Permafrost thermal state change: Average QTP permafrost MAGT in the reference period was −1.72 °C (−2.23 to −1.30), ALT 2.11 m (1.75–2.48). Under SSP245, MAGT increases by ~1.55 °C by 2050 and an additional ~0.82 °C by 2090; ALT increases by ~0.62 m by 2050 and an additional ~0.26 m by 2090. Under SSP585, ALT increases by ~1.61 m by 2090. Keeping global warming below 1.5 °C reduces permafrost warming by ~0.75 °C relative to 2.0 °C. Spatially, western and southern QTP warm >1.5 °C by 2050 (SSP245) vs ~1.2 °C in the east and north; along the Golmud–Lhasa railway corridor, MAGT increases ~1.36 °C by 2050 and a further ~1.83 °C by 2090 (ALT +0.64 m and +0.96 m, respectively). Hazard exposure: By 2050, the fraction of infrastructure in high-hazard zones is ~28.66% (7.78–52.35) under SSP245, ~63.25% (23.25–78.70) under SSP585; by 2090, ~83.52% (73.85–95.46) under SSP585. Under SSP126 in 2050, ~17.62% (5.83–26.13). Hotspots include the Yellow River source area, Qinghai-Tibet engineering corridor, and Xinjiang-Tibet highway corridor. Current inventory at risk: In permafrost areas, the QTP hosts >9,389 km of roads, 580 km railways, 2,631 km power lines, and 1,064,590 m² buildings. By 2050 (SSP245), high-hazard exposure is projected as follows: roads 38.14% (19.92–60.36), railways 38.76% (11.35–71.30), power lines 39.41% (range reported inconsistently), buildings 20.94% (14.82–31.28). Paris targets substantially reduce exposure by 2090: controlling warming to 1.5 °C roughly halves high-hazard infrastructure relative to 2.0 °C. Economic impacts: Additional costs (net present value, 2008 USD, 2.85% discount rate) to maintain service function of current infrastructure: 2008–2050, SSP245: $3.98 billion (3.33–4.75), with transport accounting for ~93% (roads 87%, rail 6%); SSP126: $3.62 billion (3.01–4.44), i.e., ~$0.35 billion lower than SSP245. 2008–2090, SSP245 without adaptation: ~$6.31 billion; with adaptation: ~$4.99 billion (20.9% savings). Under SSP126, 2008–2090 with adaptation: ~$4.10 billion; SSP370: ~$5.42 billion; SSP585: ~$5.49 billion. Without adaptation, SSP585 over 2008–2090: ~$7.23 billion; SSP370: ~$7.12 billion. Paris targets by end-century: 2.0 °C case ~$5.65 billion (without adaptation); 1.5 °C saves ~$1.32 billion relative to 2.0 °C. Adaptation benefits: Potential cost savings from adaptation are ~15.36% by 2050 (SSP245) and ~21% by 2090; range 12% (SSP126 by 2050, ~$0.45 billion) to nearly 24% (SSP585 by 2090, ~$1.74 billion). Low-grade roads generally show poor economic return for adaptation under current cost ratios. Spatial distribution: Highest cumulative additional costs (to 2090) concentrate in central-eastern prefectures (Yushu, Golog, Haixi, Nagqu); lower in northern/southeastern prefectures (Bayingolin, Lhasa, Linzhi, Qamdo, Garze). Low-carbon scenarios reduce costs across prefectures; Ngari corridor remains strategically important.

The study shows that permafrost on the QTP warms almost synchronously with air temperature due to weak surface thermal offsets (sparse vegetation, dry soils, thin snow, low organic matter), making infrastructure particularly vulnerable in warming scenarios. The integrated approach demonstrates that a large fraction of existing infrastructure will transition into high-hazard zones, driving substantial additional costs even under green pathways and ambitious warming limits. Adaptation measures are generally cost-effective at regional scale, with savings that grow under stronger warming and longer horizons, though not universally for low-grade roads. Comparing to other regions, projected economic damages on the QTP exceed Alaska’s estimates despite a smaller infrastructure inventory, likely due to faster degradation of warm, arid, climate-driven permafrost and methodological differences; they remain lower than Russia’s due to the much larger infrastructure base there. The results quantify benefits of mitigation: weaker mitigation (SSP370, SSP585) increases the share of high-risk infrastructure and additional costs, whereas SSP126 and Paris-aligned warming targets reduce both exposure and costs, underscoring the value of global emissions reductions. High permafrost maintenance costs and warming amplification on the QTP may exacerbate social inequality; targeted investments in resilient infrastructure, cost-reducing adaptation technologies, and coordinated green development policies can reduce vulnerabilities and enhance regional connectivity and resilience.

This work provides the first regionally comprehensive quantification for the Third Pole of how permafrost degradation translates into infrastructure risk and additional economic costs, and how adaptation and climate mitigation can reduce these burdens. By combining high-resolution projections, a multi-index hazard framework, and a lifespan replacement model, the study estimates multibillion-dollar additional costs through 2050 and 2090, identifies spatial hotspots, and demonstrates that strategic adaptations can save roughly 15–24% of costs, while limiting global warming to 1.5 °C yields further large savings relative to 2.0 °C. The findings support planning for resilient infrastructure, prioritizing adaptation along critical corridors, and highlight the broader benefits of low-carbon development. Future work should refine damage functions and cost estimates with richer engineering and cost datasets, incorporate dynamic changes in infrastructure portfolios and local processes (e.g., talik formation), and further integrate engineering economics to improve present value calculations and uncertainty characterization.

Uncertainties arise from: (1) Permafrost projections: ensemble statistical/machine learning models capture climate-driven changes but cannot fully resolve local nonlinear processes (e.g., taliks, ecosystem feedbacks) or local disturbances; assuming constant E-factor and static soil/terrain may under- or overestimate hazards. (2) Infrastructure database: replacement and adaptation cost data are scarce, sometimes sensitive, and compiled from heterogeneous sources; future infrastructure expansion is not included, rendering estimates conservative. (3) Damage relationships: reductions in useful life by hazard level are based on highway observations and expert judgment and applied uniformly across infrastructure types due to data gaps; this simplification may not capture type-specific sensitivities. (4) Lifespan replacement model: uses simplified annual replacement costs and a low discount rate (2.85%); present value results are sensitive to the analysis horizon because end-of-life replacements are not treated as future values; while this affects absolute costs under both cases, it likely has limited impact on additional cost differences. Overall, the estimates are considered conservative, and more detailed engineering-scale data would improve accuracy.