Twice the Atmosphere's Carbon, Thawing Unmodeled

Permafrost contains approximately 1,600 Gt of carbon — twice the amount currently in the atmosphere — and is rapidly thawing. The Arctic-Boreal permafrost region is already a net warming source, emitting ~51 Tg CH4/yr from natural sources. Including permafrost thaw and fire emissions reduces remaining carbon budgets by 25% for 1.5°C and 17% for 2.0°C targets. But we cannot accurately predict the rate, magnitude, or chemical form (CO2 vs. CH4) of future carbon release because abrupt thaw processes — which may dominate total emissions — are not represented in any climate model used for IPCC projections.

Permafrost carbon feedback is potentially the largest unmodeled positive feedback in the climate system. Methane is 80x more potent than CO2 over 20 years, making the CO2-vs-CH4 partition critically important for near-term warming. If abrupt thaw releases carbon primarily as methane rather than CO2, the warming impact could be orders of magnitude larger per unit carbon than gradual thaw models predict. The Arctic is warming 2–4x faster than the global average. Permafrost thaw also threatens Arctic infrastructure (roads, buildings, pipelines) serving ~4 million people, with estimated damage costs of $30–50 billion by 2060 in Russia alone.

Climate models include only gradual top-down thaw; sudden ground collapse (thermokarst), thaw slumps, and lake dynamics are not represented at all — yet these abrupt processes may account for the majority of near-term emissions. Winter emissions may account for up to half of annual methane totals but are almost never measured due to extreme conditions. Monitoring relies on outdated wetland maps that do not capture rapid landscape change. Scaling from point measurements to regional estimates fails because Arctic landscapes are extremely heterogeneous — ponds and wetlands cover only ~10% of area but produce roughly two-thirds of methane emissions. Ecosystem models neglect biotic influences (vegetation succession, decomposer activity, herbivore impacts) that significantly modulate permafrost carbon feedbacks. Recently discovered pathways, such as aquatic grasses acting as "methane straws" through hollow structures, were entirely absent from models.

AI-powered analysis of high-resolution satellite imagery (the Permafrost Discovery Gateway approach, funded at $8M+ by NSF and Google) for landscape-scale thaw tracking — the DARTS dataset has already mapped >43,000 thaw slumps using deep learning, dramatically increasing the known inventory. Continuous year-round monitoring stations including winter measurements. Process models that incorporate abrupt thaw, thermokarst lake dynamics, and biotic feedbacks. Integration of tree-ring records, drone surveys, ground-penetrating radar, and permafrost drilling data to reconstruct thaw histories and validate predictive models.

A student team could use publicly available satellite imagery (Sentinel-2, Landsat) and machine learning to map thermokarst features in a specific Arctic region, comparing their automated detection to the DARTS dataset and ground-truth data. Alternatively, a team could build a process model for a single thermokarst lake's carbon emissions, incorporating water chemistry, temperature depth profiles, and ebullition (bubble) flux data from published field studies. Relevant skills: remote sensing, machine learning, biogeochemistry, environmental modeling.