Gigatonne-scale carbon dioxide removal (CDR) is essential in climate scenarios that limit long-term warming to 1.5 °C, alongside trillions of dollars of investment. However, carbon removal projects frequently face under-delivery risks: forests burn and geological storage sites leak. Current carbon removal markets lack a standardised, quantitative measure of permanence risk, often relying on qualitative labels or buffer pools whose sizing is only weakly linked to quantified risk. This research proposes the carbon at risk (CaR) metric, inspired by value at risk (VaR) in finance, to provide a continuous probabilistic measure of the shortfall between contracted removals and realised storage at a specific confidence level and time horizon.

The CaR framework

The CaR metric quantifies removal shortfall risk in physical rather than monetary units. It is defined as the gap between contracted removals (Q) and the lower-tail quantile of realised removals (D) over a specified time horizon (T). CaR serves two primary functions: it quantifies the storage shortfall that will not be exceeded with a given probability (such as 95%), and it determines the size of the buffer required to make a delivery obligation safe. CaR decomposes into a contracting gap (the difference between contracted and expected removals) and a tail component (the additional shortfall in a worst-case outcome). This horizon-specific measure allows for risk assessment well beyond the terms of individual contracts, accounting for accumulating reversal hazards.

CaR assessment in practice

Operationalising CaR requires estimating the distribution of potential removals over a chosen horizon. For established approaches like afforestation, empirical calibration can be used, while emerging technologies like direct air capture with geological storage (DACCS) require process-based simulation models. The research provides two proof-of-concept applications: forest carbon and geological storage. These case studies assume contracted tonnes have already entered the reservoir and focus on reversal mechanisms rather than additionality or initial delivery risks.

Forest carbon removals

Forest carbon is modelled based on two fundamental determinants: fire and post-fire regrowth. Simulations across three regions—Papua in Indonesia, Mato Grosso in Brazil, and California—show distinct risk profiles. California and Mato Grosso have similar mean burn fractions (1.81% and 1.84%), but differing regrowth rates of 2.0% and 3.0% per year respectively. California also carries a more skewed fire-size distribution, with a largest recorded annual burn fraction of 11.3% against 4.6% in Mato Grosso. For California, single-project 95% CaR can exceed 70% over 200 years. Risk management through diversification is possible, but common drivers like climate set a floor for risk reduction; pooling 100 California projects lowers CaR by only 7 to 9 percentage points due to inter-project correlations.

Geological storage after carbon capture

Using the Storage Security Calculator, the research estimates CaR for geological storage under different regulatory scenarios over a 200-year horizon. The 95% CaR is 0.15% for well-regulated offshore storage, whereas it rises to 17% for poorly regulated onshore storage—a two-orders-of-magnitude difference. Onshore risk is driven primarily by legacy wells serving as dominant leakage pathways. Even over extreme horizons, well-regulated offshore storage remains low-risk, with 1,000-year CaR rising modestly to 0.66% and 10,000-year CaR reaching only 6.3%.

Portfolio design under risk

The two technologies differ in both risk magnitude and drivers (climate sensitivity for forests versus regulatory quality for geological storage). By combining these in a portfolio, buyers can manage tail risk cost-effectively. In a stylised simulation where forest removals cost US$40 per tCO2 and DACCS costs US$500 per tCO2, the cheapest way to meet a delivery target of 30,000 tCO2 with 95% confidence is often an interior mix of both. For example, under certain correlation conditions, the optimal portfolio consists of 61 DACCS and 45 forest projects, costing US$16,150,000. As correlations within technology types rise, the value of diversifying across different removal pathways increases.

Discussion

The CaR framework aims to transform carbon removal procurement from qualitative assessments into transparent portfolio design. By expressing permanence risks on a common scale, it enables risk-adjusted comparisons, actuarial bases for insurance pricing, and calibrated buffer sizing. Standardised risk-return characterisations could help unlock the institutional investment required for gigatonne-scale carbon removal. While data limitations remain, CaR makes assumptions explicit and auditable, providing a more robust alternative to current flat-rate buffer heuristics.