Buy a forest carbon credit and you are trusting a promise: the trees paid to stay standing will keep their carbon locked away for a century. Wildfire, drought, beetles, and windstorms do not sign that contract. A study published in Nature on May 20 puts a number on how badly the system underprices that gap, and the number is not small.
Chao Wu of Tsinghua University and the University of Utah, working with colleagues at CarbonPlan and UC Irvine, set out to map where and how much stored forest carbon is likely to be lost to natural disturbance across the contiguous United States over the next 100 years. Then they compared that risk to the financial cushion the offset market keeps on hand for exactly this problem. The cushion is too thin. By their estimate, the buffer pool of the largest US forest carbon program is likely too small by an average factor of 6.3.
What a buffer pool is supposed to do
Carbon offset programs know that some credited forest will burn or die. To cover it, they hold back a fraction of every project's credits in a shared reserve called a buffer pool. Think of it as self-insurance. If a credited forest goes up in smoke, the program cancels an equivalent amount from the pool so the atmospheric math still balances. The whole scheme only works if the pool is stocked to match the real risk.
The research team argues it is not. They combined forest inventory records, satellite observations, disturbance modeling, and machine learning to estimate the 100-year reversal risk, the chance that carbon counted as stored gets released, forest by forest. Climate change pushes that risk up across the country, and it concentrates hardest in California and the Intermountain West, where fire seasons have grown longer and hotter. Those are the same regions that host a large share of US offset projects.
Off by more than an order of magnitude at the edge
The headline factor of 6.3 is an average. Once the authors folded in uncertainty about future emissions scenarios, how severe disturbances might get, and carbon stored in pools like soil and dead wood, the shortfall ranged from 2.2 to 8.0 times too small. Even the low end means the reserve holds less than half of what the risk warrants. At the high end it is a rounding error against the exposure.
Why does this matter beyond accounting? A credit that turns out to be backed by an underfunded reserve is a credit that may not represent a real, durable ton of carbon kept out of the air. Companies buy these to cancel their own emissions. If the forest releases its carbon and there is nothing left in the pool to cover the loss, the offset quietly becomes fiction, and the climate benefit that was already sold does not exist.
The team did not just flag the problem. They produced spatially explicit maps of long-term carbon loss risk, the kind of granular input a program could use to set buffer contributions that actually reflect where a project sits and what it faces.
Where the estimate stops short
This is a modeling study of the contiguous United States, so the specific 6.3 figure belongs to CONUS forests and one program's design, not to every offset market on Earth. Projecting disturbance a century out means leaning on climate and severity scenarios that carry real uncertainty, which is exactly why the authors report a range rather than a single verdict. The work estimates risk and reserve adequacy. It does not audit individual projects or claim any particular forest will burn. And it does not argue that forests are a bad climate tool. The authors are explicit that cutting fossil fuel emissions remains the priority and that forests still matter. Their point is narrower and harder to dodge: the insurance math needs revising to survive the climate it is being asked to weather.
Nature-based solutions have been sold partly on the appeal that a living forest is a simple, visible place to park carbon. This study is a reminder that a forest is also a flammable, drought-prone, insect-vulnerable thing, and that treating it as a stable vault requires paying for the risk that it is not.
Comments