Theoretical growth models are the first basis for computing the level of carbon stocks. It is fortunate that in forests tree growth and carbon stocks are highly correlated – also including the below ground biomass which is proportional to the above ground mass by allometry. Therefore the existing forest growth models can provide a starting point for estimating both the baseline and the increased growth levels.

Unfortunately, almost all the current models lack very long-term accuracy – and permanency is key for high-integrity carbon offsets. In many models this inaccuracy rises from design: they were designed to model single rotations that would then be repeated. If the growth periods are extended much above, the accuracy drops rapidly. Therefore, for proper and trustworthy carbon accounting, Aurora Forealis must contribute to the development/refinement of a new set of models (our concrete program work on this is described in the next section under Luke.) This model refinement is also needed for measuring model uncertainty. On one hand, the currently used simple, linear statistical models are easy to understand and prod, but they cannot count for the long-term effects. On the other hand, models that are non-linear, complex, and with a lot more parameters, can potentially be more accurate, but also a lot more uncertain. And from Bayesian perspective, this uncertainty increases very rapidly.¹Thus controlling both accuracy and predictability will always be a trade-off, and often a judgmental/empirical one in that.

1. Generally, the more non-linear the model is and the more parameters it has, the more accurate it can be made, but also a lot more uncertain. ↩︎