Ecosystem model spin-up: Estimating steady state conditions in a coupled terrestrial carbon and nitrogen cycle model

Abstract

We tested a variety of methods for determining steady state solutions for Biome-BGC, a coupled model of terrestrial water, carbon, and nitrogen dynamics. Our objective was to identify methods that could reduce the computational cost of model spin-up relative to simulations running under the model's native dynamics, while retaining or improving upon the simulation quality, where quality is judged by comparison to assumed values for the true steady state solution. Two classes of methods were tested: ad hoc methods that approximate steady state by taking advantage of specific characteristics of the modeled dynamics to produce individual time trajectories through the model state space, and general multivariate minimization methods that iteratively explore multiple time trajectories through state space in search of a reasonable steady state solution. We examined the behavior of these methods for both woody and herbaceous vegetation simulations. We found that both the ad hoc and the generalized methods, parameterized appropriately, could provide reductions in computational cost of 50–75% compared to the model's native dynamics. With the exception of the generalized methods for the woody vegetation case, we also found that the quality of the simulated steady state solution was as good as or better than the native dynamics approach. The one method which performed consistently well across sites and vegetation types involved an acceleration of decomposition rates for the spin-up phase, resulting in 73 and 66% reductions in computational cost for woody and herbaceous vegetation types, respectively.