Improved performance of offline ALM simulations

Abstract

The current ACME model architecture is structured for efficient coupled simulations, but offline model land simulations suffer from slow execution times and inefficient handling of input and output data streams.  The capability to perform rapid offline land model simulations significantly improves the pace of model development, testing and uncertainty quantification (UQ) efforts.  Both offline and coupled simulations also require expensive land spinup simulations.  Here we present an alternative structure for handling input data in ALM that bypasses data-atmosphere and the model coupler and stages data in the system memory at the time of initialization.  We also improve the existing "accelerated decomposition" algorithms in ALM by introducing a climate-dependent, rather than constant, acceleration factor for soil organic matter pools.  In addition, introducing accelerated vegetation mortality helps to reduce model spinup times significantly.  Finally, a new scripting framework is presented to perform efficient single-gridcell land model simulations for model-data comparisons and benchmarking.