Implement model spin-up

[dalonsoa]

From the proposal

• We need a mechanism to allow a model to run for many years while holding everything else constant. The principal use of this is to allow initial estimates of input data to converge to a biologically stable state before starting the ‘actual’ simulation from those stable conditions.
• There are two possible mechanisms for terminating the spin up process: run for a fixed number of years or run until a set of watched variables generate stable results to a given precision. It’s probably good to have both options and they could work together.
• In most cases, the spin up process can be thought of as a set of sequential runs with initial runs that we might want to converge to a steady state before we start the simulation, and then an actual simulation using outputs from those convergence runs. For example:
  • Run models with all frozen apart from Hydrology for 10 years, save soil moisture.
  • Run models with all frozen apart from Soil for 500 years, save carbon variables.
  • Run models with all active for 100 years, using the saved outputs from the first two runs (the actual simulation).
• That could be done manually using three runs and configs but it would be useful to have a configuration mechanism that:
  • Specifies a set of spin up “epochs” with different model settings and spin up termination rules
  • Identifies which variables should pass from spin up into the main simulation.
  • In many cases, we will want to restart the ‘actual’ simulation with different scenarios after the burn in, so it might be cleaner to setup a burn in process that runs models and then updates the input data, ready for the actual simulation.
• The immediate use case here is the Hydrology model, but any model could conceivably want to use this process.

[dalonsoa]

@vgro , we will need example models to run and test if what we are implementing actually works, ideally the three scenarios described above of running hydrology only, soil only and finally the full thing.

[dalonsoa]

@vgro , do you have any update on this. I can give a start - probably next week - but proper testing will not be possible without some exemplar scenario to run.