Guardrails

[rgknox]

Issue #1090 revealed the possibility that some allometries could create unusual plant sizes or organ proportions. In that example, some trees became so large, that the stature growth integrator had a tough time generating precise results.

During review of this issue, several developers thought it could be useful, if we come up with strategies to apply optional guardrails to different aspects of the model. Ideally, the guardrails are there to prevent very unusual, and hopefully very seldom model states from occurring and breaking the model.

For instance, in the allometry example, we could set caps on maximum sizes. When the maximum size is reached, we would shunt all NPP to either reproduction, exudation, or respire it away, after the plant has replaced turnover. This is just an example. This type of behavior would kick in only for a plant or organ size that is far beyond the maximum expected. Someone may want this type of behavior if they are running a ensembles of parameters, and it is more desirable that all the runs complete, even if they have strange results, rather than one of their runs failing and breaking their workflow.

Feel free to use this space to come up with ideas for different guardrails, and in what contexts it would be useful to apply them. Note that we could keep guardrails default off, as a user defined parameter.

cc @ckoven

[rgknox]

After some discussion, there is some thought that current issues with integration errors relate to growth, may have something to do with capping in allometry and/or the use of allometries that rely on a sudden change in allocations. An example is hitting a maximum size threshold and pivoting to growing the stem diameter instead of stem and height simultaneously. So instead of applying these guardrails, we will wait and see if phasing out capping-style and sudden change style allometry in favor of asymptotic and smooth functions gets rid of these problems.

[rgknox]

closing feel, free to reopen