Non-steady-state models

[VEZY]

Some models are non-steady-state, because they change abruptly the state of the system, e.g. they modify the inputs they are given (see #107), or they remove nodes from the plant graph.

This is a problem in the current implementation of PlantSimEngine, because it is not allowed to:

1. Remove nodes, and
2. Update input variables (see Handle models that update variables (variable in both inputs and outputs) #107).

A solution to 1. would be to provide a function remove_node! that does the reverse of add_node!, that is to say, delete a node from the plant graph, delete it from the vector of nodes at its scale, and delete the references to its variables in other multiscale variables.

A potential solution to 2. would be to define some models as "non-steady-state" (e.g. using a trait). Those models would be applied without a dependency graph (as is) after all other models, and would be authorized to update the state of variables. For example in XPalm, when a leaf is harvested, we keep its node in the plant graph, but we change its state in the status, and we update some variables accordingly (e.g. biomass = 0.0, litter = previous biomass). This is forbidden in PlantSimEngine, so we use a trick: we don't put updated variables as outputs of the model (just inputs). The risk here is that we may call the model at a wrong time.

[Samuel-amap]

To give other potential examples of non-steady state models (besides pruning) :

• Other forms of plant disruption such as hedgelaying (https://en.wikipedia.org/wiki/Hedgelaying)
• Hazards such as fire
• Some potential plant-plant (honeysuckle strangulation ?) or plant-animal interactions

Some of these might affect the MTG in much more drastic ways than 'simply' removing a branch.

Also, some other interactions (root fusing) could also be modelled in a non-steady state manner, perhaps ?

[Samuel-amap]

Additionally, there is a need to be clear regarding the behaviour of later models making use of variables computed off the MTG if ever the decision is made to opt for a different approach (for instance, a biomass-dependent variable, is it or isn't it affected by the sudden change in biomass ?)

[Samuel-amap]

Another potential pitfall that has just come to mind :

Some nasty corner-cases are likely to occur and need to be handled. For instance, what if a model creates a leaf that is then immediately pruned ? How is that reflected in the outputs ?

The non-steady state stage seems like it would deal with this example without many issues, but it's quite likely some other subtle interactions could cause some very tricky behaviour to happen.