[Task] Running multiple crops: Vectorization of phenology and data of AgroManagement

[SarahAlidoost]

The parameters IDSL, DLO, DLC, and DVSEND in the DVS_Phenology class control when the simulation should stop. Depending on their values, DVS_Phenology sends a signal to the Engine to end the simulation. These parameters are closely related to AgroManagement settings such as crop_start_date, crop_start_type, crop_end_date, crop_end_type, crop_name, and max_duration.

The main issues are:

• The Engine can currently run only one crop model at a time, as defined in AgroManagement. It executes the Euler loop day-by-day for a single crop.
• The dispatcher can handle only a single signal from a single crop.

As a result, the existing tests loop over different datasets one crop at a time. Ideally, we want to run all crops together with one Engine instance. Achieving this will mainly require changes to the Engine, and later the phenology module should also be revisited.

Note: related pcse notebook is 04 Running PCSE in batch mode.ipynb. This include examples of running model with several agro management data using a loop. But we want to replace the loop with vectorization operations.

[fnattino]

With #112 I started to make sure the phenology model can be vectorized, allowing for the crop model simulations to run even after maturity (this is needed if we want to run for different locations, since not all locations will reach maturity at the same time). #113 shows a way to run several locations simultaneously using the current diffWOFOST framework (this is still for the same crop, though).

I wonder if the current system based on the agromanagement/dispatching is suitable to run several crops. The current framework creates/removes an instance of the crop model (e.g. phenology) whenever the start date of a given crop is reached. In the envisaged diffWOFOST applications it does not make sense to have several instances of a crop model, we want instead a single phenology model to run over different sets of parameters. Using a mask-based approach which indicates which elements have an active crop seems a more natural framework to work with.