This is the follow-up issue to #95, which dealt with potential evapotranspiration (PET).

So far, our evapotranspiration modularisation efforts have addressed most "main models". The only exception is lland_v3 (besides the still experimental application model lland_v4), which does not calculate AET based on previous PET estimates but determines it directly after Penmen-Monteith. Now, when modularising AET, we can finally also tackle lland_v3, which is currently of high priority for us.

All implemented PET submodels both work as submodels and stand-alone models, which simplifies testing and playing around with them a lot. At first glance, it seems AET models cannot work as stand-alone models because they need to know and want to update the states of their main models. However, maybe we can again find a solution here. Otherwise, we would need to introduce additional testing functionalities (perhaps main models designed explicitly as test models), which seems like much extra effort.

The AET models must deal with more special cases than the PET model. First thoughts:

• Different land use types often require different handling, especially water areas.
• Often, AET removes water from two storages: the interception and the soil water storage.
• Some models handle snow evaporation differently from "normal" evapotranspiration. (lland_v3 is a highly complex example of this.)