Notes on ILAMB implementation in zppy

[chengzhuzhang]

When ILAMB was first implemented, I took some notes from ILAMB developers and thought it is useful to share here (copied from #197):

ILAMB has some underlying assumptions (Thanks to @minxu74, one of the ILAMB developers, to help me understand):

• The benchmarking practice is recommended applying to historical simulations, i.e. simulation data at least partially over the time period of the source data (year 2000-). ILAMB will match the overlap of the model and obs years, except using "--model_year" to shift year to create overlapping.[This is how it is implemented in zppy] ILAMB can work with continuous time segments without pre-concantenation.
• if there is no areacella, ILAMB will use longitude and latitude information to compute areas (CellAreas function in ilamblib.py); if the sftlf variable is missing, 100% land area is assumed. This is an approximate method.
• the cmip.cfg will be used as base configuration file. This .cfg is specific for CMIP5 and 6 benchmarking. The other file ilamb.cfg is used to run ilamb in a default setting, some metrics are removed in the cmip.cfg as the observational data’s uncertainty is large from feedbacks of users and rubisco team members. But I will defer to BGC to provide a more specific ilamb config to fit the research need.