Equilibrated state in ORC restart files from training data based on compressed forcing low accuarcy

[dsgoll123]

The number of pixels in equilibrium during the re-run is much lower when the SBG restart files is derived from training using compressed forcing (RERUN06) compared to ones from training using 'global' forcing (RERUN05): 53 % vs 75% (years 201-300 of re-run)

Something must be wrong with either the ML predictions, the remapping, or the IO. To test if it is the ML I started a loocv (here: /home/surface3/dgoll/SPINUP_ML/CLEAN_TEST01/SPINacc-main_COMPRESSEDall_5.5k and /home/surface3/dgoll/SPINUP_ML/CLEAN_TEST01/SPINacc-main_REF01_LOOCV )

The applied equilibrium criteria is linear trend over 100 years < 1 g m-2 yr-1. I used the script uploaded to gitbhub (AuxiliaryTools).

All simulations I performed and their specfications are listed here: https://docs.google.com/spreadsheets/d/1zSS9X6TKbfWcKQufJAxA5tjaHEnulKXdAQYwb4aypic/edit?usp=sharing You will find the paths to the output and configuration (and more info).

[dsgoll123]

The plots show total land C in RERUN05 (x-axis) vs RERUN06 (y-axis) for all 15 PFTs from 1st year of re-run simulation.

soil carbon

vegetation carbon

[dsgoll123]

The issues has been traced to ORCHIDEE. There is a new issue #30 which replaces this one. I close this issue.