In the test for MICT version, the ML tool gives extremely high and unrealistic pool size of litter and SOC for some grid cells (e.g., Saudi Arab and Canada), mainly on PFT 1 bare soil, but could happen in other PFTs. It was not only happened on the very low PFT coverage, but for dominant PFT (e.g., bare soil in Saudi Arab). Not sure how it happened, but it is better to do some constrains.
A few ways were proposed in the meeting 10th May:
1) force 0 value for bare soil PFT 1
2) constrain the ML resulted value to e.g., 1.5 folds of the global maximum (for each pool). But to be check the super high value should be 0 or a maximum value.
In the test for MICT version, the ML tool gives extremely high and unrealistic pool size of litter and SOC for some grid cells (e.g., Saudi Arab and Canada), mainly on PFT 1 bare soil, but could happen in other PFTs. It was not only happened on the very low PFT coverage, but for dominant PFT (e.g., bare soil in Saudi Arab). Not sure how it happened, but it is better to do some constrains. A few ways were proposed in the meeting 10th May: 1) force 0 value for bare soil PFT 1 2) constrain the ML resulted value to e.g., 1.5 folds of the global maximum (for each pool). But to be check the super high value should be 0 or a maximum value.