During a code review of LPJmL it has come to the modelers attention that the "rootmoist" output variable was implemented incorrectly for ISIMIP2a and ISIMIP2b. The values as reported are far too low (by a factor of roughly 2.5 as a global average). The error is not constant across space or time so the only way to fix this would be to rerun all ISIMIP2a and ISIMIP2b simulations (which is not possible). Is is an output issue only, not a faulty process inside LPJmL. So the error should have no effect on any other output variable.
The variable should not be analysed anymore. There's no way of knowing how much the values are underestimated in each cell. Plus, the error is not constant over time, so it's not safe to look at trends instead of absolute values either.
The data has been retracted from DKRZ, PIK-Cluster and the public ESGF archive on December 7th, 2020.