Most of the BCCAQ2 datasets have the time.units attribute days since 1950-01-01 and the value of the time variable start at 0 and increment by ones.
The HadGEM2 datasets have time.units of days since 1859-12-01 and start counting upward at 32430.
This happened because the PDP-time normalizing script explicitly expects the last date a dataset covers to be January 1 of the final year, and - as an error check - doesn't normalize time values any file where it calculates this is not the case. However, the HadGEM2 dataset ends on December 30 of the final year, which has been an issue in other contexts as well.
At the time I was processing the BCCAQ2 data, I elected to just leave the HadGEM2 data unmodified, because I did not know that the HadGEM2 datasets were slightly shorter than the other modeled datasets, and I was worried about introducing an error when I tried to make them match the others.
However, the PDP's interface doesn't handle time values that do not start at 0 transparently, and it is confusing users, who reasonably expect all datasets in a group to behave similarly. The HadGEM2 files should be normalized.
Most of the BCCAQ2 datasets have the
time.unitsattributedays since 1950-01-01and the value of the time variable start at 0 and increment by ones.The HadGEM2 datasets have
time.unitsofdays since 1859-12-01and start counting upward at 32430.This happened because the PDP-time normalizing script explicitly expects the last date a dataset covers to be January 1 of the final year, and - as an error check - doesn't normalize time values any file where it calculates this is not the case. However, the HadGEM2 dataset ends on December 30 of the final year, which has been an issue in other contexts as well.
At the time I was processing the BCCAQ2 data, I elected to just leave the HadGEM2 data unmodified, because I did not know that the HadGEM2 datasets were slightly shorter than the other modeled datasets, and I was worried about introducing an error when I tried to make them match the others.
However, the PDP's interface doesn't handle time values that do not start at 0 transparently, and it is confusing users, who reasonably expect all datasets in a group to behave similarly. The HadGEM2 files should be normalized.