Keith Lindsay (Aug 19 2020 at 15:22): Keith Lindsay (Aug 19 2020 at 15:22):A couple of notebooks (003, 004) have been added to the HiRes-CESM-analysis repo that have 'quick-look' views of a handful of variables.
MARBL tracers in 004 were initialized from a x1 companion that was run for 2 JRA cycles. In 003, macronutrients and O2 were initialized from WOA13, DIC and ALK were initialized from GLODAPv2, and the remaining tracers were initialized from the x1 companion.
One observation from the notebooks is that the histograms for O2_ZMIN indicate has some very large magnitude negative O2 values in 003. I dug into this and have discovered what I think is the culprit. The bad news is that the regridding of WOA13 fields to the tx0.1v3 grid generated all zeros in the deepest tx0.1v3 layer. The problem is that all of the deepest model layer is deeper than the deepest WOA13 layer, and the vertical remapping function that I wrote generated zeros in this case. This problem is not present for GLODAPv2 interpolated fields. Unfortunately, I missed these bad values when inspecting the generated IC files :sad:. I suspect that the zero values lead to very large tracer gradients, which leads to numerical advection errors which generated the large magnitude negative O2 values. The problem is present for all WOA13 interpolated fields: O2, PO4, NO3, SiO3. All of these tracers have spurious values near where KMT reaches 62. The good news is that the spurious values seem to mix out within a handful of months.
While these values are definitely problematic in the first few months of 003 and are a reason to prefer 004 over 003, I'm not sure if the problem by itself is a show-stopper for 003. Given that 003 has run for a year, and our throughput, I do NOT advocate for restarting 003 with corrected IC.
Matt Long (Aug 19 2020 at 15:28):Thanks @Keith Lindsay; I agree with your ultimate conclusion.
Last updated: May 16 2025 at 17:14 UTC