(back to up to ScienceWorkRobin)
Filtering issues
original MOSES1 versions of FAMOUS had always had a winter high latitude cold bias (worse in the northern hemisphere), and, it transpired, an elusive stability issue in colder climate simulations. This stability issue led Bristol to change the model timestepping (see Current Work Timestepping), which I still don’t think is a great idea.
I did my glacial runs with workarounds, but they didn’t always work around the problem for everyone, so I had a a serious look at the problem around Christmas 2010. The UM uses a grid-point model in the atmosphere, which requires higher wave-number terms to be cut out at high latitudes to stop the timestepping stability requirements from becoming too restrictive as the gridbox size shrinks near the poles. This is done by fourier filtering the zonal wind fields, and there are routines that interactively determine how much filtering needs to be done on a timestep-by-timestep basis to keep the model stable. The proximate cause of the cold climate crashes was those filtering routines turning themselves off, leading to massive CFL violations at high latitudes and the rapid development of enormous instabilities.
The filtering routines have a sanity check that turns them off if they determine that an entire hemisphere needs to be filtered - really, the filtering should only operate at very high latitudes. It turns out that in FAMOUS, the low resolution seems to be stopping the gravity wave drag parameterisation from functioning correctly, which means that the top-level winds assume unrealistically high velocities. In cold climates, the meridional temperature gradients get steeper, the vertical wind shear gets greater and this problem gets worse - the unrealistic top-level tropical winds can get fast enough that variability on them can trip the sanity checks on the high latitude filtering routines and take the whole model down.
The simple fix to improve the stability of the model is thus just to cap the tropical top-level winds below this critical level - this (stratcap.mod) works well, and you can then run a snowball Earth simulation with previous crash-prone versions of FAMOUS without trouble. Manoj Joshi@Reading suggested a much better fix though, which is to put simple Rayleigh friction in the top levels of the model to do what the GWD should have done, and take the excess energy out of the wind field. This (rayleigh_fric.mod) drastically improves the vertical structure of the FAMOUS atmosphere wrt climatology (and thus allows more realistic ozone concentrations to be used), although a global average friction doesn’t reproduce the hemispherically asymmetric jets - a retuned GWD parameterisation should maybe be retained for this reason - this is being looked at.
The Rayleigh friction has a significant knock on effect on the surface climate. Even when the top-level winds weren’t crashing the model, they appeared to be forcing FAMOUS to filter much of the mid-latitude variability in the winter hemisphere (see figure), which reduced the heat transport. Fixing the top-level in this way winds thus improves FAMOUS’s winter high latitude cold bias and overall seasonality significantly, meaning that we no longer have to rely on the extremely low sea-ice albedoes to improve the annual mean northern hemisphere temps. This does, however, require a general retuning of the model, which is being conducted by Jonny Williams@ Bristol
New Standard Version XFXWB has the stratcap.mod in, and release candidate XFHCY is being tuned with rayleigh_fric.mod.
annual avg. zonal mean zonal winds for (top left) HadCM3, (top right) standard FAMOUS, (bottom) FAMOUS with Rayleigh friction included
Attach:filtering.png Δ
maximum allowed wavenumber for northern hemisphere winter in different versions of FAMOUS
