FAMOUS

(back to up to ScienceWorkRobin)

MOSES2.2 in FAMOUS

MOSES2.2 is the newer land scheme used in the UM at version 5 and up. It tiles the surface gridboxes into fractional coverage of Broadleaf Trees, Needleleaf Trees, Shrubs, C3 grass, C4 grass, bare soil, lake, urban and ice (although, if present, ice has to be either all or none of a gridbox). The soil is similar to MOSES1, although at least one difference is an implicit solver for the drainage. MOSES2.2 can also cope with carbon and use the TRIFFID dynamic vegetation model, which is why we want to use it in FAMOUS to include the land in the carbon cycle model.

Although completely rewritten, MOSES2.2 is meant to produce (scientifically) the same results as MOSES2.1, which itself was a “simple” tiling extension of the original MOSES1 scheme used in normal FAMOUS. Several different sources suggest that MOSES2.2 does in fact produce significant differences to 2.1, although no-one seems to know why. In FAMOUS, we use MOSES2.2 with version 5 of the boundary layer scheme, the same as in MOSES1 FAMOUS and as used with MOSES2.1 in other versions of UM4.5; more modern version of the UM use MOSES2.2 with a more complex version of the boundary layer that we don’t think will be suitable for FAMOUS.

Implementing MOSES2.2 in FAMOUS was a heroic task on the part of Annette. The climate it produces is, however, substantially different from MOSES1 FAMOUS. In general, there is a high latitude warm bias in winter, and a land surface warm bias during summer. These issues are still extant, although a version of FAMOUS+MOSES2.2 has been released (XFHCY) for other FAMOUS users to contribute tuning/improvements. This release candidate also includes the top level rayleigh friction that fixes the unphysical top level winds, and free-drifting sea-ice.

The high latitude winter warm bias seems to result from the reduction in northern hemisphere filtering that comes with fixing the top level winds. MOSES1 FAMOUS does have a significant winter cold bias wrt climatology, so the warm bias of MOSES2.2 is really an improvement rather than an error. In fact, JJA warm biases over Europe, Russia and Alaska are the only areas where the MOSES2.2 climate is significantly worse than MOSES1 wrt surface temperature climatologies. I am currently investigating the sensitivity of the summer time warm biases over land to the vegatation distribution.

(the following notes are copied across from the old muddle)


FAMOUS needs a land carbon cycle. The standard land scheme (MOSESI) at UM4.5 can’t do this. MOSESII can. MOSESII has subgridscale “tiles” for land points, that mean we can use a Dynamic Vegetation Model, like TRIFFID. (my sketchy understanding of the situation follows) MOSESII.1 exists as some kind of standard at UM4.5, but the code is tangled, unsupported and difficult to update. And at the very least, we need coastal tiling in there for FAMOUS work. MOSESII.2 is the cleanly rewritten version, used by all the more recent UM versions (although apparently the results weren’t as good as MOSESII.1). Various people have various mods that let them run various bits of MOSESII.2 at UM4.5, but there’s no standard at UM4.5. To complicate the picture, there are different boundary layer schemes related to the MOSESI/II choice. At UM4.5 there are BL versions 3, 5 and 6 (MOSESI), and BL7, where BL7 is BL5 + MOSESII.1. By UM5 and above there are only BLs 6 and 8, where BL6A is a rewritten version of BL6@UM4.5 (MOSESI) and BL6B is the MOSESII.2 version. BL8 is a newer BL, with similar A and B versions. The MOSESII.2@UM4.5 mods appear to give the choice of BL5 and BL6. Only BL6 and 8 have MOSESII.2 coastal tiling for UM5+. :)

Annette has collected MOSESII.2 for UM4.5 mods from people and sorted them into some kind of order. Having got a working HadAM3 with MOSESII.2 and BL5, she’s then implemented the coastal tiling for MOSESII.2@BL5 (as found for BL6@UM5+) and got the coupling to work. That was easier said than done. We decided we’d rather stick with the BL5 for consistency with HadCM3, especially given that BL6/8 depend on choices of different catagories of vertical stability profile that are probably less valid at the resolutions we’re looking at. We’ve then hacked together a restart by stitching a few of our new, FAMOUS specific fields to a reconfigured MOSESII.2 atmos dump and sacrificing some chickens. Having fixed a few issues with (I guess, now inconsistent) land parameter values resulting from the reconfiguration and the chickens, the model appears to run OK. Sane values have been observed after having run the model for 100 years. Eyeballing of run times suggests that the MOSES2 code as it stands runs about 3.5% slower than MOSES1 (about 60–90 seconds a year for FAMOUS, which adds up to about an extra day of runtime for a simulation of 1000 years).

Looking at data from a longer run, I’m not sure where I got the 0.3 cooling from. I still see the same warming over land, but it implies a global average warming of about 1 degree. There’s also a further reduction in NH sea-ice, and coverage fractions in the high north are starting to look rather low - this may need retuning. Annoyingly, comparisons of MOSES1 and our 2.2 mods with HadAM3 don’t show this warming, although the runs do use very different soil params. Cross comparisons are starting to look difficult.

Following Essery ‘03 (Explicit representation of subgrid heterogeneity in a GCM land surface scheme, J Hydr. 530–545) which compares MOSES1 and 2(.1?) in HadAM3, I’ve tried randomly overlapping clouds. This is intended to cut some of the incoming solar and reduce the warm bias inland. However, this causes rather a lot of cooling in FAMOUS, everywhere. Maybe someone forgetting to change this switch is the cause of the odd cooling that’s been observed in the non-MetO MOSES2s? On the other hand, maybe our non-standard IGBP soil parameters (see below) are the reason we differ.

It has been noted that diverting excess water through the soil rather than letting it run off the top is very beneficial to the climate over central Asia in MOSESII (I’m aware that’s not a good explanation. The modset claims: “ensure that excess soil moisuture is put into water flux out of bottom of layer, rather than decreasing water flux into top of layer”). The 6.2 mod, runoff_MII.mf77 has thus been included from the start - this is currently built in to are2f406_ctile, but may be separated out again for clarity later. Ditto Tibet_snow.mf77 (which introduces fractional snow coverage for both snow sublimation and snow melt - prevents sudden melting and warming of snow covered areas and subsequent instabilities), which also results in modifed newdecks_ctile and amv1f406_ctile. Pier-Luigi Vidale has also identified long-standing, serious errors in the parameters specified in the standard MOSESII.2 ancils for the land VMC values. Until an official fix is issued, the chickens have hacked in values from the IGBP land-cover dataset (thanks to Tom Osborne), averaged down to the relevant resolution. Initial test runs will thus be conducted with these modifications already included, as well as the climate modifications used as standard in the Summer 07 FAMOUS release.

Page last modified on August 16, 2011, at 11:05 AM by robin