Tuesday, September 15, 2015

Temperature change July to August mapped in GISS and TempLS

I was curious about the relatively small rise in the GISS August global land/ocean anomaly, relative to July. So I drew maps comparing gridded GISS with TempLS mesh for those months. The maps are based on a spherical harmonics fit to both data sets, with L=10 (121 functions). This puts them on exactly the same basis.

I included TempLS with GHCN adjusted data as well. This gives the opportunity to also show the differences between adjusted and unadjusted, and to plot the actual differences.

Nothing looked unreasonable. The big Jul-Aug difference between GISS and TempLS was in the treatment of Arctic (especially) and Antarctic. There is actually very little effect of GHCN adjustment. The month-month changes are themselves quite interesting; the patterns are quite persistent.

The GISS data I used was the 1200 km interpolated, (download gistemp1200_ERSSTv4). The TempLS runs used ERSST4 and GHCN from here. I fitted each set with spherical harmonics up to L=10 (121 functions). This gave enough resolution to make the comparisons, without risking artefacts.

In this tableau, for each set the July 2015 result is on the left, and August on the right. All data uses base years 1951-80, with conversion numbers from here.

Giss shows the Arctic going from part-warm to moderate everywhere. TempLS, however, has the Arctic warm in each month. In both datasets, East Antarctica is cold in both months, but generally a larger area in August. TempLS shows greater cooling in the East US. Both show the August warmth in S Americe, which is in fact the most noticeable change over the month.

You see a slight variation in SST with adjustment, which may seem odd, as only land is adjusted. But the spherical harmonic approx is global, so the effects can spread. The difference is small.

The next plot shows the actual differences Aug-July.

This emphasises the polar changes in GISS, with much less Antarctic change in TempLS, and almost none in the Arctic. A diffreence is not surprising, as GISS uses extra data in the south, and their interpolation scheme in the Arctic is also different. The other main differenve is in the US, where TempLS emphasises both the Eastern cooling and Western warming.

The next tableau shows the datasets differences, just for August.

Very little effect of GHCN adjustment, except for some in Antarctic. This is expected, because present time is the reference point for adjustment. The differences will be due to changes in the offsets. Between GISS and TempLS, you see another view of the polar differences, and otherwise just odd patches, mostly over land.


  1. Thanks, that's exactly the set of comparisons we need.

    If I remember correctly, tempLS throws in all of the land and ocean stations together and calculates the results on a single globe?

    Whereas GISTEMP calculated the extrapolated land temperatures separately and then blends, allowing land temperatures to full in missing regions of ocean but not the other east rewound. The physical justification being that the missing ocean regions are sea ice (at least for recent decades) and air temperatures over sea ice behave more like air temperatures over land, lacking the strong coupling to the ocean heat sink.

    I've looked at this in models, in the satellite skin temperature data and in the IABP data, and I agree with GISTEMP on the physics. However that leaves me with a problem explaining the difference between GISS and the reanalysis.

    1. Thanks, Kevin. I'm developing an ambition to do a more comprehensive set. Spherical Harmonics make a very flexible basis, because for each map you go through a stage where the info is reduced to a set of (here) 121 coefficients, which can then be rendered back on a common base - no issues of different infill etc. Linear algebra on the coefficients before rendering does the differencing.

      I worried about sea ice with the mesh method. It works out fairly well in the Arctic, because the geometry tends naturally to use land data over ice. I remove all data (NA) where they put T=-1.8, which makes no climate sense. The south is trickier, as sea nodes become more likely to influence ice regions. However, at least some degree of influence is justified.

  2. Actually, that might still be the explanation. The sea-ice as land similarity is very clear in winter when the ice is thick and covered in snow, which is an excellent insulator. However it is less good in summer - see the Rigor IABP paper.

    To get the long term tends right, the ice-as-land model is important because all of the Arctic warning has occurred in the winter. Ice melt holds the summer temperatures almost constant.

    However, to get monthly variations in summer right, we may need a hybrid model in which blending varies by season or absolute temperature. I would develop it by training against the reanalyses, but how do we validate? Against IABP?

    Very interesting - there might be some good new science in this oddity.