Friday, July 22, 2011

TempLS V2.2 and the June 2011 global average

I'll soon be releasing Ver 2.2 of TempLS. Its main new features are:
  • An ability to model global temperatures with monthly resolution, rather than annual. You can also ask for a selection of months - say, summer.
  • The graphics is internally reformatted so that each plot is generated by a function from a data list, which is available after the run. That means that plots can be tweaked, merged etc.
  • Instead of providing just one data set and inventory, you can provide several, and they will be merged. In practice, at the moment, that means that land sets (eg GHCN2, GHCN3, GSOD) and SST sets (HADSST2, HADSST3, ERSST) can be combined as desired. That will allow comparisons (a future post). It's also easier to mix in, say, USHCN.
  • A new weighting function. Previously there was uniform, 5x5 lat/lon cells, and the adaptive methods ( triangular mesh and Voronoi). The complicated ones are good, with one cell for each station in each month, but time-consuming. So I made a new one, which has latitude bands of near-squares, but added the capability that empty cells would have potential weightings distrubuted among neighboring cells that do have stations (those stations are upweighted). This removes the bias caused by empty cells.

A lot to write about, but for the moment, I'm exploiting the ability of the monthly scheme to report the current month. Fixing bugs kept me from getting ahead of the majors this month, but maybe next. I've added TempLS to the regularly monitored set.

The spatial methods work, too, and I'll show a comparison of June with Gistemp below the jump. But here's the most recent time series plot:

These anomalies are set (for plotting) to a common base period of 1979-2000. Raw numbers from TempLS are on a 1961-90 base. And here is the monthly plot for the last four years.

These plots were done using GHCN v2 and HADSST2. Soon I'll do a post with GHCN v3 and HADSST2/ERSST.

The spatial analysis was done using as series of about 12x12 spherical harmonics. It isn't as detailed as the Gistemp one that follows, but gets the main features right.

And here is the GISS comparison.

In a future plot, I'll set out the least squares theory for the monthly adaption.


  1. Why does it take the Brits so long to update their series? Too many tea breaks?

  2. How large are the differences for different weighting functions?

  3. Different weighting functions - I'm working on a post that compares different datasets - I'll try to cover that too. It should only matter when coverage is getting sparse.