Sunday, April 14, 2013
TempLS is a program I wrote for getting global average surface temperature anomalies. I've been using it every month to calculate an index from GHCN/ERSST data, which I compare with GISS when it comes out. It differs from those usually used in that it associates offsets with stations and does a least squares fit, so avoids fittinga fixed anomaly period into station data. The math basis is described here. And here is a somewhat dated overview.
I've been in discussion at Lucia's (see also recent threads here), and it seemed to me that a method she was proposing for anomaly reconstruction for paleo could be implemented in TempLS. Although TempLS is oriented toward monthly data with seasons, it can fix on just one month of the year, so all I had to do was get data in the right input format. As with my earlier recons, I did not attempt the Monte Carlo variation of dating and calibration. I used the published dating.
It works quite well - I'll show early results below. It avoids the dimple. I use the same 5x5 weighting scheme. However TempLS has extensive facilities for weighting by different methods, such as unstructured triangular meshes, which are well suited for sparse sets like these.
The Marcott et al data comes from the Supplementary Materials spreadsheet. In these plots I have shifted all curves to a 4500 to 5500 BP anomaly base for plotting. However, I am using the Marcott et al CI's as from the spreadsheet, and the TempLS calc did not use this period, so I did not modify the CI's to reflect this shift. I smoothed all data using a century moving average. This cuts 40 years off each end. The reason I smoothed the Marcott data is that it is given to only two significant figures, so is whiskery when plotted unsmoothed.
Here is the plot over the full period with the TempLS and Marcott reconstruction with shaded CI's. One thing to note is that the TempLS solution is dimple-free. The overlap is fairly good - TempLS strays a bit at the ends. I'm checking for possible reasons.
And here is the plot with just the CI's for comparison. TempLS is mostly lower, but again does not include error due to calibration and dating uncertainty.
Here are versions restricted to the last two millenia.
Posted by Nick Stokes at 10:24 PM