Other recent reconstructions have concentrated on land stations too, perhaps because, as Zeke says, the context was the examination of claims that drift in selection of stations creates biases.
The least squares method used here makes the incorporation of ocean data easier, because there is no intermediate gridding step. So I tried it, I downloaded HadSST2 data, and added a set of 2592 (36x72) synthetic stations - one at the centre of each grid cell. HAD uses the same 5x5 grid as I use for weighting for station density. So where the station sits in an ocean-only cell, it is fully weighted as the sole representative of that cell, and where ocean overlaps land, the new station is weighted according to the number of land stations in the same cell. If the cell is land only, there is no new data, so that cell does not add to the analysis.
New codeThis all required some new coding. That's tied up with some other changes, so I'll post ver 1.3 shortly. The change was mostly in the preprocessor, which rearranges the Had data to the GHCN format and creates the new station file.
A summary of this series of posts is here.
TrendsHere is the table of trends from my analysis:
Update: I hope to be routinely quoting standard errors for trends. At present these are based on OLS - no correction for correlation of residuals. So they are smaller than the AR4 values below.
I chose the time intervals for comparison with Table 3.3 from the AR4:
|Temperature Trend (oC per decade)|
|CRU/UKMO (Brohan et al., 2006)||0.047 ± 0.013||0.075 ± 0.023||0.234 ± 0.070|
|NCDC (Smith and Reynolds, 2005)||0.063 ± 0.022||0.245 ± 0.062|
|CRU/UKMO (Brohan et al., 2006)||0.038 ± 0.014||0.068 ± 0.017||0.092 ± 0.038|
|NCDC (Smith and Reynolds, 2005)||0.066 ± 0.009||0.096 ± 0.038|
|CRU/UKMO (Brohan et al., 2006)||0.042 ± 0.012||0.071 ± 0.017||0.163 ± 0.046|
|NCDC (Smith and Reynolds, 2005)||0.064 ± 0.016||0.174 ± 0.051|
|GISS (Hansen et al., 2001)||0.060 ± 0.014||0.170 ± 0.047|
Here my plots are compared with the published gridded data, taken from WFT. The smoothing uses the 13-point filter from AR4 Chap 3
The anomaly base periods have been adjusted to match in each case to the published data.