Some are kept up to date, within a few days, and it is this aspect that interests me. They are easily integrated over space (regular grid, no missing data). I do so with some nervousness, because I don't know why the originating organizations like NCAR don't push this capability. Maybe there is a reason.
It's true that I don't expect an index which will be better than the existing. The reason is their indirectness. They are computing a huge amount of variables over whole atmosphere, using a lot of data, but even so it may be stretched thin. And of course, they don't directly get surface temperature, but the average in the top 100m or so. There are surface effects that they can miss. I noted a warning that Arctic reanalysis, for example, does not deal well with inversions. Still, they are closer to surface than UAH or RSS-MSU.
But the recentness and resolution is a big attraction. I envisage daily averages during each month, and WebGL plots of the daily data. I've been watching the recent Arctic blast in the US, for example.
So I've analysed about 20 years of output (NCEP/NCAR) as an index. The data gets less reliable as you go back. Some goes back to the start of space data; some to about 1950. But for basically current work, I just need a long enough average to compute anomalies.
So I'll show plots comparing this new index with the others over months and years. It looks good. Then I'll show some current data. In a coming post, I'll post the surface shaded plots. And I'll probably automate and add it to the current data page.
Update: It's on the data page here, along with daily WebGL plots.
More on reanalysisReanalysis projects flourished in the 1990's. They are basically an outgrowth of numerical weather forecasting, and the chief suppliers are NOAA/NCEP/NCAR and ECMWF. There is a good overview site here. There is a survey paper here (free) and a more recent one (NCEP CFS) here.
I've been fosussing on NCEP/NCAR because
- They are kept up to date
- They are freely available as ftp downloadable files
- I can download surface temperature without associated variables
- It's in NCDF format
There are two surface temperature datasets, in similar layout:
Then I tried sig995. That's a reference to the pressure level (I think), but it's also labelled surface. It goes back to 1948, and seems to be generally more recent. So that is the one I'm describing here.
Both sets are on a 2.5° grid (144x73) and offer daily averages. Of course, for the whole globe at daily resolution, it's not that easy to define which day you mean. There will be a cut somewhere. Anywhere, I'm just following their definition. sig995 has switched to NETCDF4; I use the R package ncdf4 to unpack. I integrate with cosine weighting. It's not simple cosine; the nodes are not the centers of the grid cells. In effect, I use cos latitude with trapezoidal integration.
ResultsSo here are the plots of the monthly data, shown in the style of the latest data page with common anomaly base 1981-2010. The NCEP index is in black. I'm using 1994-2013 as the anomaly base for NCEP, so I have to match it to the average of the other data (not zero) in this period. You'll see that it runs a bit warmer - I wouldn't make too much of that.
|NCEP/NCAR with major temperature indices - last 5 months |
|NCEP/NCAR with major temperature indices - last 4 years |
Here is an interactive user-scalable graph. You can drag it with the mouse horizontally or vertically. If you drag up-down to the left of the vertical axis, you will change the vertical scaling (zoom). Likewise below the horizontal axis. So you can see how NCEP fares over the whole period.
Recent months and daysHere is a table of months. This is now in the native anomaly bases. NCEP/NCAR looks low because it's base is recent, even hiatic.
The mean for the first 13 days of November was 0.173°C. That's down a lot on October, which was 0.281°C. I think the reason is the recent North American freeze, which was at its height on 13th. You can see the effect in the daily temperatures:
Anyway, we'll see what coming days bring.
Update (following a comment of MMM). Below is a graph showing trends in the style of these posts - ie trend from the x-axis date to present, for various indices. I'll produce another post (this graph is mostly it) in the series when the NOAA result comes out. About the only "pause" dataset now, apart from MSU-RSS, is a brief dip by GISS in 2005. And now, also, NCEP/NCAR. However, the main thing for this post is that NCEP-NCAR drifts away in the positive direction pre 2000. This could be that it captures Arctic warming better, or just that trends are not reliable as you go back.