Thursday, September 11, 2014

Extended Trend Viewer

I have been maintaining regularly an active trend viewer. A reduced image looks like this:

The triangle shows with color shading the trend from any starting year to end year, in a range that you can choose (from 1999, 1989, 1960 or 1900 to now). There are settings that show just trend, trend with significance masked, or CI's (upper or lower) or t-values. And there are now 15 datasets - monthly temperature series.

Each triangle has an accompanying plot of the time series. The active aspect is that you can choose a time range, and numerical information about the trend will be shown, and colored markers will show the trend line on the graph. You can choose either by clicking in the triangle, or using controls in the graph.

I have now added four new data sets. They are from Cowtan and Way, BEST Land/Ocean, NOAA SST, and a new TempLS set. I'll describe each in detail, and give links to the sources.

Cowtan and Way

Kevin Cowtan's home page is here. They took HADCRUT 4, which uses a grid but omits cells that have no data, and used different schemes for estimating the empty regions. The motivation is that recently there has been warming in the Arctic which HADCRUT was not detecting properly. They used kriging interpolation, and hybrid schemes making use of satellite data. I'm showing the kriging results here.

BEST Land/Ocean

Berkeley Earth started with land only, and that remains their prime product. They only update their land/ocean data annually. They use HADSST3 for ocean, and as with all land/ocean indices, this has a dominant effect.


I'd really like to be able to show the OI V2 SST, which is kept very current. But although it's easy to download spatial data, for some reason the global index can only be got interactively. So I'm showing the regular product which comes out late in the month.


TempLS is my least squares global temperature index  program. My recent post gives some background, and a comparison. It is basically an area weighted regression, and I have been using a traditional scheme whereby the weight is assigned according to each stations share of its gridcell area. This gives similar performance to actual gridded schemes, and that recent post was to show how well it follows NOAA.

But this has the same faults that Cowtan and Way were trying to fix. No weight is assigned to allow for empty cell area, of which there is much in the Arctic. TempLS offers various weighting scheme, one of which is essentially finite element integration on an irregular triangular mesh. This has no problems with empty cells, and no problem with cells getting small at the poles. I have felt that Cowtan and Way's kriging was overkill, and that any interpolation (here linear) would do much the same. This is an opportunity to test. I'll write more on this soon.


Here is a table of links to the data files.

HadCRUTHadCRUT 4 land/sea temp anomaly
GISSloGISS land/sea temp anomaly
NOAAloNOAA land/sea temp anomaly
UAH5.6UAH lower trop anomaly
RSS-MSURSS-MSU Lower trop anomaly
TempLSgridTempLS grid weighting
TempLSmeshTempLS mesh weighting
BESTloBEST Land/Ocean
C&WkrigCowtan/Way Had4 Kriging
GISS TsGISS Ts Met stations temp anomaly
CRUTEMCRUTEM CRU global mean Station anomaly
NOAAlaNOAA land temp anomaly
NOAAsstNOAA sea temp anomaly


  1. Very nice work, Nic. Is TempLS shifted relative to the other figures? I wonder if there's an indexing issue.

  2. Nick

    Well done, this looks really useful. Is it intended for others to use? If so how do we operate it, can you give us an example or two?


    1. Tony,
      Yes, it's certainly intended for all to use. The image here is just a pic. Have you checked the main page? There are three links at the top to earlier posts, which have various degrees of explanation. There is also explanation below the main picture.

      But mainly, it's a matter of experimenting. The big help is that whenever you click on the triangle or the graph controls, the numbers on the left of the table part tell you what is happening, and the dots on the time series graphs show the range and trend.

      So what you can do:
      1. Pick a data set (radio button) eg GISSlo etc
      2. Pick a time range (eg 1960-now)
      3. Pick a mode. These are mainly explained in those earlier posts. The default is trend, but you can grey out the statistically insignificant, or show CI values, or t values. For example with t-values, I've shown 1.96 with dark grey, -1.96 with light. Those are 95% levels, and values on one side are significantly above - reddish (or below, blue) zero, other side not.
      4. Click in the triangle. You'll see the red and blue dots on the graph jump into position. The text should report the same dates. This is the best way of getting used to the triangle. The (imaginary) line joining red and blue dots is the trend line. If you have plain trend, the positive parts will be reddish. Click to see what that looks like in the time series.
      5. The graph controls are the red and blue lines, and the <<>> nudgers. Click on the red line to shift the red dot to that time. Nudge with the red <<>>. If you click on the > neare3st the center, the red dot moves up one month. Next > moves it two, next 4 etc. Same with blue.

      The purple nudger moves both in the same direction, keeping the interval the same. The gold moves them in opposite directions, keeping the centre the same.

    2. Tony,
      I thought of an explanation of the triangle with compass points that might help. The x-axis is end time of the period; y axis is the start. So the E edge has trends for periods ending now. As you move S, you push back the start year. At S edge, the period starts at start of time selected (eg 1999). If move W, push back the end year. Move SW, and the start and end are pushed back equally. Move NW and they move oppositely, so the mid-period stays the same. At the NW edge, the period is 1 year.

      The red nudger is equiv to a N-S move, blue to E-W, purple to NE-SW, and gold to NW-SE.