Data Comparisons

HadCRUT: Down slightly from January.

13 Comments

HadCRUT NH+SH posted their temperatures anomaly for February:

  1. The anomaly was 0.460 C, down from 0.495 C in January. ( Note that the January temperature was revise from 0.470 to 0.495.)
  2. According to my tally, this was the 7th warmest February 4 in the HadCRUT NH&SH record. February anomalies are circled in the graph above; as you can see most of those warm Februaries occurred this century.
  3. If computed using data since Jan 2001, both the simple trend based on time only and the trend computed using a multiple regression involving both MEI and time remain negative. The nominal trend of 0.2C/decade remains outside the ±95% confidence intervals for the MEI corrected trend; these confidence intervals computed based the assumption that the residuals for the observations are ARMA(1,1).

Graphs requested in comments

Figure 2: Carrick requested a graph of the residuals for the OLS fit to time and the multiple regression to (time, mei).
Figure 2: Carrick requested a graph of the residuals for the OLS fit to time and the multiple regression to (time, mei).

13 thoughts on “HadCRUT: Down slightly from January.”

  3. Andrew_KY: Rutherford must have done experiments in a lab. 🙂

    When doing a lab experiment, it is wise to apply statistical reasoning before designing the experiment. One of the most important things to understand is that you should know what level of effect you want to detect if it exists, and then take enough data that you would have a very good chance of detecting that effect at a statistically significant level.

    So, for example, way back in 1960, if you wanted to figure out how long it would take to detected 0.1 C/decade of warming if it existed, you estimate the amount of noise that would appear in future data, and do an analysis to figure out the statistical power of a test as a function of the time you accumulate trend data. It’s computations like these that result in rules of thumb like “wait 30 years”.

    Of course, you could do this test, and if it turned out that warming was 0.1 C/dec you’d have an answer before 30 years. Or if noise was lower because no volcanoes erupted after you started your experiment, while your noise estimate assumed they might, you could detect a trend of 0.1 C/decade in less time.

    The fact that you thought it would take you 30 years to have a good chance of detecting the trend wouldn’t mean you couldn’t report that you’d detected a statistically significant when you did detect it.

  6. “If your experiment needs statistics, you ought to have done a better experiment.”

    So are you advocating this position, Andrew, or just being “yourself”?

  7. Lucia:

    Do you want a temporal plot? A histogram? I can do either.

    I would prefer the temporal plot.

  8. Carrick–
    This is since 2001.

    I can easily create since 1980. Other start years require a bit more tweaking.

    This graph [observation – (m_time*time + m_mei*mei + constant) ]

  9. Thanks, Lucia. If it isn’t being too much of a bug, can you show the crowd what the residuals look like if you don’t correct for MEI?

  10. Here:
    Figure 2: Carrick requested a graph of the residuals for the OLS fit to time and the multiple regression to (time, mei).

    I also adjusted the legend to include information about the estimate of the standard error of the “noise”. If you read the values, you’ll see the “noise” is lower if we include MEI in the fit. The main reason the uncertainty intervals drop is that the autocorrelations drop. (For longer periods in the fit, MEI has a noticeable effect on the rms of the residuals too. I could do a correction by “freezing” the factor multiplying MEI, but I don’t. )

  11. So are you advocating this position, Andrew, or just being “yourself”?

    Carrick,

    I am always ‘myself’! Who else would I be? ‘squash consumer’? ‘pea resister’? ‘donut monger’? 😉

    I am an advocate of the position that a statistical analysis is only valid if you know your data. Ahem. (cough)

    Andrew

  12. It just gets worse and worse for statistics and climate. As a biologist the papers I published usually had to have at least p < 0.001 with ONE variable cause=effect.
    http://wattsupwiththat.com/2010/03/20/science%E2%80%99s-dirtiest-secret-the-%E2%80%9Cscientific-method%E2%80%9D-of-testing-hypotheses-by-statistical-analysis-stands-on-a-flimsy-foundation/
    Lucia sorry to be continually spoiling this statistics fest! You give however a very fair balanced view of state of the art statistics in Climate change hahaha

  13. Unfortunately Climate is like measuring your heart rate.. it changes occasionally.. eventually it will become extremely boring and NO ONE will be interested. So I predict WUWT, CA, RC, and unfortunately this site. ect will vanish into “ol soldiers never die they only fade away” scenario LOL

Comments are closed.