Apr 15 2008
AtmozTemp: Just the Mean of the Other Temps
Yesterday I posted a file online that made it easy to compare the temperature anomalies from GISS, Hadley, RSS, and UAH. I’ve taken that one-half step further today and took the mean and standard deviation of the 4 temperatures for each month, and included them in the climate metrics data file. This allows for easy plotting and the use of error bars.
|
(Click to make really big)
Note that due to the large horizontal size of that image, it appears that the temperature has not risen much. However, the trend is almost exactly the same as that when computed only on the GISTemp and Hadley data.
The error bars give an estimate of the uncertainty in the global mean temperature. If all four global temperature metrics were the same, there would be no error bars on this plot. The error bars do not factor in errors in the temperatures in each of the four temperature products. This would increase the size of the error bars.
|
By plotting the standard deviation, there are times when the 4 metrics don’t agree that well. The largest divergence is around 1998 during the El Nino event. This is because the GISS temperatures during this period do not have the same spike in the data that the other 3 time series have.
Similarly, recently there has been an increase in the standard deviation due to the ‘divergence’ of the satellite data from the surface temperature-based data. The satellite data (RSS, UAH) still show temperature anomalies around zero, while both of surface products (GISS, Hadley) have rebounded to near previous levels. As with the other ‘divergences’, this shouldn’t last too long or have a large impact on calculated long-term trends.
(No Ratings Yet)
Loading ...Related Posts:
7 Responses to “AtmozTemp: Just the Mean of the Other Temps”
To reduce spam, comments are automatically closed 30 days after the last comment. If you would like to comment on any closed thread, please use the contact form at the top of this page.
Nice work. It would be interesting if the standard deviation for this La Nina would be similar to the last El Nino. We’ll see over the coming season.
Very nice, been playing with your methods myself. Also included the NCDC numbers normalized in the same manner.
Here’s a very interesting exercise. Calculate the average anomaly by month for just the sat-based numbers, and the average for just the land-sea based numbers, and then take a look at the variance between those two averages. [Hint: its not as big as I expected]
Seems to me this is just the variance between the series and not a measure of the noise or variability.
Eli, you are sort of correct. When determining errors, the variance between different methods of measurement is assumed to be the error OF the methods of measurement (assuming they have good agreement with one another). However, all errors and noise are cumulative, so this is only PART of the total error. Other factors would indeed add additional error for the purposes of this exercise.
Just curious - these global average temperature series are the mean of mean of minimum and maximum temps, as an anomaly from mean of the last 30 years at that station, at measurement points world-wide, right? Weighted for geographic coverage, but essentially the mean.
If so, isn’t that a misleading method? Wouldn’t the median of the anomalies be a better measure of global warming? If not, why not?
Thanks in advance
Hello Atmoz!
Using the EBCF method for you graph and ignoring the 1998 El Nino Bump, I get a saw-tooth plot upto 1995.8 and then a step function. There is a mirror plane at 1985.
Got any data before 1980 to see if there is a peak ca 1976?
BTW: EBCF=Eye-Ball Curve Fitting. Would a fancy computer curve- fitting program pick out the saw-tooth wave form and the mirro plane?
If I remember correctly from one of my undergrad stats courses, variance increases with the mean of a measurement, so coefficients of variation are compared instead. For example, diversity in male heights wouldn’t be compared to diversity in female heights in a class via standard deviation — the standard deviations of each would be divided by their means. You can’t just subtract the difference to standardize the measurement between the two sexes and then look at the standard deviations.
Your case is an interesting variation on the simple example I just gave. You’re comparing diversity over time, but there is a trend with time. Shouldn’t you adjust for the trend by division rather than by simply subtracting the trend out?
[Reply: I'm not subtracting the trend (slope) out. I'm substracting an arbitrary bias (offset, y-intercept).]