My last two posts have been about the surface stations, and this one follows in its footsteps. This is also being discussed in two threads at ClimateAudit. Most of the discussion seems to center on how the temperatures are corrected for several known biases. The image above illustrates the importance of correcting for known biases.

This photograph is obviously of a road with several hairpin bends. There was a long exposure time, such that when the automobiles were driving towards the camera the lights appeared white, and when they were driving away from the camera they appeared red. Suppose you knew this a priori, and were interested in the reflective properties of the road in the visible portion of the spectrum. Without accounting for the artificial lights, the characteristics of the road are difficult to discern.

The temperature record also has known biases. In 1991, Quayle, et al. wrote a paper titled Effects of Recent Thermometer Changes in the Cooperative Station Network. They looked at the statistics of how much the temperature changed due to the systematic change from one thermometer to another. They concluded that the switch biased both the maximum and minimum recorded temperatures. The yearly average maximum temperature was 0.40 Celcius lower than the temperature measured with the previous thermometer. The yearly average minimum temperature was 0.28 Celcius higher, and the yearly average mean temperature was 0.06 Celcius lower.

Temperature Trend Distribution

If the date of the switch between the CRS and MMST thermometers is known, a simple change can be made to either the temperatures before or after the switch. I don’t really want to redo the analysis in their paper, so I’ll look at some other things that I thought were interesting. One has to do with changes in the distribution of trends in the USHCN after several of the adjustments, and the other is looking at theoretical changes in trends due to the thermometer switch.

This figure shows the changes in the distribution of temperature trends in the USHCN at several steps in the adjustment process. The raw data (which coincidently is already processed for bad points) is shown in black. The blue curve represents the distribution of trends after the time of observation (TOBS) bias is corrected. And the red curve represents the distribution after the FILNET stage of the adjustment process. (The legend says MMTS; this was a mistake.)

From the USHCN documentation, “[t]he temperature data are adjusted for the time-of-observation bias (Karl, et al. 1986), which occurs when observing times are changed from midnight to some time earlier in the day.” There is a systematic bias in the TOBS correction that causes the temperatures trends to be higher after the correction is applied. However, when applied to all station is relatively small. This could be because the bias is small, or the time-of-observation bias could be either positive or negative for each station, but averaged over the entire US these errors tend to average themselves out.

MMTS, SHAP, and FILNET Corrections

The corrections applied after TOBS are more interesting in that there is a larger corrections. The difference between the red curve and the blue curve actually represents three different corrections. The first has to do with the change in thermometers from CRS to MMTS. This adjustment is known as MMTS. The second adjustment is related to the station history (SHAP), and the third adjustment fills in missing data (FILNET). Because the data after the MMTS and SHAP adjustments are not readily available, this analysis cannot say which adjustments have the most effect in the FILNET data.

The MMTS adjustment is known to cause a rise in the average temperature trend in the US. The FILNET adjustment should not have a large effect on the temperature trend distribution since it is just filling in missing data. The SHAP adjustment likely adjusts the temperature for changes in station latitude and altitude. SHAP may include other changes, but I’m not certain (I haven’t read the paper). Therefore, I don’t know which is causing the majority of the changes seen between the FILNET and the TOBS adjustments.

This figure shows the theoretical bias correction needed for several scenarios. First, I assumed that the temperature record consisted of two different levels, with a step at some “hinge” point. For the purposes of this analysis, I chose three “hinges” corresponding to the years 1984, 1990, and 1996. Because different stations have different record lengths, this will affect the trend bias. For instance if a temperature record started in 1900, the bias in the trend would be less than a station that had its record start in 1950.

The plot shows the bias correction needed for each station as a function of the station start date (the first record in the temperature data). As can be clearly seen, the longer the record, the less this bias affects the temperature trend. There also exists a point where if the temperature record is too short, this bias is not as significant. Imagine a station that has a start time after the “hinge” point; there will be no bias.

References

Karl, T. R., C. N. Williams, Jr., P. J. Young, and W. M. Wendland. 1986. A model to estimate the time of observation bias associated with monthly mean maximum, minimum and mean temperatures for the United States. J. Clim. Appl. Meteor. 25:145-60.

Quayle, R.G., D.R. Easterling, T.R. Karl, and P.Y. Hughes, 1991: Effects of Recent Thermometer Changes in the Cooperative Station Network. Bull. Amer. Meteor. Soc., 72, 1718–1723.

(No Ratings Yet)

 Loading ...

Related Posts:

Trackback URI |