Lon Glazner, a fellow blogger and local electronics engineer made some comments about my post on the NASA/CSU study on California temperatures. Well that got me started…so below are Lon’s comments and my reply along with a fun technical challenge. For those of you that read this blog, but disagree with my views, I invite you to read this carefully.
You make a number of good points. Particularly in the fact that the writers may have applied changes in urban temperature measurements over large regions for graphical impact.
As someone who has designed and built electronic temperature sensors I have certain concerns about the data itself.
Unless temperature sensors are regularly calibrated I think it is unreasonable to expect accuracy of greater than a couple of degrees.
Even some that are calibrated may not have good accuracy. The LM34 which is a commonly used semiconductor for measuring temperature is +/-2 degrees F. This is pretty typical of analog or digital semconductor sensors. The temperature error for this part is also non-linear, and so it’s not a simple offset that you have to account for during data collection. Furthermore, there are lots of additional errors that can creep into a temperature measuring device beyond the sensor itself.
One could argue that numerical analysis done on data points would tease out errors. But if a scientist doesn’t know the exact accuracy of a temperature sensor then they couldn’t account for errors in their system.
Some of the temperature sensing stations may be very accurate and regularly calibrated. But maybe they’re not?
I have a hard time trusting that the data is accurate to the level of identifying 1 or 2 degree changes over decades. This is especially true since the techniques of making these measurements have changes over that time frame.
Lon, thank you for the comments. FINALLY somebody who understands the kind of biases that creep into temperature measurements!
I’m innately familiar with National Semi’s LM34 and it’s accuracy problems. One of my early jobs at my university as a research assistant was to create remote electronic weather stations. I soon learned how inaccurate many electronic devices can be in temperature measurement.
The problem with the National Weather Service temperature data sets (and world data sets too) is that they are full of biases and errors that I’m not sure have been accurately accounted for. People such as Jim Price, from CSUC who is on the IPCC say they have been, yet nobody has shown me any hard evidence of such. I’d be a lot less skeptical if I could see how the IPCC accounted for temperature measurement biases. But they won’t share.
Some people that I try to explain this to accuse me of splitting hairs. But these bias problems in temperature measurement are quite real.
What works against my arguments about the difficulty in getting accurate temperature records is the everyday simplicity of temperature and its common measurement. We live by temperature, we have it reported constantly, we all have thermometers at home, we measure our childrens fevers with thermometers, we barbeque with thermometers.
Measuring temperature is easy right? You just stick the thermometer in whatever gas, liquid, or solid you want to measure the temperature of and voila’ there it is. People tend to think of thermometers as perfect devices. Some very expensive calibrated thermometers, are close to perfect, especially when taking measurements in a closed system, like a fermenatation vat at Sierra Nevada.
But in an open system in our atmosphere, there are many many more biases that can affect the measurement within a few inches or feet of the thermometer. Here’s just a few:
- Reflected sunlight from nearby building or objects
- Re-radiated infrared from nearby cement or asphalt surfaces or the ground itself (which is why airports make terrible places for temperature measurement)
- The structure that the thermometer is mounted to, can conduct heat to the thermometer
Now add to that:
- Accuracy of the thermometer itself
- Linearity of the thermometer over its measurement range
- Long term repeatability of the thermometer’s accuracy
- Long term repeatability of the thermometer’s linearity
And then we have urban effects such as:
- Localized vegetatation removal or addition over time
- Localized building changes over time
- Localized asphalt or concrete surfaces addition or removal
And finally within the global temperature records data set we find instances of:
- Changing the location of the weather station and/or its thermometer
- Changing the thermometer itself at some point – i.e. repair/replace
- Changing the thermometer type, from mercury, to electronic (thats been done at thousands of weather stations worldwide)
- Variations in temperature measurement devices from country to country, even though the World Meteorological Organization has specifications, they are not always followed.
- Changes in thermometer shelter, different types of paint over time, all which have different absorptive and reflective properties.
- Changes in the observer recording the temperature, some may round up, others round down numbers. BTW for about 75 years, all temperature records were manually recorded.
Ok with all these biases and possible errors that you have to account for to make long term temperature measurement reflect the true temperature of the location, can you be absolutely sure of the data integrity? Especially when you are looking for trends that may be 1 degree or less over 50-100 years? I can tell you that I’ve looked at these climatological data sets, and NONE of them come with a calibration record for the thermometer, or even a description of the make/model used at that location. There are notations in the records that say things like “station relocated to accomodate construction” or “thermometer replaced” which can give clues to the data integrity possibly changing but the climate researcher is left to make a judgement call on the viability of the data without anything to gauge the sensor or its local environment.
Or lets try a thought experiment Lon, you’ve been commissioned by the IPCC to make a new thermometer for use around the world at climate measurement stations. As an electrical engineer, could you design an air temperature thermometer that is:
- Linear to within 0.1% over a temperature range of -20F to 120F
- Accurate to within 0.1 degree F over that same range
- Repeatable in linearity and accuracy defined above for a period of 20 years. Or even 10 years.
- Identical withing the specs above, so that if one fails, it can be immediately swapped with another one from parts stock with no worry about introducing bias
Ok there’s your challenge. Could you do it?