How to make good scientific measurements (and how not to)

One of the principles of good scientific investigation is to eliminate as many variables as possible.  When measuring in a lab, you should avoid changing out thermometers, because they're all different.  If you do, you have to calibrate them first to show they measure the same way the old one did.  You also need to make sure you don't change the location of the thermometer and, if you have multiple thermometers, that they are distributed appropriately.

In 2009, former meteorologist Anthony Watts decided to look at the placement of US surface temperature stations.  He found that "the majority of official climate thermometers in the USA were improperly sited by the government’s own standards."  There were many particularly odd sites. Sometimes the site had more building or remodeling occurred near the site, they were now in parking lots, or near air conditioners:

 or the shade of trees or even near a child's basketball hoop:

(Full report)

This causes the sensor's accuracy to change.  If it's near another heat source, it is warmer than expected.  If it's in the shade all the time, it's likely cooler during the summer.  One cannot have confidence in a thermometer's accuracy if it's improperly placed.
 

Why is this important?

Accuracy in measurement is extremely important in scientific investigations. If the accuracy of a measurement is plus or minus 0.5˚, for example, then any change withing 1˚ could be irrelevant.  

A good example is when you step on a bathroom scale, see your weight, step off, then step back on.  Sometimes the second reading is slightly different.  It's not that you gained or lost weight in a moment, but the device isn't that precise.  

All scientific equipment have manufacturer specifications that show their accuracy.  That accuracy can be checked in a lab setting, as was done in this 2010 American Meteorological Association (AMS) journal article.  (Portions available without membership are here.)  It says regarding the Automated Weather Service (AWS), "the HMP35C sensor in the AWS network can experience more than 0.28°C errors in temperatures from -30° to +30°C [-22° to 86°F]. Beyond this range, the RSS error increases from 0.4° to 1.0°C."

Look at the NOAA graph below:

Notice the temperature scale on the left deals in 2/10th's of a degree.  A critical scientist needs to ask, "Are the measurements of our digital thermometers accurate accurate enough to determine there really is a significant change or is it just statistical "noise" in this example?  Using the data from the AMS journal above, AWS temperature changes are only significant if greater than 0.56°C (0.28°C above the measurement or 0.28°C below it), and up to 2.0°C if they are beyond the -30° to +30°C (-22° to 86°F) temperature range.

Also, given we're now using digital sensors, which are supposedly more accurate devices--ignoring any issues above about improper placement of such devices--does the accuracy of measurements before we began digital thermometers that accurate to determine if there's really that significant a change?