# Analysis of Biases in Fallbrook Temperatures

The plots below show a comparison of my temperatures in 1997 to the official Fallbrook temperatures:

 (Click on graph for bigger and better image.) (Click on graph for bigger and better image.)

Two lines are shown for comparison: one that represents identical temperatures, and one where my temperatures are 10o hotter than the official temperatures. (10o is an interesting comparison since it could result from a typo in the tens digit.)

Note:

• The generally good agreement between my temperatures and the official temperatures, with my temperatures being slightly higher in almost all cases.

• The clear trend in the high temperature comparison for lower differences at very high temperatures.

• My low temperatures are somewhat higher on average than the official low temperatures at very low temperatures.

The following plots show histograms of the temperature differences between my temperatures and the official temperatures for various ranges of official temperatures:

 (Click on graph for bigger and better image.) (Click on graph for bigger and better image.)

The histograms are color-coded with colder official temperatures bluer. The ranges are as follows:

Temperature Plot
ColorHighLow
Blue<=67<=40
Green75-8540-45
Red>=9045-50

These histograms bear out the conclusions reached above. Specifically:

1. My high temperatures are about 4o higher than the official Fallbrook temperature for official temperatures less than 90o and about 2o higher than the official Fallbrook temperature for official temperatures over 90o.

My "high temperature thermometer" is located underneath my porte cochere, and the concrete underneath it traps solar heat and may keep the temperature there a bit warmer. However, the following plot shows that there is no noticeable effect versus the amount of sunshine, which leads me to question whether there is any significant bias due to this cause. Thus this bias may simply be due to a bias in my thermometer, or a real difference in temperature between my location in the Winterwarm area compared to the location of the official Fallbrook temperature near downtown.

The blue histogram shows the difference between my high temperatures and the official high temperatures for all days where the sun shown less than 50% of the day. The red histogram is for days that were 100% sunny.

The plots below use data from 1995 through 1997 to explore the high temperature bias further. The leftmost plot below shows that the high temperature bias is only weakly correlated with date. The rightmost plot shows the bias has the same correlation with temperature as found in 1997.

 (Click on graph for bigger and better image.) (Click on graph for bigger and better image.)

The plots below show the median average of the bias versus month, averaged over all years, for both my high and low temperatures:

 (Click on graph for bigger and better image.) (Click on graph for bigger and better image.)

2. My low temperatures are about 2o higher than the official Fallbrook temperature for official temperatures over 40o. Below that temperature, my low temperatures can range to over 10o warmer than the official temperature, which is probably caused by the location difference. (My location is the Winterwarm area, not the Wintercold area...)

However, it could also be caused by typos in the printed Fallbrook temperatures, since typos resulting in temperatures 10o too high would remove the data from the comparison below 40o, and typos too low would cause a similar bias. I am very suspicious of the claimed Fallbrook low temperature datum of 33o, when I measured 43o and didn't observe any frost on any local roof at the time.

My "low temperature thermometer" is located in a natural area fairly close to my house and close to a large Canary Island Pine, which may produce a slight high bias for the low temperatures. I don't believe the magnitude of this effect is large, but I'll place another high-low thermometer in a more open area sometime to check it.

3. I am absent from Fallbrook about 9 days per month, usually for 3 days per week for 3 weeks, and hence the high and low recorded temperature for that period of 3 days will be recorded accurately, but two days of "lower highs" and "higher lows" were not recorded. This biases the mean high and low temperature record slightly high and low, respectively.

I quantified this error by removing all measurements next to days without measurements, which eliminates not only the readings during my absences but an additional day on either side. (This removed more data than necessary, but made it easier to carry the analysis.) The results showed biases less than 0.2o, which is well within the noise level.

I was a bit surprised that the bias was so low, but it is probably due to the high correlation of temperatures from day to day and the relatively small amount of missing data.

Go To: