Measuring Snow Is Hard!

All observations are bad, but some are useful

- Paraphrased version of George Box quote

There is a tendency in meteorology to treat observations, especially those that are collected in place by weather stations, as ground truth.  Meteorology is a data starved field, which contributes to the tendency to place great value on any observations, but of course observations are not perfect and one needs to consider the possibility of error.  

There are two major sources of error in meteorological observations.  The first is observational error, which is the difference between what is measured and the true value.  Such errors can by systematic, meaning that they occur regularly, or random, which means they vary from observation to observation. A good example of a systematic error is the tendency for thermometers that are not well ventilated to overheat and record a higher temperature than that in the free air when they are in the sun on days with light winds.  A good example of a random error is that related to round off.  For example, many observations are recorded the nearest whole degree (fahrenheit or celcius) rather than with absolute precision, and this introduces some error to every observation.  

The second is representativeness error, which arises from the observation simply not being representative of the area or location to which it is being applied.  For example, we often use the temperature at the Salt Lake City airport as indicative of the temperature throughout the Salt Lake Metropolitan area.  While easy and convenient, this observation often times departs dramatically from temperatures in other parts the metro area, especially during inversions.   

Yesterday provided a good example of how both observational and representativeness errors can complicate the interpretation of snow observations.  These observations are collected by automated stations similar to the one below, which is our Alta-Atwater site (click to enlarge).  There is an ultrasonic snow depth sensor that measures snow depth from a stick mounted over a white board in the lower left of the photo.  Typically such boards are wiped and repositioned either following storms or at regular intervals of no frequently than every 6 hours.  The white cylinder on the platform is a precip gauge that measures snow-water equivalent.  

Here are automated snow amount, water equivalent, and water content observations from four locations in Little Cottonwood Canyon from midnight to 10 am.  These observations are based on ultrasonic snow depth sensors, which measures snow depth on a white board, and precipitation gauges that measure snow water equivalent.  

Alta-Collins (~9700 ft): 6 inches, .23" SWE, 3.8%
Alta-Atwater (~8800 ft): 8.4 inches, .18" SWE, 2.1%
Alta Guard (~8800 ft): 5.9 inches, .31" SWE, 5.2%
Elberts (~7600 ft): 3.15 inches, .09" SWE, 2.9% 

What is to be believed?  The Alta-Atwater and Alta Guard sites are within a couple hundred yards of each other, yet Alta-Atwater has more snow, less SWE, and lower water content.

There are many factors that contribute to the variability seen above.  One is that there were real variations in snowfall that contribute to some of the differences between the two sites.  Two is that the snow depth and precip measurements are essentially point measurements.  We had very low density snow yesterday and even with a small amount of wind can transport snow.  Thus a single measurement might not be representative of the snowfall in the area around the observing site.  Think about how flow around the platform and other obstacles might affect what is measured by the instruments at the Alta-Atwater site.  The effects are probably not large, but they aren't zero.  

Third, the snow-depth measurements aren't fully consistent.  The snow masurement at Alta-Atwater is based on a difference in measured snowdepth at two times with no wiping of the board during the recording period.  The snow measurement at Alta Guard is based on the addition of two measurements taken before and after a wiping of the board.  Each operating group has different approaches and times that they do this.  

Fourth, most precipitation gauges are prone to undercatch, meaning that flow across the gauge prevents all the snow from falling into it and being recorded.  The winds were not strong yesterday, but the snow density was very low, and low density snow is very prone to undercatch.  Some of the variations in SWE probably reflect variations in undercatch.  For example, we have noticed that our site at Alta-Atwater is very prone to undercatch, probably because it is mounted a few meters above the ground, and this might explain the anomalously low SWE and water content at the site.  

Fifth, low-sensity snow can be a poor reflector of sound, and this can lead to measurement errors by the ultrasonic snow depth sensor. 

I can add a few more, but I think you get the point.  There are lots of potential error sources and this complicated the interpretation of automated snow measurements yesterday.  For this reason, snow geeks typically supplement automated measurements with those taken manually. 

Yup, 250 years after the industrial revolution, and the old fashioned way is still the best way to measure snow.  Of course, manual observations have the disadvantage of not being on demand and all the time.