Teller Top Snow Season Start: 2016 @ Tripod

...Separating random depth measurement noise from estimated snow depth.

The way the snow sensor works is that it sends out a series of ultrasonic pulse. The sound pulses reflect off the vegetation / snow / surface below. The yellow cone shape to the left is roughly the sensing area of the instrument. The sensor accurately measures the timing between pulse and the return of the reflection. The speed of sound (meters / second) is a known quantity that varies as a function of air temperature. This knowledge (and math) allows the return time to be converted to a distance measurement between the sensor and surface below.

This plot shows the raw depth measurement (B) plus several types of filtering (C, D, E, F). There are just a couple of things to note. One is the September/ October noise is quite high between readings. This is due in part to the vegetation under the sensor (despite trimming the grass in September!) and what goes into making a depth measurement. From just depth information it is difficult to declare a snow onset date. So, meteorological data is used to narrow the window. Knowing though that snow season begins 11/2, one thing that is obvious is that once there is enough snow on the ground, the measurement to measurement noise drops quite a bit. Just looking at 9/15 data vs 11/14. Estimating the start of the snow season with precision is difficult using just this data set however.

This plot has air temperature in blue and liquid precipitation in violet. I previously use the air temperature to declare the end of the rain season but also easily visible is air temperature remain near freezing until after the snow begins to fall. It appears in 2016 that snow was accumulating while temperatures were still above freezing (about +2 Celsius)... which can happen. But just from air temperature it isn’t clear with enough precision when the snow season starts. Though we can confidently say it is early November.

This slide shows shortwave and longwave radiation data. The longwave is yellow / violet and the shortwave is green/red. Here, in these variables it become more clear when the snow season begins.

The best helper for determining the onset of snow is the shortwave radiation data. The green line shows the energy coming from the sun. The red line shows the energy reflected by the ground surface. The ratio is called the albedo. It is always less than 1.0 but the value for vegetation can vary from 0.1 to 0.3. Looking at the relative difference between Incoming and outgoing… if you look at this data all the way back to September 2016 the albedo is much closer to 0.1 than 1.0. That changes at the peak on 11/3 (remember this data is UTC timezone) Solar noon at the site is about 2:30PM Alaska standard time (or maybe a bit later I can’t remember offhand). That corresponds to 11:30PM UTC. Anway, 11/2 albedo looks to be about 0.22 and 11/3 it is about 0.72. Good clean snow has an albedo about 0.9. So 0.72 suggests fresh snow with grass protruding.

This plot is albedo (again, the ratio of reflected shortwave radiation to incoming shortwave radiation from the sun) during the daytime hours. It is super noisy in part due to the low angle of the sun at this time of year. But a cursory review, it appears there might have been some snow on 10/28 that melted off and then the snow season arrives between 11/2 and 11/3. This meshes with what we eyeballed looking directly at shortwave radiation on the previous slide.

Looking second at the longwave (yellow / violet). The violet is measuring the sky and the yellow the energy of the ground surface. When the values are close together (10/26 to 10/29) that indicates cloudy conditions as the sky is nearly the same temperature as the ground. So, looking more closely at 11/2 to 11/3, it appears that later in the day of 11/2 clouds form and we know from albedo that snow accumulates.

Longwave Radiation Data Analysis

So, knowing albedo changes between 11/2 and 11/3 plus seeing the clouds form mid 11/2, we can zoom in on the snow depth data to look more closely at the data on either side of this point. The general things to note here are again, trimmed grass depth measurements are quite noisy as you see on 11/1. By 11/3 the data is much less noisy and the transition looks to happen around noon on 11/2. So, I declared that the start of the season. Also, you see that black spike around 2-4pm on 11/2. Data like this indicates enough snow falling to create an imaginary snow surface at 5-6cm. It isn’t real, it’s just that the density of blowing snow near the ground surface is high enough that the sound pulse is reflecting off this interface rather than the actual snow surface.

Question Part 2: how do we see a negative snow depth and can it be real? That’s basically the folly of working with real data. As you can see, even though there is less noise once the snow begins to fall, it is difficult to declare an official point where snow depth is 0.0cm. The filtering algorithms help with smoothing this a bit but there really isn’t a good way beyond declaring everything negative to be zero.