Using Machine Learning to Classify Stratigraphic Layers of Snow According to the Snow Micro Pen Device

Frolov Denis Maksimovich Scientific Associate, Faculty of Geography, M. V. Lomonosov Moscow State University 119991, Russia, g. Moscow, ul. Leninskie Gory, 1, of. 1904B denisfrolovm@mail.ru Other publications by this author     Seliverstov Yurii Germanovich Researcher, Faculty of Geography, Lomonosov Moscow State University 119991, Russia, Moscow, Leninskie Gory str., 1 yus5@yandex.ru Koshurnikov Andrei Viktorovich Researcher, Faculty of Geography, Department of Cryolithology and Glaciology, Lomonosov Moscow State University 119991, Russia, Moscow, Leninskie Gory str., 1 koshurnikov@msu-geophysics.ru Gagarin Vladimir Evgen'evich Scientific Associate, Department of Geocryology, M. V. Lomonosov Moscow State University 119991, Russia, Moscow, Leninskie Gory str., 1 gagar88@yandex.ru Other publications by this author

Nikolaeva Elizaveta Sergeevna Student, Department of Cryolithology and Glaciology, Lomonosov Moscow State University 119991, Russia, Moscow, Leninskie Gory str., 1 nikolaeva_lizaveta@mail.ru Other publications by this author

Introduction

In light of the ongoing climate warming, 2023 was the warmest year on record. It was almost 1.5 °C warmer than the same value for the base observation period. This warms up the world's oceans and increases the evaporation of water vapor from its surface and the intake of water vapor into the atmosphere. The reduction of the area of sea ice in the Arctic is also increasing, as well as the area of open water and the evaporation of water vapor into the atmosphere. An increase in the concentration of water vapor in the atmosphere leads to disruption of circulation processes, and extreme natural phenomena become more frequent. For example, abnormal snowfalls in Europe (called "snow chaos") and on the European territory of Russia on December 3–4, 2023, as well as the subsequent cold and heavy snowfalls again, were associated, according to the authors, with intense evaporation from the surface of open water of the seas and oceans into the atmosphere.

Snow cover has long been monitored at the meteorological observatory site by the staff of the Geographical Faculty of Moscow State University ^{[1, 2]}. The problem of classifying the layers in the snow column has been studied and is also being investigated by many authors [3–8]. This paper attempts to use AI methods to classify stratigraphic layers of the snow column according to the measurement data of the snow micro pen device, similar to, for example, in the article ^{[9, 10]}.

Materials and methods

In the 2023–2024 winter season in Moscow, the actual temperature in the first half of November, according to observations, was 6.2 °C [https://rp5.ru/]. However, the average monthly temperature in November in Moscow is -0.5 °C [http://www.pogodaiklimat.ru /], and thus, the deviation from the norm for the first half of the month was +6.7 °C. In Moscow, in the first half of November 2023, the weather was abnormally warm, exceeding the November norm by several degrees. According to the new climatic norms (1991–2020), a steady transition of air temperature through 0 °C towards negative values has been taking place in Moscow since November 12. However, in 2022, such a transition occurred on November 15, and in 2023, it happened on November 17. So, in 2023, the arrival of the meteorological winter in Moscow (that is, the moment when there is a steady transition of average daily temperatures through zero toward negative values) occurred only on November 17. On November 23–24, snow began to fall intensely, and snow cover was established.

So on November 24, the thickness of the snow cover in Moscow was 8 cm, and on the morning of November 25, it reached about 15 cm, which is the highest value for November 25 since 2004 and is usually only reached by December 17. The temperature in Moscow on November 24 and 25 was about -6 °C to -8 °C degrees. On the afternoon of the 26th, snow did not increase, and the temperature rose to -3 °C. It was snowing heavily on the evening of the 26th and on the night of the 27th. On the morning of November 27, there was also freezing rain; on the afternoon of November 27, there was a slight thaw to +2 °C degrees.

The snow cover thickness in Moscow on November 27 at the VDNH weather station was 21 cm. On the night of November 27, about 16 mm of precipitation fell. There was also a light snowfall on November 28. On the morning of November 29, it got colder in Moscow, dropping to -10 °C. The snow thickness in Moscow at VDNH on the morning of November 29 was 18 cm, and at the time of research at the meteorological site on November 29, three snow horizons and a crust were found in the snow column. The results of studying the snow cover at the Moscow State University Meteorological Observatory site on November 29, 2023, are shown in Figure 1 and Table 1:

Fig. 1. The structure of the snow column on 11/29/2023 at the Moscow State University meteorological site

Table 1. The structure of the snow column at the site of the Moscow State University Meteorological Observatory on November 29, 2023.

The temperature at the time of researching the snow column was below (0 cm) -0.2 °C, at 10 cm -2.2 °C, at 20 cm -5.6 °C, and 23 cm -4.8 °C.

In the days following the observation, the 29th and the 30th, there was light snowfall, but on the night of December 3 to 4, almost 15 mm of precipitation fell in Moscow, and the thickness of the snow cover at the VDNH weather station increased to 36 cm. (Figure 2).

Fig. 2. Changes in air temperature and snow cover thickness in Moscow according to the VDNH weather station in the winter of 2023–2024.

In the following days, from December 5, the weather was quite cold (-10 °C and below).

At the time of the study on December 7, about seven layers were observed in the snow column (Fig. 3).

Fig. 3. The structure of the snow column on 7.12.2023 at the Moscow State University meteorological site

Table 2. The snow column structure at the Moscow State University Meteorological Observatory site on December 7, 2023.

The temperature at the time of researching the snow column was below (0 cm) 0 °C, at 10 cm -2.2 °C, at 20 cm -3.6 °C, and 30 cm -5.4 °C, and at 33 cm -10.2 °C.

The weather was also quite cold (about -10 °C), and there was heavy snowfall in the following days. At the time of this study on December 14, the thickness of the snow cover was 41 cm (31 cm remained at VDNH, Fig. 2), and eight layers were observed in the snow column (Fig. 4).

Fig. 4. The structure of the snow column on 12/14/2023 at the Moscow State University meteorological site

Table 3. The snow column structure at the Moscow State University Meteorological Observatory site on December 14, 2023.

The temperature at the time of researching the snow column was below (0 cm) -0 °C, at 10 cm -2.2 °C, at 20 cm -2.8 °C, and 30 cm -6.8 °C, and at 40 cm -9 °C.

On the night of December 14th, 2023, a cyclone formed over the Balkans and traveled to Moscow. This was the third cyclone since the beginning of the snow accumulation. In Moscow, even before the night's snowfall, the height of the snowdrifts was high—on December 14 at the VDNKH weather station, the height of the snow cover was 31 cm. A day before December 15, another 7 cm was added, mainly due to overnight snowfall, and the 38 cm figure became not just large but record-breaking. Before that, the highest snowdrifts on December 15 were recorded more than 100 years ago, in 1919, at 32 cm high. Other Moscow weather stations noted an even higher snow height: Baltschug, 43 cm, and Moscow State University,49 cm! Kolomna had the most snow in the Moscow region, at 47 cm.

At the same time, the air temperature rose by the evening of Sunday, December 17, as the capital began to warm, and in the following days, the air temperature ranged from 0 to +2 degrees. There was a long thaw, rain, and snowmelt. On December 21, at the VDNH weather station, the snow cover settled to 24 cm (that is, by 15 cm); at the Moscow State University Meteorological Observatory, the snow cover settled to 28.5 cm (from 49 cm; there was a reduction of almost 20.5 cm). The observed heavy snow accumulation and subsequent intense snowmelt created weather conditions and the threat of flood in areas around bodies of water in Moscow and the Moscow region.

Machine learning is the use of mathematical data models to help a computer learn without direct instructions. It is considered a form of artificial intelligence (AI). Machine learning uses algorithms to identify patterns in the data. The main machine learning algorithms are linear regression, logarithmic regression, decision tree, K-nearest neighbor method (Fig. 5), and others.

Figure 5. Basic machine learning algorithms and their representation.

The essence of these algorithms is well documented in the literature [11–12].

Results and discussion

In the same work, an attempt was made to use AI methods to classify stratigraphic layers of snow according to measurements from the snow micro pen device. The snow micro pen (or snow micro penetrometer) is a device that allows you to determine the hardness of snow layers in increments of 4 micrometers by pushing a probe with a mechanical resistance sensor into the snow [13–15] (Figure 6).

Figure 6. External view of the operation of the snow micro pen device (www.slf.ch/en)

The data sent from the device is processed, and by comparing it with the data gathered through direct snow drilling, a comparison of the classified stratigraphic layers of the snow column was made. In the future, it will be possible, based on the available classified data of the device of stratigraphic layers of the snow column, by clustering K-nearest neighbors, to classify stratigraphic layers according to the newly obtained data of the device without involving additional manual drilling (Figure 7).

Fig. 7. Classification of stratigraphic layer types according to the snow micro pen’s data

Figure 6 shows that the shapes of ice crystals in the snow column resulting from metamorphism (roundedfacetedthawed) differ both in density and in parameters obtained as a result of processing data from the snow micro pen device (MPFN)—the average resistance force SD(N)—its standard deviation, and cv is its covariance. This makes it possible to cluster the processed instrument data and type new instrument measurement data without involving the results of direct manual drilling.