1. Introduction
In Arctic regions during extremely cold conditions, temperature inversions to tens of degrees can form due to radiative cooling, which can lead to the air becoming saturated (Malingowski et al. 2014). Ice particles can nucleate in the boundary layer under these conditions sometimes leading to diamond dust or, under colder conditions, ice fog. Ice fog is a type of fog composed of ice particles that generally forms at temperatures colder than −30°C (American Meteorological Society 2012b; definition of ice fog) but can exist at warmer temperatures (Gultepe et al. 2015). Ice fog particles are generally smaller than 100 μm with smaller (sub-20 μm) droxtal-shaped particles common in dense ice fog (Schmitt et al. 2013). Water vapor sources, especially related to anthropogenic activities, can significantly enhance ice fog (Benson 1970; WMO 2017; definition of ice fog). Diamond dust is defined as small ice crystals falling from an apparently cloudless sky (American Meteorological Society 2012a; definition of diamond dust). Diamond dust is common during strong thermal inversions when the air aloft reaches saturation either through mixing with cooler air or by radiative cooling.
Generally, boundary layer ice particles are classified as either diamond dust or ice fog, yet according to Gultepe et al. (2017), the separation between the two classes is “blurry.” The American Meteorological Society (AMS) and WMO definitions of diamond dust both refer to very small ice crystals that precipitate out of a clear sky. Domine et al. (2011) and Girard and Blanchet (2001) both use a 30-μm lower cutoff size for diamond dust with sizes of up to 300 μm being suggested as in Domine et al. (2011). Optical effects such as halos and light pillars are often associated with diamond dust, suggesting that the formation mechanism often leads to some ice particles with reflective surfaces that can become oriented as the particles fall. Ice fog, on the other hand, is often described as being composed of smaller ice particles with no associated halos or other optical effects but is often associated with low visibility situations. Given the ambiguity in the separation between the classifications and given that ice particles often exist in the boundary layer when conditions do not meet either definition, the term “boundary layer ice particles” will be used to describe our measurements throughout this publication.
A comprehensive understanding of boundary layer ice particles is important for a number of reasons. Curry et al. (1990) found that ice fog can significantly impact the surface energy budget. Similarly, Girard and Blanchet (2001) found that diamond dust could significantly enhance the downward flux of infrared radiation. Since optically thick ice fog layers can be confined to the lowest 10–50 m of the atmosphere, surface travel can be significantly disrupted. Ice fog can severely disrupt transportation as reduced visibility can halt airfield operations and vehicular traffic can be impacted by ice fog. Gultepe et al. (2019) and Gultepe and Heymsfield (2016) suggested that accidents in the Arctic related to reduced visibility could increase substantially as air traffic increases in the future. The fact that anthropogenic activities can both enhance and be impacted by ice fog makes it increasingly important to understand.
Bowling et al. (1968) show that upper-level cold air advection can be a precursor to ice fog formation. Subsequent radiative cooling leads to ice particle nucleation and fog formation. Surface radiative cooling often leads to a strong temperature inversion with air temperatures at the surface being much colder than aloft (Malingowski et al. 2014; Gultepe et al. 2014). Due to the low amount of sunlight reaching the surface during winter, the Arctic boundary layer can be extremely stable, usually with strong surface-based temperature inversions (Serreze et al. 1992). The stability of strong temperature inversions severely reduces vertical mixing, leading to vapor and pollution accumulating near their sources at the surface (Cesler-Maloney et al. 2022). The strength of the inversion, as well as the thinness, makes it difficult for forecast models to accurately predict ice fog events (Kim et al. 2014). In urban environments when the relative humidity is at saturation with respect to ice, numerous pathways exist for ice particle formation due to anthropogenic influences. Homogeneous nucleation of liquid water droplets directly into ice crystals occurs at around −38°C (Gultepe et al. 2017). Heterogeneous nucleation, a process involving an additional catalyst particle through immersion freezing, or contact freezing can convert liquid particles to ice at warmer temperatures. Deposition nucleation, the direct transfer of vapor to an ice nucleating particle, and subsequent growth bypassing the liquid state could be a substantial pathway as well in polluted regions.
Boundary layer ice particle microphysical characteristics have been studied for decades. Thuman and Robinson (1954) collected ice fog particles on oil-covered slides and coined the term “droxtal” to represent a compact ice particle, which they concluded was formed by freezing a liquid drop. Kim et al. (2014) noted that homogeneous freezing of sub-micrometers haze droplets could explain the very high concentrations observed in dense ice fog. Hexagonal column and plate-shaped ice crystals are often observed in high-resolution imagery (Schmitt et al. 2013; Lawson et al. 2006; Gultepe et al. 2014). Large hexagonal plate-shaped crystals often lead to light pillars around streetlights at night where flat surfaces on oriented ice crystals reflect light to the viewer.
In this article, we present the results from a 3-yr boundary layer cloud particle measurement campaign. Through three winters, surface and near-surface measurements were conducted continuously and microphysical measurements were conducted when conditions were favorable for boundary layer particle formation. The resulting dataset includes data collected from initiation through dissipation of several cold events. The ice particle microphysical instrumentation used is very well suited to the microphysical characterization of ice particles in the typical size range, and additional shape classification has been conducted using advanced machine learning techniques. Microphysical characteristics, particle concentrations, and particle size distributions are presented based on surface observations and boundary layer conditions as determined by meteorological temperature profilers and photographic observations. With particle shape information on nearly 3 million particles, the dataset is one of the largest collected.
2. Data and methods
a. Field site
Figure 1 shows a map of the Fairbanks, Alaska, region. The map includes instrument locations and significant local sources of vapor and particulate matter such as power plants. Fairbanks is situated at the northern edge of the Tanana Flats region, which is a large flat region between the Alaska Range to the south and a ring of small hills to the north. The research project was conducted at Fort Wainwright Army Base (FW) on the east side of Fairbanks. The goal of the research project was to study the impact of the FW power plant on ice fog; thus, measurements were conducted nearby. Generally, the FW power plant did not directly influence the microphysical measurements, but occasionally, weak winds sent the power plant plume toward the measurement location. The power plant plume can directly impact the Richardson Highway to the south. Instrumentation was located at the Cold Regions Research and Engineering Laboratory (CRREL) Alaska Regional Office building approximately 1 km northwest of the power plant near the base hospital.
Google map of the Fairbanks region. Yellow squares indicate the location of regional power plants. The red circle indicates the CRREL building, the location of two weather stations, the PPD-2K, two axis panorama cameras, and the ATTEX MTP-5HE. The eastern red triangle indicates the location of MTP-5PE. The northern red triangle indicates the Birch Hill site with a weather station and an axis panorama camera.
Citation: Journal of Applied Meteorology and Climatology 63, 6; 10.1175/JAMC-D-23-0190.1
b. Instrumentation
Several instruments were deployed for this study (Table 1). Two weather stations were located at the CRREL building at 2 and 10 m. For the 2019/20 winter, a meteorological temperature profiler (MTP-5HE) was also located at the CRREL building. A third weather station was located near the top of the Birch Hill Ski Area, approximately 150 m in altitude higher than the CRREL location and 4 km to the northwest. Two axis panoramic cameras at the CRREL building collected images of the atmospheric conditions. One camera collected images to the east including a view of Fort Wainwright Hospital and the plume from the FW power plant, and the second camera collected images to the north where street lights easily illuminated ice fog at night. The 180° panoramic images from both cameras were saved to a local computer every 10 min throughout the observation period. A third axis panorama camera was collocated with the Birch Hill weather station. The Birch Hill camera collected images to the south, overlooking all of FW.
Instrumentation type and location.
A second meteorological temperature profiler (MTP-5PE) was installed at the east end of Ladd Army Airfield at FW in February 2020. The MTP-5PE was pointed directly west along the runway and operated there until May 2022. The MTP-5 instruments passively measure microwave radiation from which the temperature profile is estimated for the lowest 1000 m of atmosphere. Both MTP-5 models provide a temperature profile every 5 min. For the purposes of this publication, factory calibration is used, which provides temperature information at least every 50 m through the lowest 1000 m. The 12 profiles per hour were averaged when compared to other datasets, which were averaged by hour as well. While the MTP-5 instruments are relatively new instruments, early studies indicate that the instruments work well especially in the lowest 500 m of the atmosphere (Cadeddu et al. 2024).
The ice particle microphysical measurements were conducted at the CRREL building using a Particle Phase Discriminator mark 2, Karlsruhe edition (PPD-2K), developed at the University of Hertfordshire, United Kingdom. This instrument is similar to the instrument described in Kaye et al. (2008). The PPD-2K is a laboratory version of the small ice detector-3 (SID-3; Ulanowski et al. 2012, 2014). A brief description of the operation of the PPD-2K is given here. For a complete description of the operation of the PPD-2K, see Vochezer et al. (2016) and the aforementioned publications. The PPD-2K operates by detecting laser light that is scattered through interactions with individual atmospheric cloud particles. A frequency-doubled neodymium-doped yttrium–aluminum–garnet (Nd:YAG) laser shines a 100-mW 532-nm laser beam through a sample volume. A vacuum pump pulls 5 L m−1 of air into the instrument. A specially designed inlet nozzle accelerates and focuses the sampled air into the laser beam so that all particles in the sampled air intersect the laser beam at roughly the same speed. When a cloud particle passes through the laser beam, the scattered light is measured in two ways. A beam splitter sends a portion of the light to a trigger detector, and the additional light goes to the imaging camera. The trigger detector operates continuously, and the intensity of light observed by the trigger detector is used to estimate the size of all particles that pass through the sample volume. If the trigger detector signal is higher than a minimum threshold, a 582 × 592 charge-coupled device (CCD) camera is signaled to collect an image of the light that is scattered by the particle. While the trigger detector operates continuously counting and sizing all particles, the camera samples at 30 Hz meaning that depending on the actual particle concentration, a fraction of particles pass through the sample volume without their scattering pattern being imaged.
During the 3-yr study, approximately 3 million scattering patterns were recorded by the PPD-2K. The PPD-2K was not operated continuously; rather, it was operated during time periods when ice particles were anticipated. In total, the PPD-2K was operated for approximately 315 h with varying measurement strategies from constant operation to operating 5–15 min every half hour. As data for this publication are averaged over 1-h periods, this led to 673 h during which the PPD-2K operated. Table 1 shows the dates and gives short descriptions of each of the measurement periods during which the PPD-2K was operated.
c. Methods
The PPD-2K data were analyzed as in Vochezer et al. (2016). For particle sizing calibration of the trigger light scattering signal, water droplets were pulled into the PPD-2K and were sized by comparing the scattering patterns to the Mie theory (Mie 1908) to get an unambiguous determination of the droplet size. Droplet sizes were then related to the trigger intensity, and a relationship was determined between trigger intensity and particle size, which was applied to all particles. As ice particles typically scatter less light in the near-forward direction, sizes are likely underestimated for ice. Jang et al. (2022) show that there can be up to 40% error in sizing when a Cloud and Aerosol Spectrometer (CAS) probe calibrated for liquid drops is used to size ice crystals. They used ice particle phase functions calculated for randomly oriented hexagonal crystals. As the sizing detector on the CAS has a narrower angle of view than the PPD-2K, it would not be appropriate to use their correction factor. Data presented herein are sphere-equivalent sizes without any correction. Equivalent sphere particle size distributions were calculated for different length time periods depending on the quantity of data. In the most dense cases, size distributions were calculated every 10 s, while in thin cases, it was necessary to use 5 min of data for a statistically reasonable particle size distribution. The second component of the PPD-2K measurement, the scattering patterns, can be used to extract shape information. Complex particles (often referred to as rough-surfaced particles) have scattering patterns characterized by speckled patterns (Schnaiter et al. 2016). Experiments at the Aerosol Interaction and Dynamics in the Atmosphere (AIDA) cloud chamber show that speckled scattering patterns often occur when ice nuclei have many surfaces that promote crystal growth, such as dust or homogeneously frozen acid solution droplets. The degree of complexity can be estimated with the gray-level co-occurrence matrix (GLCM) method described in Lu et al. (2006). Vochezer et al. (2016) identify pristine particles by first determining the directional intensity of scattered light and then applying a fast Fourier transform (FFT) to identify symmetries. If the strongest FFT coefficient is 2, 4, or 6, the particle is thought to be a hexagonal column or plate.
To extract as much information as possible, machine learning (ML) was used to categorize the PPD-2K scattering patterns. The PPD-2K ML study is being submitted as a companion study to this one (Schmitt et al. 2023, manuscript submitted to Artif. Intell. Earth Sci.). Figure 2 shows the overall breakdown of all particle shapes as derived by the ML study. The training datasets for the ML study and how they differ from previously described particle types are noted below. The largest category is labeled rough. Rough particles have either surface roughness such as layered or hopper growth or are polycrystals (Schnaiter et al. 2016). A particle with a roughened surface has many edges that introduce stochastic light diffraction, resulting in speckled scattering patterns. The ML study attempted to identify the same subset of particles as the GLCM analysis. Sublimating particles are those that scatter light as if all sharp edges have been rounded by sublimation (Schnaiter et al. 2016). While these are easy to identify by eye, the ML algorithms are thus far the only way to identify the particle type automatically. Irregular scattering patterns generally do not have speckles, nor are they obviously hexagonal. Scattering patterns classified as saturated generally have a high percentage of saturated pixels in the image, but typically are speckled and therefore likely rough. Spherical particles scatter light according to the Mie theory and have near-perfect concentric rings. Details on the ML study and direct comparisons between analysis techniques can be found in Schmitt et al. (2023, manuscript submitted to Artif. Intell. Earth Sci.).
Pie chart showing the percentages of particles classified into each category by the ML algorithms for the full dataset. Note that in the remainder of the figures, percentages less than 4% are omitted from the figures.
Citation: Journal of Applied Meteorology and Climatology 63, 6; 10.1175/JAMC-D-23-0190.1
The panorama camera images were also analyzed using ML with a similar strategy as described in Schmitt et al. (2023, manuscript submitted to Artif. Intell. Earth Sci.). A sufficient training dataset was only possible for nighttime images. As daylight hours can be as low as 3:41 in the winter in Fairbanks, there were approximately twice as many images in the winter nighttime dataset (December through March). Figure 3 shows example images from the panorama cameras with the area used in the convolutional neural network (CNN) analysis being indicated with the white rectangles. For the camera ML study, a training dataset including 250 images per category was identified from the full dataset, which had on the order of 30 000 images total. The Visual Geometry Group-16 (VGG-16) convolutional neural network was used for the study. Figure 3 also shows example images from each of the categories for the cameras. Categories included snowing, cloudy, clear, fog, and power plant plume size. Two confusion matrices and the statistics for different observed conditions are shown in Fig. 4. While the total number of images analyzed is similar for both datasets, there are some differences that are not initially intuitive. The focus of the east camera was to understand the plume so the area analyzed was specific to the plume. For the north camera, the sky was often clear when the plume was present in the east camera view. This explains why the north camera had more occurrences of clear skies. The small number of images categorized as “snowing” was further investigated. It was determined that light snowfall was extremely difficult to see in the images, suggesting that the snowing category is actually “heavily snowing.” The fog training dataset was purposely set up to detect very light fog with the assumption that temperature screening could also be used to separate fog from snow events since snow in Fairbanks typically occurs at warmer temperatures (Hartl et al. 2023). The purpose of the camera study was to characterize atmospheric conditions during our measurement periods. Uncertainties that could arise (such as snowing conditions being identified as ice fog) were not important as the PPD-2K was only deployed when there was no chance of snowfall.
Example images from axis panoramic cameras showing the full image and the cropped region used in the ML study and example images from each of the ML categories.
Citation: Journal of Applied Meteorology and Climatology 63, 6; 10.1175/JAMC-D-23-0190.1
Confusion matrices for the two panoramic camera ML studies. Each row of the matrices is for one labeled category, which was classified as shown in columns. Bar graphs showing the frequency of occurrence of each categorized sky condition.
Citation: Journal of Applied Meteorology and Climatology 63, 6; 10.1175/JAMC-D-23-0190.1
The CNNs appeared to be performing well when applied to unlabeled data. To confirm this, saliency maps were created in order to assess if the models were focusing on the important image characteristics. The process of constructing saliency maps involves extracting values from a CNN during the classification process (Przybylo et al. 2022). Then, the saliency map can be compared to the image being classified. The saliency map should highlight the areas of the image that are most important for the classification of the image, which should also correspond to how a person would identify the condition. Several individual images and their associated saliency maps are shown in Fig. 5. Subjectively, the saliency maps suggest that the CNNs were working as expected with the critical regions being highlighted.
Saliency maps and associated images for several categorized images. Brighter areas on the saliency maps indicate the areas of the image that were most important for categorization using CNN. The saliency map corresponding to the fog image shows that the CNN identifies the locations where the fog is visible in the lights, while the plume area is highlighted in the plume images.
Citation: Journal of Applied Meteorology and Climatology 63, 6; 10.1175/JAMC-D-23-0190.1
3. Results and discussion
a. Microphysics during cold events: Case studies
In this section, we will present data from some of the more significant cold events during the project. All of the datasets have been averaged by hour through all of the measurement periods. The two case studies are the final two from Table 2 (IOP11 and IOP12), which both occurred in 2022. These events were chosen because they included significant measurement time before and after the main cold periods. The PPD-2K was operated on a 15-min on/15-min off cycle throughout both IOPs. IOP11 took place from 3 to 9 January 2022 and included an initial cold period followed by a warmup and then a second colder time period. The coldest temperature measured during the period was −41.5°C, and approximately 800 000 PPD-2K scattering pattern images were saved during the period. As the sun was only above the horizon 4–5 h per day during this time period, there was very little impact from solar input. Figure 6 shows data from this episode. The particle habit mixture shows some interesting trends that are specific to cold periods. Note that the highest particle concentrations coincide with high occurrences of particles categorized as sublimating (7 and 8 January). This is likely due to the high concentrations necessitating that the saturation stays very close to 100%. On the microscopic level, this leads to edges being slightly supersaturated, while flat surfaces are slightly subsaturated. The result is that sharp edges slowly sublimate over time, while flat surfaces collect ice, thus leading particles to trend toward quasi-spherical. This is analogous to the well-known vapor pressure change over curved surfaces described by the Kelvin equation (Thomson 1871), which mathematically describes that the implications of water molecules at a curved surface have less opportunity to bond with other water molecules when compared to a flat surface. Pristine particles are present in higher relative concentrations during periods when the temperature is relatively constant or cooling (late 4 January through the morning of 6 January). During this period, the near-surface temperature was steadily cooling, suggesting a particle formation mechanism similar to diamond dust. The temperature profiler data show a relatively deep inversion (the line marks the warmest level on each profile) for the majority of the time period with the inversion thinning during the final few days. An interesting feature is the warm period in the middle of the period (6 January). This coincides with a short time period of overcast skies. The arrival of the cloud layer coincided with a significant increase in the surface temperature and elimination of the inversion, followed by a warm period while the cloud was present and then surface cooling began again after which the coldest temperatures of the period were registered. The cloud layer was high, optically thick, and very uniform. Interestingly, the camera ML did not identify the initiation of the cloudy period and only began to identify the cloudiness once the sky started to lighten. The ML CNN identifies clouds by reflected light from Fairbanks, and the cloud layer was too high to reflect much urban light. Nevertheless, visual scrutiny of the camera images shows that this was the only overcast period during the 7-day period. The main cold period ended late on 8 January 2022 with warming at the surface. Note that the proportion of particles categorized as sublimating reduced quickly at the time coincident with the start of the warming (late 8 January). This also coincided with a sharp reduction in particle concentration, suggesting that the low-level droxtal-shaped particles all sublimated away, leaving only the larger particles that formed at higher altitudes.
Cold periods when the PPD-2K was operated to collect microphysical measurements. The date, coldest observed temperature, approximate duration of IOP (hours of operation of the PPD-2K), the number of PPD-2K images collected, and a brief description of the events are included.
Data from IOP11 from 3 to 10 Jan 2022. Time series starts at midday on 3 Jan with midnights indicated on the x axis. Panels show particle shape from the ML study, PPD-2K particle size distributions, and the temperature profile graphic from the MTP-5PE for the time period with the warmest temperature indicated by the line. The two line graphs show the total particle concentration observed by the PPD-2K and the surface temperature (°C) measured at the PPD-2K location (blue) and at the MTP-5PE location (orange).
Citation: Journal of Applied Meteorology and Climatology 63, 6; 10.1175/JAMC-D-23-0190.1
The second case study, IOP12, took place from 28 January to 4 February 2022 and is shown in Fig. 7. This event was more episodic in nature with the surface temperature warming during daylight periods and then cooling again at night. The diurnal variation is clearly visible in the surface temperature plots in the bottom panel. The lowest temperature recorded was −39.5°C, and the PPD-2K captured 400 000 images. Each day after sunrise, the surface temperature started to warm. Warming at the 50- and 100-m levels as measured by the MTP-5 lagged the surface warming. The ice particle concentrations were not as high during this period, but there were interesting aspects to the ML classification. The particles classified as sublimating, in this case, occurred during the warming periods, suggesting that they were particles that were losing their sharper edges due to warming conditions. Notice the spikes in the sublimating categories on 29, 30, and 31 January that are aligned with the daily surface temperature increases. Pristine particles were present in relatively high percentages except during the warming periods.
As in Fig. 6, but for the IOP12 period between 27 Jan and 3 Feb 2022.
Citation: Journal of Applied Meteorology and Climatology 63, 6; 10.1175/JAMC-D-23-0190.1
b. Microphysics versus conditions
Understanding the impact of different conditions on the microphysics of boundary layer ice particles is of substantial interest. Here, we investigate boundary layer particle habit and particle size distributions under different conditions by sampling from the entire dataset. As before, hourly data are used. For each hour when measurements were conducted, all data collected during that hour are averaged and then those values are averaged for the combined statistics. This prevents heavier cases from dominating the statistics when concentrations are highly variable. Table 3 provides statistical information for each of the subsets of data presented graphically in the remaining figures. Note that for particle concentrations, the 25th and 75th percentiles are given rather than the standard deviation, which can be nonsensical with skewed data.
Additional data for each of the datasets are presented in Figs. 9–11. The median and mean concentrations and the 25th and 75th quartiles are in number per liter. The average inversion height and strength are the average of the inversion heights and strengths of all of the cases (not the height of the average temperature profile) in meters above the surface and the temperature difference (°C). The median mass diameter is the average of the values calculated for each particle size distribution in micrometers. The mean and standard deviation (in parentheses) of the temperatures measured at the PPD-2K location and the characteristic condition that was used to sort the data.
Figure 8 shows example graphs that will be created for each of the conditions explored. The data presented in Fig. 8 are from 664 of the 673 h during which the PPD-2K was operating (indicating that nine of those periods registered particle concentrations of 0). The average concentration for all periods was 76.0 L−1 with a maximum concentration of 2065 L−1. The temperature histogram indicates that the majority of the sampling occurred at temperatures colder than −20°C although some sampling occurred at temperatures as warm as −16°C. The average sphere equivalent particle size distribution shows particles from the lower size limit (∼8 μm) up to about 100 μm with a lot of variability as shown by the individual size distributions in light gray. Actual ice particle sizes were likely larger given the uncertainty when converting sphere equivalent sizes to ice particle sizes (Jang et al. 2022). Corrections have not been applied to the data as a correction factor has not been published for the PPD-2K. The total concentration histogram shows the number of cases in logarithmically spaced bins with the mode being around 10 L−1. While concentrations lower than this do not impact visibility, this along with the temperature show that the observational dataset includes a lot of edge cases. For all observations, the average temperature profile shows a 10.3°C temperature inversion of the warmest temperature in the profile averaging 497 m above the surface with a lot of variability throughout the profile.
Example graphs from the full observed dataset. These figure types will be duplicated for different conditions in the final figures. (a) Particle size distribution for ice particles observed during selected condition. There is one gray power spectral density (PSD) for each hour when the condition was met, and the black PSD is the average. (b) As in (a), but for MTP-5 temperature profile data. (c) Histogram of temperatures observed for the selected condition. (d) Histogram of the logarithm of the ice particle concentrations measured by the PPD-2K for the selected condition.
Citation: Journal of Applied Meteorology and Climatology 63, 6; 10.1175/JAMC-D-23-0190.1
The simplest way to segregate the data for comparison is by temperature and by time of day. Figure 9 shows the results of these comparisons. The top four rows of data show the observations for different temperature ranges. As would be expected, the highest concentrations (average and peak) are at the coldest temperatures, while the warmer temperatures have average concentrations an order of magnitude lower. The only substantial difference in the average particle size distributions for the four temperature ranges is that the coldest temperature range lacks particles larger than 50 μm, while the warmer temperatures include particles close to 100 μm. The particle size distributions for the coldest temperature range (from −35° to −40°C) show distinctly higher concentrations at the smallest sizes. The particle habit chart indicates a substantially higher relative concentration of particles classified as sublimating for the coldest temperatures. This is likely due to the previously mentioned saturation effect that leads to sharp edges being eroded when particles spend a lot of time very close to saturation. In total, there were 20 time periods when the concentration exceeded 500 L−1 and 16 came when the surface temperature was colder than −33°C. Note that spherical particles made up only 0.32% of the particles during the coldest measurements yet increased to 1.5% for the two warmer temperature ranges. From the temperature profiler data, the strongest inversions are at the coldest temperatures, but two distinct types of temperature profiles appear in the −35° to −40°C graph. One set of profiles shows temperatures continuing to warm above the 1000-m level, while the other main group shows the warmest temperatures at ∼200 m. Further investigation led to the determination that the first group was from the early stages of the longest-lasting cold period during the 3 years of measurements (IOP1 in January 2020). The temperature inversions become weaker and less deep for warmer temperatures. Note that the temperature data used for sorting the dataset were not collocated with the temperature profilers, so the temperature at the lowest altitude for the MTP might be warmer or colder than the temperature range for the category. See the temperature line graphs in Figs. 6 and 7 to see examples of the temperature difference between the two locations.
Characteristics of ice particle populations observed at different temperatures and times of day. The first four rows are data from observations binned by 5°C temperature ranges. The next four rows are data from observations binned in 6-h blocks throughout the day.
Citation: Journal of Applied Meteorology and Climatology 63, 6; 10.1175/JAMC-D-23-0190.1
The final four rows of data in Fig. 9 show the data sorted by time of day. Data were only used if the temperature was below −20°C. The data are separated into four 6-h periods throughout the day. The highest concentrations were seen during the 0600–1200 LT time period. The early morning hours (0000–0600 LT) had the lowest particle concentrations. Interestingly, the nighttime hours (1800–2400 and 0000–0600 LT) both had higher percentages of pristine particles, while particles classified as sublimating were observed at higher percentages during the 1200–1800 LT time period. During sunlight hours, the temperature warms, which could lead to decreased ice concentrations, pollution and water vapor sources, specifically from vehicular traffic, are significantly increased.
From time to time, there appears to be an increase in concentrations centered around the 40-μm size bin in the particle size distribution. There does not appear to be any systematic reason for this bump although these elevated values tend to occur during periods when there are higher concentrations of pristine-shaped particles (notice the higher concentrations in the size distribution in Fig. 7 do align with higher percentages of pristine particles). Particle size distributions with this feature are mixed throughout the full dataset without any apparent preference. The slight increase in large particles on 0830 LT 7 January 2022 (Fig. 6) was recorded at the same time that very strong light pillars were observed in the north panorama camera image.
Figure 10 shows the results of investigating extremes in the dataset. The first two rows of data show data from the extreme cases of particle concentrations. The top row shows data from the 10 cases when the concentration averaged more than 1000 L−1. These were some of the coldest periods when there were high concentrations of particles classified as sublimating. The second row shows the characteristics when particle concentrations were less than 10 L−1. Lower concentrations lead to less restricted growth with the available vapor meaning that the sharper edges can grow, leading to higher relative concentrations of rough particles. The next two conditions of interest were the characteristics observed during the fastest warming and fastest cooling time periods. Data from the largest 100 temperature changes in the previous 6-h period are displayed. For the cooling periods, data include all time periods when the temperature had cooled more than 4.25°C in the previous 6 h, while for the warming time periods, a temperature increase of 4.25°C over the previous 6 h was necessary to identify the 100 most extreme cases. The particle size distributions for both datasets do not appear too different except that several from the cooling dataset had much higher concentrations than the others. For particle shapes, the main difference was the percentage of particles classified as sublimating. While previously, sublimating particles thought to be droxtals have had relatively higher percentages in colder scenarios, they are less numerous in the cooling case. For the warming periods, the overall particle concentrations are significantly lower and the sublimating particles are likely to have that shape due to sublimation. Pristine particles were observed in higher percentages during cooling periods.
Data were sorted to identify the most extreme time periods for each situation. The first two rows show data from the time periods with the highest and lowest average particle concentrations. The next two rows show the 100 time periods when the temperature had increased or decreased the most in the previous 6 h. Rows 5–8 show data from more extreme temperature inversion cases based on the MTP-5 temperature profiles.
Citation: Journal of Applied Meteorology and Climatology 63, 6; 10.1175/JAMC-D-23-0190.1
The relationship between the temperature profile measured by the MTP-5 instruments and ice particle microphysical properties is also shown in the lower part of Fig. 10. The inversion height and strength were both estimated based on the MTP-5 temperature profiles. Some temperature profiles showed two temperature peaks. To identify the inversion altitude, first the maximum temperature in the profile was identified. If that location was below 500 m, that height was taken to be the inversion altitude. If it was higher than 500 m, then the maximum temperature below 500 m was identified. If the maximum temperature below 500 m was warmer than the temperature at 500 m, then the lower peak was taken to be the inversion height. The surface temperature was subtracted from the temperature at the inversion height to determine the inversion strength. For the full dataset, the maximum MTP-5-derived inversion strength was greater than 21°C. In Fig. 10, the 100 most extreme cases of each type are considered. The 100 strongest inversions were identified as having inversion strengths of at least 17.3°C, while the 100 weakest inversions had temperature differences of less than 3.7°C. The 100 deepest inversions were defined as time periods when the inversion height was at least 700 m above the surface, and the 100 shallowest inversions were identified as times when the inversion height was below 200 m. As with previous comparisons, only time periods with surface temperatures colder than −20°C were considered. As would be expected, the time periods with deep inversions and strong inversions had substantially higher ice particle concentrations than times with shallow or weak inversions. This was more pronounced in the small size ranges. While there did not appear to be substantial differences in the ice particle shapes, “rough” particles accounted for a higher percentage in both the weak inversion cases and the shallow inversion cases, while pristine particles were observed in higher percentages in the strong and deep inversion categories. Note that the deep and strong temperature inversions also include some of the coldest temperatures as compared to the other two cases.
For the use of the panorama camera imagery results, a few caveats need to be considered. As mentioned, only the nighttime images were characterized using ML; thus, using the camera ML data to sort observations restricts results to nighttime cases. Additionally, camera observations were made every 10 min so there were six measurements during each hour. Figure 11 shows the microphysical characteristics observed under different conditions identified by the cameras. As mentioned in the methods section, the camera dataset was significantly smaller, and therefore, the results of the ML study are not anticipated to be as robust. For example, there were only two 1-h periods when fog was detected in all six images from a single hour. Figure 11 shows the microphysical characteristics when the power plant plume was present or absent compared to when at least one of the north camera images detected fog during the hour. The particle size distributions indicate that the observable fog is typically associated with higher concentrations of particles at smaller sizes and strong though shallower inversions. When the plume was easily identified, the MTP-5 measurements showed strong deep inversions and a higher percentage of pristine particles, whereas particles observed during ML-defined clear time periods tended to be less pristine and these time periods coincided with very shallow inversions. The average concentration during clear periods was substantially lower (as would be expected), but the maximum concentrations were similar. The other major difference is that the fog-detected cases included a very high percentage of pristine particles and particles classified as sublimating, thought to be droxtals. The rough category has the highest percentage for clear (no plume) periods. Of all of the categories shown, the fog category included the lowest percentage of liquid particles.
Based on ML results, the first three rows show the microphysics from when the power plant plume was observed compared to when no plume was observed by the east camera and when fog was present in the north camera. The final three rows show data from when specific particle types were more dominant in the microphysical dataset.
Citation: Journal of Applied Meteorology and Climatology 63, 6; 10.1175/JAMC-D-23-0190.1
The final data presented are from time periods when a particular particle type was present in higher concentrations. For each category, the 100 periods with the highest percentage of that category are shown. High percentages of rough particles were typically associated with deeper although on average weaker temperature inversions and modestly warmer time periods (T > −34°C). Pristine particles were also associated with deeper but stronger inversions and warmer temperatures (note that the bump at 40 μm is more prominent in this case). Particles categorized as sublimating were associated with a bimodal distribution of temperatures. As stated before, the cold periods are likely cases of droxtal-shaped particles, while the warmer periods are suspected to be due to actual sublimating particles.
4. Summary and conclusions
Boundary layer cloud particle measurements have been conducted over three winters in Fairbanks, Alaska. The focus was to investigate boundary layer ice particles as opposed to falling snow. Particle measurements were conducted throughout the winter starting in December 2019 until March 2022. The PPD-2K was deployed when the temperature was expected to drop below −30°C and when no precipitation was expected. In total, the PPD-2K was operated for 315 h. Ice particle habits and boundary layer conditions have been classified using machine learning. Among the numerous findings detailed in the results and discussion section, a few key points stand out. First, although observations were conducted at temperatures as warm as −15°C, liquid water droplets were very rarely observed with only 1.3% of all observed particles displaying spherical shapes. The dearth of liquid particles, even at temperatures significantly warmer than the −38°C homogeneous nucleation temperature threshold, suggests that ice nucleation pathways are abundant. Even when observations were limited to temperatures warmer than −20°C, the liquid percentage was 4.3%. Second, ice fog conditions (when the temperature was the coldest) tended to be characterized by much high concentrations than warmer periods and high percentages of the particles have similar scattering characteristics to particles thought to be sublimating based on comparisons with data from the AIDA cloud chamber. These are thought to be droxtals, which have been observed in ice fog studies in the region since the 1950s. It is possible that the high concentrations of ice particles are present due to high urban nuclei concentrations, which activate uniformly (haze droplets as in Kim et al. 2014; Shaw 1983), and then, due to the high numbers, can only grow minimally. Shaw (1983) measured the concentrations of at least 170 haze particles per cubic centimeter (well above the highest ice particle concentration measured in this study), and Kim et al. (2014) found that nucleation of haze particles was the best way to account for the observed concentrations in their modeling study. Since the ice particle concentrations are so high during the coldest events, the relative humidity is likely in equilibrium. This would lead to the aforementioned process of the erosion of sharp edges on the crystals, thus making them appear to be quasi-spherical. Light scattering calculations based on near-spherical particles as shown in Schnaiter et al. (2016) are similar to sublimating particles; thus, the high concentrations of particles that appear to scatter as sublimating particles could be the droxtal-shaped crystals, which have been observed in Fairbanks ice fog studies since the 1950s (Thuman and Robinson 1954).
Acknowledgments.
This material is based upon work supported by the Broad Agency Announcement Program from the U.S. Army Cold Regions Research and Engineering Laboratory (ERDC–CRREL) under Contract W913E521C0017 from the U.S. Army Basic Research Program (Program Element 0603119A, Ground Advanced Technology). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Broad Agency Announcement Program or ERDC–CRREL. The authors would also like to thank Dominique Pride, Heike Merkel, and Tom Douglas for their leadership in this project.
Data availability statement.
Data are available at https://akclimate.org/projects/icefog/.
REFERENCES
American Meteorological Society, 2012a: Diamond dust. Glossary of Meteorology, https://glossary.ametsoc.org/wiki/Diamond_dust.
American Meteorological Society, 2012b: Ice fog. Glossary of Meteorology, https://glossary.ametsoc.org/wiki/Ice_fog.
Benson, C. S., 1970: Ice Fog: Low Temperature Air Pollution. Vol. 121, Corps of Engineers, US Army, Cold Regions Research and Engineering Laboratory, 198 pp.
Bowling, S. A., T. Ohtake, and C. S. Benson, 1968: Winter pressure systems and ice fog in Fairbanks, Alaska. J. Appl. Meteor. Climatol, 7, 961–968, https://doi.org/10.1175/1520-0450(1968)007<0961:WPSAIF>2.0.CO;2.
Cadeddu, M., M. Stuefer, T. Wong, and C. Schmitt, 2024: Boundary layer temperature profiles from a V-band scanning radiometer in Barrow, AK. MicroRad Conf., Alexandria, VA, Institute of Electrical and Electronics Engineers.
Cesler-Maloney, M., W. R. Simpson, T. Miles, J. Mao, K. S. Law, and T. J. Roberts, 2022: Differences in ozone and particulate matter between ground level and 20 m aloft are frequent during wintertime surface-based temperature inversions in Fairbanks, Alaska. J. Geophys. Res. Atmos., 127, e2021JD036215, https://doi.org/10.1029/2021JD036215.
Curry, J. A., F. G. Meyers, L. F. Radke, C. A. Brock, and E. E. Ebert, 1990: Occurrence and characteristics of lower tropospheric ice crystals in the Arctic. Int. J. Climatol., 10, 749–764, https://doi.org/10.1002/joc.3370100708.
Domine, F., and Coauthors, 2011: The specific surface area and chemical composition of diamond dust near Barrow, Alaska. J. Geophys. Res., 116, D00406, https://doi.org/10.1029/2011JD016162.
Girard, E., and J.-P. Blanchet, 2001: Microphysical parameterization of Arctic diamond dust, ice fog, and thin stratus for climate models. J. Atmos. Sci., 58, 1181–1198, https://doi.org/10.1175/1520-0469(2001)058<1181:MPOADD>2.0.CO;2.
Gultepe, I., and A. J. Heymsfield, 2016: Ice fog, ice clouds, and remote sensing. Pure Appl. Geophys., 173, 2977–2982, https://doi.org/10.1007/s00024-016-1380-2.
Gultepe, I., and Coauthors, 2014: Ice fog in Arctic during FRAM–ice fog project: Aviation and nowcasting applications. Bull. Amer. Meteor. Soc., 95, 211–226, https://doi.org/10.1175/BAMS-D-11-00071.1.
Gultepe, I., and Coauthors, 2015: A review on ice fog measurements and modeling. Atmos. Res., 151, 2–19, https://doi.org/10.1016/j.atmosres.2014.04.014.
Gultepe, I., A. J. Heymsfield, M. Gallagher, L. Ickes, and D. Baumgardner, 2017: Ice fog: The current state of knowledge and future challenges. Ice Formation and Evolution in Clouds and Precipitation: Measurement and Modeling Challenges, Meteor. Monogr., No. 58, Amer. Meteor. Soc., https://doi.org/10.1175/AMSMONOGRAPHS-D-17-0002.1.
Gultepe, I., and Coauthors, 2019: A review of high impact weather for aviation meteorology. Pure Appl. Geophys., 176, 1869–1921, https://doi.org/10.1007/s00024-019-02168-6.
Hartl, L., C. Schmitt, T. Wong, D. A. Vas, L. Enterkin, and M. Steufer, 2023: Long-term trends in ice fog occurrence in the Fairbanks, Alaska, region based on airport observations. J. Appl. Meteor. Climatol., 62, 1263–1278, https://doi.org/10.1175/JAMC-D-22-0190.1.
Jang, S., and Coauthors, 2022: Theoretical calculations of directional scattering intensities of small nonspherical ice crystals: Implications for forward scattering probes. Remote Sens., 14, 2795, https://doi.org/10.3390/rs14122795.
Kaye, P. H., E. Hirst, R. S. Greenaway, Z. Ulanowski, E. Hesse, P. J. DeMott, C. Saunders, and P. Connolly, 2008: Classifying atmospheric ice crystals by spatial light scattering. Opt. Lett., 33, 1545–1547, https://doi.org/10.1364/OL.33.001545.
Kim, C. K., M. Stuefer, C. G. Schmitt, A. Heymsfield, and G. Thompson, 2014: Numerical modeling of ice fog in interior Alaska using the weather research and forecasting model. Pure Appl. Geophys., 171, 1963–1982, https://doi.org/10.1007/s00024-013-0766-7.
Lawson, R. P., B. A. Baker, P. Zmarzly, D. O’Connor, Q. Mo, J.-F. Gayet, and V. Schherbakov, 2006: Microphysical and optical properties of atmospheric ice crystals at South Pole Station. J. Appl. Meteor. Climatol., 45, 1505–1524, https://doi.org/10.1175/JAM2421.1.
Lu, R.-S., G.-Y. Tian, D. Gledhill, and S. Ward, 2006: Grinding surface roughness measurement based on the co-occurrence matrix of speckle pattern texture. Appl. Opt., 45, 8839–8847, https://doi.org/10.1364/AO.45.008839.
Malingowski, J., D. Atkinson, J. Fochesatto, J. Cherry, and E. Stevens, 2014: An observational study of radiation temperature inversions in Fairbanks, Alaska. Polar Sci., 8, 24–39, https://doi.org/10.1016/j.polar.2014.01.002.
Mie, G., 1908: Beiträge zur Optik trüber Medien, speziell kolloidaler Metallösungen. Ann. Phys., 330, 377–445, https://doi.org/10.1002/andp.19083300302.
Przybylo, V. M., K. J. Sulia, C. G. Schmitt, and Z. J. Lebo, 2022: Classification of cloud particle imagery from aircraft platforms using convolutional neural networks. J. Atmos. Oceanic Technol., 39, 405–424, https://doi.org/10.1175/JTECH-D-21-0094.1.
Schmitt, C. G., M. Stuefer, A. J. Heymsfield, and C. K. Kim, 2013: The microphysical properties of ice fog measured in urban environments of Interior Alaska. J. Geophys. Res. Atmos., 118, 11 136–11 147, https://doi.org/10.1002/jgrd.50822.
Schnaiter, M., and Coauthors, 2016: Cloud chamber experiments on the origin of ice crystal complexity in cirrus clouds. Atmos. Chem. Phys., 16, 5091–5110, https://doi.org/10.5194/acp-16-5091-2016.
Serreze, M. C., J. D. Kahl, and R. C. Schnell, 1992: Low-level temperature inversions of the Eurasian Arctic and comparisons with Soviet drifting station data. J. Climate, 5, 615–629, https://doi.org/10.1175/1520-0442(1992)005<0615:LLTIOT>2.0.CO;2.
Shaw, G. E., 1983: On the aerosol particle size distribution spectrum in Alaskan air mass systems: Arctic haze and non-haze episodes. J. Atmos. Sci., 40, 1313–1320, https://doi.org/10.1175/1520-0469(1983)040<1313:OTAPSD>2.0.CO;2.
Thomson, W., 1871: On the equilibrium of vapour at a curved surface of liquid. Philos. Mag., 42, 448–452, https://doi.org/10.1080/14786447108640606.
Thuman, W. C., and E. Robinson, 1954: Studies of Alaskan ice-fog particles. J. Appl. Meteor., 11, 151–156, https://doi.org/10.1175/1520-0469(1954)011<0151:SOAIFP>2.0.CO;2.
Ulanowski, Z., E. Hirst, P. H. Kaye, and R. Greenaway, 2012: Retrieving the size of particles with rough and complex surfaces from two-dimensional scattering patterns. J. Quant. Spectrosc. Radiat. Trans., 113, 2457–2464, https://doi.org/10.1016/j.jqsrt.2012.06.019.
Ulanowski, Z., P. H. Kaye, E. Hirst, R. S. Greenaway, R. J. Cotton, E. Hesse, and C. T. Collier, 2014: Incidence of rough and irregular atmospheric ice particles from Small Ice Detector 3 measurements. Atmos. Chem. Phys., 14, 1649–1662, https://doi.org/10.5194/acp-14-1649-2014.
Vochezer, P., E. Järvinen, R. Wagner, P. Kupiszewski, T. Leisner, and M. Schnaiter, 2016: In situ characterization of mixed phase clouds using the Small Ice Detector and the Particle Phase Discriminator. Atmos. Meas. Tech., 9, 159–177, https://doi.org/10.5194/amt-9-159-2016.
WMO, 2017: International Cloud Atlas: Manual on the observation of clouds and other meteors. WMO-407, https://cloudatlas.wmo.int/en/ice-fog.html.
