Statistics of glinting clouds observed by DSCOVR and geostationary satellites

This study examines how frequently the specular reflection of sunlight—that is, sun glint—reveals the presence of ice crystals that maintain a steady horizontal orientation. The study analyzes data from the Earth Polychromatic Imaging Camera (EPIC) onboard the Deep Space Climate Observatory (DSCOVR) spacecraft and from collocated images taken by geostationary satellites. The analysis of spatio-temporal variations in glint frequency over vegetated land surfaces reveals that (a) year-to-year variations are modest with no clear trends; (b) glints typically occur 7%–8% more frequently than previously estimated; (c) glints are most frequently observed during the May-August period, and over Asia. The results also show that glint frequency drops for very high (>12–13 km) clouds but otherwise displays little sensitivity to geostationary satellite-provided cloud parameters, namely altitude, optical thickness, and particle size. This is because glints come from horizontal crystals near cloud tops whereas geostationary satellites characterize the entire cloudy column. This suggests that glint-free passive satellite observations are not well-suited for estimating the likelihood of horizontal ice crystals and underlines the importance of analyzing direct sun glint observations from satellite instruments such as EPIC.

1 Introduction

While most crystals in ice and mixed-phase clouds tumble through the air and are randomly oriented at any given moment, aerodynamic forces make large crystals of certain shapes—especially hexagonal plates—float at a systematic horizontal orientation. This orientation is stable because, as (Katz, 1998) put it, “…if the plate tilts, the wake of the leading edge partly shields the trailing edge from the flow, reducing the drag on it; the resulting torque restores the horizontal orientation … ”.

Although horizontally oriented particles (HOPs) often represent only a minority of particles in HOP-containing clouds (Bréon and Dubrulle, 2004; Noel and Chepfer, 2004), they are widely recognized to be important for several reasons. One reason is their radiative effect: Theoretical simulations show that HOPs and similar but randomly oriented particles can result in very different cloud albedos and surface irradiances (Takano and Liou, 1989); recently this difference was shown to be quite significant in a long-term ground-based observational dataset (Stillwell et al., 2019). HOPs also greatly influence polarization in instruments such as the Global Precipitation Measurement (GPM) Microwave Imager (GMI) and can cause significant errors in microwave retrievals of cloud ice water path (Gong and Wu, 2017). In addition, observing HOPs can provide information on ice cloud characteristics such as particle shape, the level of turbulence, or the likely presence of upward air motion, and even on the significance of radiative cooling in crystal growth (Zeng et al., 2021). Furthermore, horizontal orientation has a strong influence on the fall speed of ice crystals and hence on the lifetime of ice clouds (e.g., Heymsfield and Iaquinta, 2000; Westbrook, 2008; Zeng et al., 2021).

The horizontal orientation of crystals can most readily be detected by observing the strong specular reflection of light the multitude of such crystals cause. Several studies examined HOPs by analyzing vertically pointing ground-based lidar data (e.g., Sassen and Benson, 2001; He et al., 2021). Other studies examined Cloud-Aerosol Lidar with Orthogonal Polarisation (CALIOP) lidar observations from the first year of the CALIPSO (Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation) satellite mission, when the instrument pointed almost exactly vertically and was able to observe the specular reflection of its lidar pulses from HOPs (e.g., Noel and Chepfer, 2010; Zhou et al., 2012). Researchers even explored using statistical proxy techniques to extend CALIOP information on HOPs to later years, during which CALIOP was tilted sideways by 3° to avoid the saturation effects caused by strong HOP specular reflection (Kikuchi et al., 2021).

The strong specular reflection of sunlight by HOPs is called “sun glint” or “subsun”. This reflection has been observed via regular photography from the ground (e.g., Lynch et al., 1994) and onboard aircraft (e.g., Lynch and Livingston, 2001, p. 160; Können, 2017), It was aso observed by satellites on low-Earth orbit, for example when the POLDER (Polarization and Directionality of the Earth’s Reflectances) instrument provided multiangle polarized measurements (e.g., Chepfer et al., 1999; Bréon and Dubrulle, 2004; Noel and Chepfer, 2004).

Despite the multitude of observations, it is unclear how common HOPs are. As (Zhou et al., 2013) put it, “Previous studies have provided estimates of the fraction of clouds containing HOPs using different methods and data sets, but the uncertainties remain quite large.” It is clear, however, that HOPs are not just occasional components of ice clouds; for example, researchers estimated that HOPs were present in more than 40% of ice clouds and in more than 50% of ice cloud-containing POLDER pixels (Chepfer et al., 1999; Bréon and Dubrulle, 2004). Researchers also found that the prevalence of HOPs varies with conditions; most notably, HOPs are more abundant above −25 °C than at colder temperatures, although recent estimates suggested higher-than-previously-thought HOP occurrence even in very cold clouds (e.g., Kikuchi et al., 2021).

The importance of HOPs motivated fruitful efforts in a wide variety of research areas ranging from the analysis of remote sensing data (as discussed above) to the building of ground-based instruments for HOP observations (e.g., Borovoi et al., 2008; Forster et al., 2017) or to the development of theoretical methods for calculating the radiative properties of HOPs (e.g., Borovoi and Kustova, 2009; Zhou et al., 2012; Saito and Yang, 2019; Brath et al., 2020).

In recent years, several studies provided new insights about HOPs by analyzing images taken by the Earth Polychromatic Imaging Camera (EPIC) onboard the Deep Space Climate Observatory (DSCOVR) spacecraft (e.g., Marshak et al., 2018). DSCOVR orbits the Sun at the L1 Lagrangian point, which is located about 1.5 million km away from Earth, right near the straight line connecting the Sun to Earth. This position allows EPIC to constantly view almost the entire sunlit side of Earth. EPIC captures about 22 multispectral images per day between late April and early September, and about 13 images per day during the rest of the year, when the spacecraft can transmit less data as the ground communication station (located in Virginia, United States) peering toward the Sun can see DSCOVR for shorter periods during the shorter days. The spatial resolution of EPIC images is around 8 km at the image center and gets coarser toward the edge of the Earth disk.

Importantly for glint detection, EPIC uses a filter wheel to take images at ten wavelengths (ranging from 317 to 780 nm) during a roughly 7-min-long period. The small time-difference between EPIC channels helps glint detection because during the time between the capture of subsequent images, the Earth’s rotation brings a different geographical location to the so-called specular spot—the spot where EPIC can observe the specular reflection from horizontal surfaces. Therefore, the glint will appear at a different location for each EPIC wavelength—which means that the presence of glint can be identified if a location’s brightness jumps for the wavelength for which the location is near the specular spot.

The concept of using this jump for identifying glints—and the HOPs that cause them—was used in several studies (e.g., Marshak et al., 2017; Li et al., 2019) and in developing the glint mask for the EPIC operational sun glint product (Várnai et al., 2021). The various statistical analyses and case studies provided numerous insights, for example that (1) glints caused by HOPs are prevalent throughout the tropical latitudes where EPIC can provide glint observations, over both land and ocean (e.g., Marshak et al., 2017; Li et al., 2019); (2) EPIC observations at oxygen absorption bands can help distinguish glints from clouds and glints from smooth water surfaces, and the data over land is consistent with clouds occurring at mid-range altitudes such as 5–8 km (e.g., Marshak et al., 2017; Várnai et al., 2020a); (3) glint-caused reflectance excess is often stronger (but is limited for a narrower range of angles) for high clouds than for cloud-free but often rough ocean surfaces (Várnai et al., 2020a), which results in the average reflectance excess being three times stronger over land than over ocean (Várnai et al., 2020b); (4) reflectance excess is weaker at wavelengths where scattering or absorption by the air above clouds is strong (i.e., short wavelengths with strong Rayleigh scattering and oxygen absorption bands) (5) time-averaged reflectance excess maps (centered on the specular spot) can help validating geolocation accuracy even far from coastlines, and indicate gradual improvement in newer and newer versions of EPIC datasets, reaching better than one pixel accuracy for the current version (Kostinski et al., 2021); (6) glints from HOPs improve the sensitivity of cloud detection but cause sudden jumps in the retrieved cloud altitude and optical thickness values provided in the operational EPIC cloud product (Várnai et al., 2024).

Most recently, a paper in this special issue provides insights into the relevant physical processes (such as ice crystal wobbling) by analyzing the angular width of glints (Kostinski et al., 2025). It does this by examining the way the reflectance excess varies with the glint angle (δ), defined as the angle between the actual satellite view direction and the direction that would look straight into the specular reflection from a perfectly horizontal surface.

This study seeks to help better understand spatial and temporal variations in the prevalence of glints (and the HOPs that cause them) and seeks to gain further insights about the conditions that favor or discourage the presence of HOPs (and the glints they cause). Such knowledge is sought because earlier studies suggested that HOP prevalence may vary not only with temperature (e.g., Noel and Chepfer, 2010; Hirakata et al., 2014) but also with other relevant cloud and environmental parameters such as location and season (Chepfer et al., 1999; Várnai et al., 2020b; Kikuchi et al., 2021), vertical wind speed (Noel and Chepfer, 2010), or cloud type identified through cloud top pressure and optical depth (Cho et al., 2008). The prevalence of HOPs can also be suspected to depend on other parameters such as turbulence that can disrupt the horizontal orientation (Klett, 1995; Bréon and Dubrulle, 2004), supersaturation that can affect the habit of forming ice crystals (Bailey and Hallett, 2009), or the cloud particle effective radius that may indicate the abundance of large (>100 µm) particle sizes needed for horizontal orientation (Klett, 1995; Bréon and Dubrulle, 2004).

The outline of this paper is as follows. First, Section 2 describes the data used, then Section 3 examines temporal and spatial variations in glint frequency. Subsequently, Section 4 examines the relationships between cloud optical properties and glint prevalence. Finally, Section 5 presents a brief summary.

2 Dataset

As mentioned above, this study examines sun glints using data from the EPIC instrument onboard the DSCOVR spacecraft and from imagers onboard several geostationary satellites.

Unless noted otherwise, the study uses all EPIC images obtained from June 2015 (the start of DSCOVR science operations) until the end of 2024. Specifically, we use three EPIC Level 2 data products: The sun glint product (Várnai et al., 2021), the cloud product (Yang et al., 2019), and the composite cloud product (doi: 10.5067/EPIC/DSCOVR/L2_COMPOSITE_02).

The EPIC Level 2 glint product is used to identify the observed glints, as well as the underlying surface type and the glint angle. Unless noted otherwise, we use the glint mask of blue channel (443 nm) EPIC images. This channel is best suited for our study because, as EPIC uses a rotating spectral filter wheel to obtain images at its 10 wavelengths, blue images are taken at least 3 min earlier than the images of other wavelengths. Due to the rotation of Earth, this implies that the blue images are taken under significantly different sun-view geometry than the images of other wavelengths. As discussed in Marshak et al. (2017), this results in blue glints appearing at different locations than the glints of other wavelengths. As a result, the specular spots of blue images are observed at other wavelengths without cloud glint effects. This is important because the EPIC cloud product does not use 443 nm data over land, and so the EPIC cloud product can provide accurate data for the location of blue image glints. (Over land, EPIC uses 388 nm data for cloud detection, 680 nm data for estimating cloud optical thickness, and 680, 688, 764, and 780 nm data for retrieving cloud altitude (Yang et al., 2019). Since all non-blue images are taken at times and sun-view geometries that are close to at least one of these wavelengths, the glints at these wavelengths can significantly affect cloud retrieval results at the location of glints observed at non-blue EPIC wavelengths (Várnai et al., 2024).)

The EPIC Level 2 composite product is used to obtain additional information on the clouds that cause the glints observed by EPIC. This product contains cloud parameters for each pixel in EPIC images, but these cloud parameters are retrieved not based on EPIC data but based on collocated geostationary satellite data. For the tropical locations where EPIC can observe cloud glints, the composite product includes cloud parameters from the GOES-13 and -15, METEOSAT-7 and -10, MTSAT-2, and Himawari-8 satellites. Using this product helps because its data can benefit from (a) the different, glint-free view direction of geostationary satellites at the location of EPIC blue glints, (b) the thermal infrared observations that help cloud detection and cloud altitude measurements, (c) the near-infrared observations that allow retrievals of cloud phase and particle size, and (d) the higher spatial resolution (up to 2 km). We note that, to minimize the impact of temporal changes such as wind drift or cloud development and decay, we use geostationary satellite data only when it is obtained within 15 min of EPIC blue channel observations.

3 Temporal and spatial distribution of glints

As EPIC took observations for a more than 10-year long period, it is now possible to get an initial glimpse of interannual variability in glint occurrence—that is, in the occurrence of ice clouds between ≈25° North and ≈25° South latitudes (see Figure 2 in Marshak et al., 2017) containing horizontally oriented particles. For this, we examine the probability that pixels near the specular spot (with glint angles less than 0.3°) contain a blue glint detected in the EPIC operational glint product. Considering the 9 years (2016–2024) in which EPIC data is available for all seasons, Figure 1 reveals modest year-to-year variations in the order of about 15%, but no clear trends. While the lack of a clear trend may be due to the year-to-year random variations masking any systematic shift in cloud properties, it is reassuring that the results do not suggest a strong drift, as such drifts could have raised the possibility of gradual changes not only in cloud properties but also in instrument performance.

Figure 1

Figure 1. Annual mean fraction of pixels where a blue glint was detected, expressed in percents. The figure is based on all pixels over vegetated land where the blue image glint angle δ < 0.3°.

Perhaps most importantly, Figure 1 shows that glint occurrence in 2017 was a bit lower than typical; this is worth noting because 2017 is the year that was examined in several glint studies (e.g., Marshak et al., 2017; Kostinski et al., 2021; Várnai et al., 2020b; Várnai et al., 2024). This means that the findings of those studies slightly underestimate the typical long-term average amount of glints (and horizontally oriented particles, HOPs). In other words, glints and HOPs appear to be about 7%–8% more frequent than estimated in the mentioned earlier studies.

Finally, we note that the error bars in Figure 1 indicate statistical sampling uncertainty estimates that are calculated by assuming that each EPIC image is fully independent from the other images. Accordingly, images are randomly distributed into 10 batches, separate results are obtained for each batch, and the standard deviation from the 10 batches (divided by the square root of 10) is used as our uncertainty estimate. To improve the reliability of our estimates, we repeat this procedure 100 times, each time distributing EPIC images into 10 batches in a different random way. Finally, we set the error bars to be the mean of the 100 estimated standard error values. This process is used for all error bars in all figures, as well as for the uncertainty estimates in all tables.

The increasingly long EPIC time series also offers new opportunities for examining seasonal changes in glint occurrence. When doing this, we need to keep in mind that, as discussed in Marshak et al. (2017), EPIC views glints at different latitudes in different seasons. This is because EPIC can observe glint always near the center of the sunlit Earth disc, and the Earth’s tilt brings northern latitudes (up to 23°) to the center for half a year from March to September, and southern latitudes in the other half of the year. As a result, the specular spot drifts with the season—making seasonal and latitudinal variability inherently intertwined and causing glint observations to be available during the local wet season/summer.

Combining data for the 2015–2024 period, Figure 2 shows strong variations in monthly mean glint probabilities, with a clear maximum from May to August. This result is opposite to the findings in Várnai et al. (2020b), as that study found more glints in December-January-February (DJF) than in June-July-August (JJA). As Table 1 suggests, the opposite behavior can be attributed to a difference in considered surface types: While the 2020 study examined glints over all land surfaces, Figure 2 considers glints over vegetated land only. This becomes important because the table shows that, compared to DJF, JJA glint abundance increases over vegetated land but decreases over all land. The difference is that in JJA, as the specular spot moves north, vast desert areas (especially the Sahara Desert) will be included in the all-land dataset, but not in the vegetated land dataset. As deserts have few clouds and cloud glints (Várnai et al., 2021), including them reduces the rate of glint occurrence, which decreases JJA glint occurrence and reverses the annual cycle in glint observations.

Figure 2

Figure 2. Monthly mean fraction of pixels where a blue glint was detected for each month of the year. Panel (A) is based on all pixels observed during 2015–2024 over vegetated land where the blue image glint angle δ < 0.3°. To illustrate seasonal shifts in the location of specular spots suitable for glint observations (always near the EPIC image center), the three insets show the areas seen in some EPIC images taken on March 21, June 21, and 21 December 2016, respectively. The red curve in Panel (B) shows the same, but for 2017 only. Also for 2017 only, the blue curve shows the glint fraction (i.e., the red curve) normalized by the cloud fraction (which is shown by the gray curve).

Table 1

Table 1. Fraction of pixels with blue glints over land. The table is based on all EPIC images from 2017, considering areas where the glint angle δ is less than 0.3°.

We note that the prevalence of ice crystals in clouds appears somewhat higher according to the blue curve in Figure 2B than it was in earlier analyses of POLDER observations, which found horizontally oriented crystals in roughly half of ice-containing POLDER pixels (e.g., Chepfer et al., 1999; Bréon and Dubrulle, 2004). The difference likely comes from EPIC having a coarser spatial resolution and especially from always observing tropical regions during the local summer/wet season.

The blue curve exceeding 100% in Figure 2B indicates that in April and May, glints were detected more frequently than clouds. This likely comes from glints off thin ice clouds that were not detected by the EPIC operational cloud masking algorithm (thus further illustrating that glints can help improve the sensitivity of cloud detection (e.g., Várnai et al., 2024)) but glints off small water bodies such as lakes may also contribute.

Finally, the large number of EPIC observations collected over the past 10 years facilitates examining not only temporal and latitudinal, but also longitudinal variations. For example, one can divide the globe into three latitude zones covering different continents: The Americas between 30° and 180° West, Africa between 30° West and 50° East, and Asia-Australia between 50° and 180° East. (Since specular spots and hence glint observations are limited to areas between 25°North and 25° South, Europe and Antarctica are not considered.)

Table 2 compares glint occurrence over each continent during 2022–2024. (These years are convenient to use because the EPIC operational glint product includes longitude data since 2022.) The table indicates that in all 3 years, glints are least prevalent over Africa and most prevalent over Asia. (Although Australia is also included with Asia, its impact is small, as much of the continent is covered by desert, whereas Table 2 is only for vegetated land.) This distribution is consistent with MODIS observations (e.g., King et al., 2013) of ice cloud frequency over the three continents: For most of the year, at the latitude of the specular spot, MODIS reports the lowest and highest ice cloud covers over Africa and Asia, respectively, with the Americas in the middle (Figure 3). The sparsity of glints in May and June over vegetated parts of Africa may be related to the dominance of low (and thus warm) or thin clouds (e.g., Figures 8, 11 in King et al., 2013), which were found to host fewer HOPs in earlier studies (e.g., Noel and Chepfer, 2010; Hirakata et al., 2014). Naturally, having more ice clouds over Asia can help increase the number of glints from ice clouds there.

Table 2

Table 2. Fraction of pixels with blue glints over vegetated land. The table is based on all EPIC images from 2022 to 2024, considering areas where the glint angle δ is less than 0.3°.

Figure 3

Figure 3. MODIS monthly mean ice cloud fraction at the constantly changing location of EPIC specular spots, plotted for each month of the year. The figure is based on the MODIS Level 3 monthly cloud product and considers all 1° by 1° latitude-longitude grid points where the EPIC specular point occurred over vegetated land. The calculations considered all data from the 2022–2024 period.

4 Cloud properties at EPIC-detected glints

In addition to examining spatiotemporal variations in glint observations, it is also worth examining whether glint appearance is related to various cloud properties. The goal is to better understand how the presence of the glint-causing HOPs may be related to satellite-observable cloud properties. This effort follows the path of earlier studies that, as noted in the introduction, used a variety of satellite datasets to examine the impact of cloud properties such as temperature on HOP prevalence (e.g., Noel and Chepfer, 2010; Hirakata et al., 2014).

Due to their different view directions, the geostationary and EPIC UV and near-infrared observations used for cloud detection and cloud characterization are not affected by glints in the areas where EPIC observes blue glints. As a result, geostationary satellites and EPIC can both provide cloud properties with their usual accuracy at the locations where blue glints indicate the presence of HOPs.

Consequently, any relationship between glint occurrence and satellite-reported cloud properties could come not from the effects of glints (i.e., specular reflection), but from certain cloud conditions (such as altitude, optical thickness, or particle size) being associated with the presence of HOPs. For example, glints may appear more frequently in clouds at a certain altitude if the typical temperatures at that altitude are conducive for the formation plate-shaped ice crystals (which can often maintain horizontal orientation).

4.1 Cloud height

We first examine the relationship between the frequency of glint detections and cloud heights provided by EPIC and by collocated geostationary satellites. The operational EPIC cloud product includes two height retrievals; we use the one based on observations at the oxygen B-band. (The other height value is based on oxygen A-band observations and is therefore more sensitive to variations in the albedo of vegetated surfaces.) The collocated geostationary satellites provide cloud heights based on thermal infrared observations (Heidinger, 2012).

Examining data for small glint angles (δ < 0.3°) where glint detection is most effective, Figure 4A shows that EPIC and the geostationary satellites provide very different estimates for cloud heights at the location of blue image specular spots. One important reason for the differences is that while the thermal infrared signal used by geostationary satellites comes from the cloud tops, EPIC provides its cloud heights based on oxygen absorption information that often comes from deep inside clouds. Accordingly, the EPIC dataset is called cloud effective height (and not cloud top height).

Figure 4

Figure 4. Relationship between satellite-retrieved cloud height and either the total number of pixels (A) or the fraction of pixels where a blue glint was detected (B). The figure is based on all pixels over vegetated land where the blue image glint angle δ < 0.3°. The orange and black lines are for cloud height values provided by EPIC or geostationary satellites, respectively. Panels (A) and (B) are based on all EPIC and collocated geostationary data collected over vegetated land. The black (geostationary) curves are based on data from June 2015 to June 2018; the orange (EPIC) curves are based on data from 2017 only. In Panel (A), the main plot is for all considered pixels; the inset is for blue glint pixels only, with the vertical scale extending to 1,200. Panel (C) shows the fraction of cases when ice clouds were detected by CloudSat and CALIPSO from 2007 to 2010 during June-July-August. Panel (D) also shows the fraction of cases when ice clouds were detected by CloudSat and CALIPSO from 2007 to 2010, but for December-January-February. In both panels (C) and (D), the horizontal dashed lines show 3 km altitude, and the vertical solid lines mark 25° North or South, the poleward limit of EPIC glint observations during these seasons. Panels (C) and (D) are custom-tailored plots based on Figures 4F,H in Hong and Liu (2015). © American Meteorological Society. Used with permission.

However, Figure 4B shows that the two datasets are in rough agreement on the probability of blue glint detection for various cloud altitudes. They both feature a modest bump centered around 9 or 11 km, a sharp drop at higher altitudes, and a fairly constant probability at low altitudes (with some sampling noise). These features can be interpreted as follows.

First, the bump around 9 or 11 km altitude is probably caused by the increased presence of crystal shapes that can maintain a steady horizontal orientation (e.g., hexagonal plates and columns), which were shown to form mainly at freezing but not too cold temperatures (e.g., Noel and Chepfer, 2010; Hirakata et al., 2014).

Second, the drop at higher altitudes is probably due to the very cold temperatures there. This is because very cold temperatures (especially below −30 °C) foster the formation of crystals (such as needles) that tumble randomly and cannot provide the steady horizontal surfaces needed for strong glint reflection (e.g., Bailey and Hallett, 2009).

Finally, the relatively constant glint probability at lower altitudes could, in principle, come from two reasons. Finding the right reason is especially important for altitudes below 3 km, where, as shown in Figures 4C,D that are based on CloudSat and CALIPSO observations (see Hong and Liu, 2015), we cannot expect any ice clouds (and hence glint-causing HOPs) at the tropical latitudes of EPIC glint observations. The first option is erroneous glint detections, for example if a pixel was much brighter in the blue image than in the red image that was captured a few minutes later because clouds dissipated or moved away. Figure 5, however, disproves this hypothesis. Specifically, the figure shows that even for low reported cloud altitudes, the probability of glint detection drops sharply with increasing glint angle—which would not be the case if glint detections were caused not by the presence of glints but by clouds drifting with the wind or by any other factor independent of glint angle. Therefore, we must turn to the other option—namely that cloud height retrievals by passive sensors often do not provide the altitude where glints are formed. Instead, they often return the height of optically thick low-altitude clouds even if there is a thin, glint-causing ice cloud above them. This implies that glint observations may help future studies in detecting the presence of thin ice clouds that are not identified in current retrievals based on passive satellite observations.

Figure 5

Figure 5. Fraction of cloudy pixels over vegetated land where a blue glint was detected, plotted as a function of glint angle (δ). The figure is based on all EPIC images from 2017. Cloud altitude is from collocated geostationary satellite data.

4.2 Cloud optical properties

Next, we consider the impact of cloud optical properties on the probability of glint detection. Figure 6A shows that in the tropical regions around EPIC specular spots, EPIC and geostationary satellites provide differing distributions of cloud optical thickness. The differences come from using different (i) ice-cloud particle-shape assumptions, (ii) forward radiative-transfer models, and (iii) retrieval methods, and also from (iv) EPIC having a coarser spatial resolution, which can result in a stronger plane parallel bias. (This bias comes from the combination of subpixel variability and the nonlinear relationship between observed reflectance and cloud optical thickness (e.g., Zuidema and Evans, 1998; Várnai and Marshak, 2001)).

Figure 6

Figure 6. Relationship between satellite-retrieved cloud optical thickness and either the number of cloudy pixels (A) or the fraction of cloudy pixels where blue glint was detected (B). The figure is based on data provided by EPIC and collocated geostationary satellites for pixels where the EPIC glint angle δ < 0.5° over vegetated land. The orange and black curves are based on all available data from 2017 and from the June 2015-June 2018 period, respectively.

In turn, Figure 6B shows that—except for a peak for thin (and probably heterogeneous) clouds for EPIC—both satellite datasets show a roughly constant probability of glint detection for a wide range of optical thicknesses. This is because glints are single scattering phenomena caused by the specular reflection of direct sunlight near cloud tops (where direct sunlight is abundant)—and it does not affect glints whether or not there is a thick cloud (illuminated by diffuse sunlight) below the glint-producing top layer. Therefore, the total optical thickness does not correlate with glint prevalence.

Next, Figure 7A shows the distribution of cloud particle size retrieved by geostationary satellites near the location of EPIC specular spots. Lacking suitable spectral channels, EPIC does not provide particle size retrievals but—as Figure 7A shows—geostationary satellites provide two of them: One for ice clouds only and one for the total cloud population. The coincidence of the two curves for large particle sizes in Figure 7A indicates that effective radii larger than about 25 µm are retrieved only for ice clouds.

Figure 7

Figure 7. Panels (A) and (B) Relationship between satellite-retrieved cloud particle size and either the number of cloudy pixels (A) or the fraction of cloudy pixels where blue glint was detected (B). These panels are based on geostationary satellite images collocated with EPIC, considering all data from the June 2015-June 2018 period over vegetated land where the EPIC glint angle δ < 0.5°. Particle size is provided only by geostationary satellites but is available separately for all clouds and for ice clouds only. Panel (C) Mean cloud effective radius [µm] as a function of cloud altitude and optical thickness, based on CloudSat and CALIPSO observations. This panel is a custom-tailored plot based on Figure 10B in Hong and Liu (2015). © American Meteorological Society. Used with permission.

More importantly, though, Figure 7B reveals that the probability of glint detection does not vary systematically with retrieved particle size. This can be attributed to two factors. First, glint is produced by HOPs, which constitute only a small fraction of the total particle population even inside glint-producing clouds (Bréon and Dubrulle, 2004; Noel and Chepfer, 2004)—whereas geostationary satellite retrievals provide effective radii values for the entire particle population. Second, glints are produced near cloud tops (in the layer where direct sunlight is still abundant), whereas the signal for geostationary satellite retrievals of particle size come from a thick layer extending deep inside clouds (Platnick 2000)—and particle size may be quite different deep inside clouds than at the glint-producing top layer (e.g., van Diedenhoven, et al., 2016; Khatri et al., 2018).

Finally, we note that although Figure 7C shows that particle size varies systematically with both cloud altitude and optical thickness, the lack of correlation between glint occurrence and particle size (Figure 7B) is consistent with our earlier results, as glint occurrence is not correlated with either cloud altitude or cloud optical thickness (Figures 4B, 6B).

5 Summary

To help better understand the radiative and microphysical properties and processes of ice clouds, this study searched for information on the orientation of ice crystals in clouds. Specifically, the paper sought insights about the prevalence of sun glints created by the specular reflection of sunlight from clouds that contain horizontally oriented ice crystals.

With this goal in mind, the paper examined statistics of glints identified by the operational sun glint product of the EPIC instrument onboard the DSCOVR spacecraft (https://epic.gsfc.nasa.gov/science/products/glint). The study first examined spatial, seasonal, and interannual variations in glint detection frequency, and then it also examined how this frequency varies with cloud parameters such as altitude, optical thickness, and particle size. In addition to using EPIC data, the analysis also considered collocated observations by geostationary satellites (GOES-13 and -15, METEOSAT-7 and -10, MTSAT-2, and Himawari-8) that observed the glint-causing clouds from different, glint-free view directions allowing high-quality cloud property retrievals. The study focused on glints over vegetated land surfaces, where glint detection is most reliable and frequent.

The results revealed modest year-to-year variations but no statistically significant overall trend over the available 9 full years of EPIC data (2016–2024). The results also indicated the frequency of glint detection was 7%–8% lower in 2017 than during the 2016–2024 period as a whole. Since several earlier studies analyzed glints only from 2017, this finding implies that cloud glints (and the horizontal ice crystals causing them) are typically 7%–8% more frequent than suggested by the earlier studies.

Seasonal variability was found to be much stronger than interannual variability, with a distinct peak in glint frequency from May to August. This seasonal variability is the opposite of earlier results which, for all land areas, indicated a maximum glint frequency around January. The apparent contradiction was explained by seasonal shifts in the latitude and underlying surface type of EPIC glint observations (e.g., Marshak et al., 2017). For example, the dearth of glints over the (less cloudy) Sahara Desert suppressed the May-August glint frequency for the “all-land” results of earlier analyses, but was not included and thus had no such suppressing impact in the current study of vegetated surfaces.

Even when deserts were excluded and only vegetated land areas were considered, the results indicated that glint frequency is highest over Asia and lowest over Africa, with the Americas in between. This finding was consistent with MODIS observations of vegetated areas at EPIC glint latitudes, which showed the most ice clouds over Asia and the least over Africa.

The combination of EPIC cloud and glint products and the combination of DSCOVR and geostationary satellite data both showed that the frequency of glint detection drops for cloud altitudes above 11 km, presumably because very high clouds are too cold for the formation of larger ice crystals that can maintain a horizontal orientation. On the other hand, glint frequency remained steady for lower altitudes, including even at the lowest altitudes where the temperature was above freezing. This could be attributed to EPIC and geostationary satellites reporting the altitude of optically thick low clouds and missing thin ice clouds above them that may have caused glints. Also, except for clouds with EPIC-reported optical thicknesses between 1 and 3, glint frequency was similar for all cloud optical thicknesses reported by either EPIC or geostationary satellites. The likely explanation is that while glints are a single-scattering phenomena that occur only near cloud tops, passive satellite instruments are much more sensitive to the total optical thickness that may include underlying cloud layers irrelevant to glint formation. Finally, glint frequency was also similar for all cloud particle sizes reported by geostationary satellites. As for optical thickness, this is because geostationary satellites report mainly on a thicker cloud column extending well below the glint-causing horizontal ice crystals—plus the data from these satellites characterizes the entire particle population and not only the small fraction of crystals that maintain a horizontal orientation and thus create glints.

Overall, these findings suggest that collocated glint-free passive satellite observations are not well-suited for indicating the conditions in glint-causing clouds, as they are sensitive to the entire atmospheric column and not just the glint-causing cloud layers and ice particles. This likely precludes using such satellite data as a proxy to estimate the likelihood of horizontally oriented ice crystals and underlines the need for direct glint observations that can identify these crystals. While the EPIC instrument has proven well-suited for providing such data, other satellite instruments can also help wherever their sun-view geometry allows glint observations. For example, the Hyper-Angular Rainbow Polarimeter #2 (HARP-2) on the PACE satellite can provide glint data including even polarization information, but imagers on numerous geostationary and polar orbiter satellites—such as the Advanced Baseline Imager (ABI) or MODIS—can also reveal statistical information on glints and glint-causing horizontal ice crystals even if individual glints cannot be identified with confidence. Different passive satellites offer different benefits for cloud glint analysis, and the choice of instrument should depend on the study goals. For example, polar orbiters can statistically characterize glints at high latitudes, EPIC can identify individual near-noon glints throughout the tropics, and geostationary satellites can help reveal diurnal variations if morning glints observed to the east of a satellite are compared to afternoon glints observed to the west of it. Thus, glint data from a multitude of passive instruments can complement the information on horizontal ice crystals from active instruments such as CALIOP and can help, for example, to characterize any diurnal variations in the frequency of horizontally oriented ice crystals.

Data availability statement

Publicly available datasets were analyzed in this study. This data can be found here: https://asdc.larc.nasa.gov/project/DSCOVR/DSCOVR_EPIC_L2_GLINT_01.

Author contributions

TV: Conceptualization, Data curation, Formal Analysis, Funding acquisition, Investigation, Methodology, Software, Writing – original draft, Writing – review and editing. AM: Formal Analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Visualization, Writing – review and editing. AK: Conceptualization, Formal Analysis, Funding acquisition, Investigation, Methodology, Visualization, Writing – review and editing.

Funding

The author(s) declared that financial support was received for this work and/or its publication. This work was supported in part by the NASA DSCOVR project, the NASA ACMAP program, and the National Science Foundation grant AGS-2217182.

Acknowledgements

We are grateful to Richard Eckman and to members of the NASA DSCOVR team for their help and support.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The author AM declared that they were an editorial board member of Frontiers at the time of submission. This had no impact on the peer review process and the final decision.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

References

Bailey, M. P., and Hallett, J. (2009). A comprehensive habit diagram for atmospheric ice crystals: confirmation from the laboratory, AIRS II, and other field studies. J. Atmos. Sci. 66, 2888–2899. doi:10.1175/2009JAS2883.1

CrossRef Full Text | Google Scholar

Borovoi, A., and Kustova, N. V. (2009). Specular scattering by preferably oriented ice crystals. Appl. Opt. 48, 3878–3885. doi:10.1364/ao.48.003878

PubMed Abstract | CrossRef Full Text | Google Scholar

Borovoi, A., Galileiskii, V., Morozov, A., and Cohen, A. (2008). Detection of ice crystal particles preferably oriented in the atmosphere by use of the specular component of scattered light. Opt. Express 16, 7625–7633. doi:10.1364/oe.16.007625

PubMed Abstract | CrossRef Full Text | Google Scholar

Brath, M., Ekelund, R., Eriksson, P., Lemke, O., and Buehler, S. A. (2020). Microwave and submillimeter wave scattering of oriented ice particles. Atmos. Meas. Tech. 13, 2309–2333. doi:10.5194/amt-13-2309-2020

CrossRef Full Text | Google Scholar

Bréon, F.-M., and Dubrulle, B. (2004). Horizontally oriented plates in clouds. J. Atm. Sci. 61, 2888–2898. doi:10.1175/JAS-3309.1

CrossRef Full Text | Google Scholar

Chepfer, H., Brogniez, G., Goloub, P., Francois, M. B., and Flamant, P. H. (1999). Observations of horizontally oriented ice crystals in cirrus clouds with POLDER-1/ADEOS-1. J. Quant. Spectrosc. Radiat. Transf. 63, 521–543. doi:10.1016/S0022-4073(99)00036-9

CrossRef Full Text | Google Scholar

Cho, H.-M., Yang, P., Kattawar, G. W., Nasiri, S. L., Hu, Y., Minnis, P., et al. (2008). Depolarization ratio and attenuated backscatter for nine cloud types: analyses based on collocated CALIPSO lidar and MODIS measurements. Opt. Express 16, 3931–3948. doi:10.1364/OE.16.003931

PubMed Abstract | CrossRef Full Text | Google Scholar

Forster, L., Seefeldner, M., Wiegner, M., and Mayer, B. (2017). Ice crystal characterization in cirrus clouds: a sun-tracking camera system and automated detection algorithm for halo displays. Atmos. Meas. Tech. 10, 2499–2516. doi:10.5194/amt-10-2499-2017

CrossRef Full Text | Google Scholar

Gong, J., and Wu, D. L. (2017). Microphysical properties of frozen particles inferred from global precipitation measurement (GPM) microwave imager (GMI) polarimetric measurements. Atmos. Chem. Phys. 17, 2741–2757. doi:10.5194/acp-17-2741-2017

PubMed Abstract | CrossRef Full Text | Google Scholar

He, Y., Liu, F., Yin, Z., Zhang, Y., Zhan, Y., and Yi, F. (2021). Horizontally oriented ice crystals observed by the synergy of Zenith- and slant-pointed polarization lidar over wuhan (30.5°N, 114.4°E), China. J. Quant. Spect. Radiat. Transf. 268, 107626. doi:10.1016/j.jqsrt.2021.107626

CrossRef Full Text | Google Scholar

Heidinger, A. (2012). ABI cloud height algorithm theoretical basis document. NOAA/NESDIS/STAR. Version 3.0.

Google Scholar

Heymsfield, A. J., and Iaquinta, J. (2000). Cirrus crystal terminal velocities. J. Atmos. Sci. 57, 916–938. doi:10.1175/1520-0469(2000)057,0916:CCTV.2.0.CO;2

CrossRef Full Text | Google Scholar

Hirakata, M., Okamoto, H., Hagihara, Y., Hayasaka, T., and Oki, R. (2014). Comparison of global and seasonal characteristics of cloud phase and horizontal ice plates derived from CALIPSO with MODIS and ECMWF. J. Atmos. Ocean. Technol. 31, 2114–2130. doi:10.1175/JTECH-D-13-00245.1

CrossRef Full Text | Google Scholar

Hong, Y., and Liu, G. (2015). The characteristics of ice cloud properties derived from CloudSat and CALIPSO measurements. J. Clim. 28, 3880–3901. doi:10.1175/JCLI-D-14-00666.1

CrossRef Full Text | Google Scholar

Katz, J. I. (1998). Subsuns and low reynolds number flow. J. Atmos. Sci. 55, 3358–3362. doi:10.1175/1520-0469

CrossRef Full Text | Google Scholar

Khatri, P., Iwabuchi, H., and Saito, M. (2018). Vertical profiles of ice cloud microphysical properties and their impacts on cloud retrieval using thermal infrared measurements. J. Geophys. Res. 123, 5301–5319. doi:10.1029/2017JD028165

CrossRef Full Text | Google Scholar

Kikuchi, M., Okamoto, H., and Sato, K. (2021). A climatological view of horizontal ice plates in clouds: findings from nadir and off-nadir CALIPSO observations. J. Geophys. Res. 126, e2020JD033562. doi:10.1029/2020JD033562

CrossRef Full Text | Google Scholar

King, M. D., Platnick, S., Menzel, W. P., Ackerman, S. A., and Hubanks, P. A. (2013). Spatial and temporal distribution of clouds observed by MODIS onboard the terra and aqua satellites. IEEE Trans. Geosci. Remote Sens. 51, 3826–3852. doi:10.1109/TGRS.2012.2227333

CrossRef Full Text | Google Scholar

Klett, J. D. (1995). Orientation model for particles in turbulence. J. Atmos. Sci. 52, 2276–2285. doi:10.1175/1520-0469(1995)052<2276:OMFPIT>2.0.CO;2

CrossRef Full Text | Google Scholar

Können, G. P. (2017). Rainbows, Halos, coronas and glories: beautiful sources of information. Bull. Amer. Meteor. Soc. 98, 485–494. doi:10.1175/BAMS-D-16-0014.1

CrossRef Full Text | Google Scholar

Kostinski, A., Marshak, A., and Várnai, T. (2021). Deep space observations of terrestrial glitter. Earth Space Sci 8, e2020EA001521. doi:10.1029/2020EA001521

CrossRef Full Text | Google Scholar

Kostinski, A., Marshak, A., and Várnai, T. (2025). Constraining orientation statistics of ice crystals in clouds with observations from deep space. Front. Remote Sens. This Issue. doi:10.3389/frsen.2025.1705235

CrossRef Full Text | Google Scholar

Li, J.-Z., Fan, S., Kopparla, P., Liu, C., Jiang, J. H., Natraj, V., et al. (2019). Study of terrestrial glints based on DSCOVR observations. Earth Space Sci. 6, 166–173. doi:10.1029/2018EA000509

CrossRef Full Text | Google Scholar

Lynch, D. K., and Livingston, W. (2001). Color and light in nature. 2nd edition Cambridge, U. K: Cambridge Univ. Press.

Google Scholar

Lynch, D. K., Gedzelman, S. D., and Fraser, A. B. (1994). Subsuns, Bottlinger’s rings and elliptical halos. Appl. Opt. 33, 4580–4589. doi:10.1364/AO.33.004580

PubMed Abstract | CrossRef Full Text | Google Scholar

Marshak, A., Várnai, T., and Kostinski, A. (2017). Terrestrial glint seen from deep space: oriented ice crystals detected from the lagrangian point. Geophys. Res. Lett. 44, 5197–5202. doi:10.1002/2017GL073248

CrossRef Full Text | Google Scholar

Marshak, A., Herman, J., Szabo, A., Blank, K., Cede, A., Carn, S., et al. (2018). Earth observations from DSCOVR EPIC instrument. Bull. Amer. Meteorol. Soc. 99, 1829–1850. doi:10.1175/bams-d-17-0223.1

PubMed Abstract | CrossRef Full Text | Google Scholar

Noel, V., and Chepfer, H. (2004). Study of ice crystal orientation in cirrus clouds based on satellite polarized radiance measurements. J. Atmos. Sci. 61, 2073–2081. doi:10.1175/1520-0469(2004)061<2073:SOICOI>2.0.CO;2

CrossRef Full Text | Google Scholar

Noel, V., and Chepfer, H. (2010). A global view of horizontally oriented crystals in ice clouds from cloud-aerosol lidar and infrared pathfinder satellite observation (CALIPSO). J. Geophys. Res. 115. doi:10.1029/2009JD012365

CrossRef Full Text | Google Scholar

Platnick, S. (2000). Vertical photon transport in cloud remote sensing problems. J. Geophys. Res. 105, 22919–22935. doi:10.1029/2000JD900333

CrossRef Full Text | Google Scholar

Saito, M., and Yang, P. (2019). Oriented ice crystals: a single-scattering property database for applications to lidar and optical phenomenon simulations. J. Atmos. Sci. 76, 2635–2652. doi:10.1175/JAS-D-19-0031.1

CrossRef Full Text | Google Scholar

Sassen, K., and Benson, S. (2001). A midlatitude cirrus cloud climatology from the facility for atmospheric remote sensing. Part II: microphysical properties derived from lidar depolarization. J. Atmos. Sci. 58, 2103–2112. doi:10.1175/1520-0469(2001)058<2103:AMCCCF>2.0.CO;2

CrossRef Full Text | Google Scholar

Stillwell, R. A., Neely, R. R. I. I. I., Thayer, J. P., Walden, V. P., Shupe, M. D., and Miller, N. B. (2019). Radiative influence of horizontally oriented ice crystals over summit, Greenland. J. Geophys. Res. 124, 12141–12156. doi:10.1029/2018JD028963

CrossRef Full Text | Google Scholar

Takano, Y., and Liou, K. N. (1989). Solar radiative transfer in cirrus clouds. Part II: theory and computation of multiple scattering in an anisotropic medium. J. Atmos. Sci. 46, 20–36. doi:10.1175/1520-0469(1989)046<0020:SRTICC>2.0.CO;2

CrossRef Full Text | Google Scholar

van Diedenhoven, B., Fridlind, A. M., Cairns, B., Ackerman, A. S., and Yorks, J. E. (2016). Vertical variation of ice particle size in convective cloud tops. Geophys. Res. Lett. 43, 4586–4593. doi:10.1002/2016GL068548

PubMed Abstract | CrossRef Full Text | Google Scholar

Várnai, T., and Marshak, A. (2001). Statistical analysis of the uncertainties in cloud optical depth retrievals caused by three-dimensional radiative effects. J. Atmos. Sci. 58, 1540–1548. doi:10.1175/1520-0469(2001)058<1540:SAOTUI>2.0.CO;2

CrossRef Full Text | Google Scholar

Várnai, T., Kostinski, A., and Marshak, A. (2020a). Deep space observations of Sun glints from marine ice clouds. IEEE Geosci. Rem. Sens. Lett. 17, 735–739. doi:10.1109/LGRS.2019.2930866

CrossRef Full Text | Google Scholar

Várnai, T., Marshak, A., and Kostinski, A. (2020b). Deep space observations of cloud glints: spectral and seasonal dependence. IEEE Geosci. Rem. Sens. Lett. doi:10.1109/LGRS.2020.3040144

CrossRef Full Text | Google Scholar

Várnai, T., Marshak, A., and Kostinski, A. (2021). Operational detection of sun glints in DSCOVR EPIC images. Front. Remote Sens. 2, 777806. doi:10.3389/frsen.2021.777806

CrossRef Full Text | Google Scholar

Várnai, T., Marshak, A., Kostinski, A., Yang, Y., and Zhou, Y. (2024). Impact of sun glint off ice clouds on DSCOVR EPIC cloud products. IEEE Trans. Geosci. Remote Sens. 62, 1–11. doi:10.1109/TGRS.2024.3400253

CrossRef Full Text | Google Scholar

Westbrook, C. D. (2008). The fall speeds of Sub-100 micron ice crystals. Quart. J. Roy. Meteor. Soc. 134, 1243–1251. doi:10.1002/qj.290

CrossRef Full Text | Google Scholar

Yang, Y., Meyer, K., Wind, G., Zhou, Y., Marshak, A., Platnick, S., et al. (2019). Cloud products from the earth polychromatic imaging camera (EPIC): Algorithms and initial evaluation. Atmos. Meas. Techn. 12, 2019–2031. doi:10.5194/amt-12-2019-2019

PubMed Abstract | CrossRef Full Text | Google Scholar

Zeng, X., Gong, J., Li, X., and Wu, D. L. (2021). Modeling the radiative effect on microphysics in cirrus clouds against satellite observations. J. Geophys. Res. 126, e2020JD033923. doi:10.1029/2020JD033923

PubMed Abstract | CrossRef Full Text | Google Scholar

Zhou, C., Yang, P., Dessler, A. E., Hu, Y., and Baum, B. A. (2012). Study of horizontally oriented ice crystals with CALIPSO observations and comparison with monte carlo radiative transfer simulations. J. Appl. Meteor. Climatol. 51, 1426–1439. doi:10.1175/JAMC-D-11-0265.1

CrossRef Full Text | Google Scholar

Zhou, C., Yang, P., Dessler, A. E., and Liang, F. (2013). Statistical properties of horizontally oriented plates in optically thick clouds from satellite observations. IEEE Geosci. Remote Sens. Lett. 10, 986–990. doi:10.1109/LGRS.2012.2227451

CrossRef Full Text | Google Scholar

Zuidema, P., and Evans, K. F. (1998). On the validity of the independent pixel approximation for boundary layer clouds observed during ASTEX. J. Geophys. Res. 103, 6059–6074. doi:10.1029/98JD00080

CrossRef Full Text | Google Scholar