It is known both empirically and from analyses of cloud observations that cloudiness changes over the course of the day even in the absence of strong dynamic forcing. Over land, low cloud cover tends to increase after a minimum in the early morning, reaching a peak around noon, and then decreasing later in the afternoon. Conversely, low cloud cover over the ocean exhibits an opposite cycle, with the highest cloud fraction during the night (Cairns, 1995) and minimum around noon (Delgado-Bonal et al., 2020a; Delgado-Bonal et al., 2021).

Besides the cloud fraction, diurnal variation is cloud height are crucial for understanding the global energy balance because high and low clouds may have different impacts on trapping thermal infrared radiation. Cloud height exhibits a minimum around midday for low clouds and a steady increase from morning to evening for high clouds. These general characteristics have continental/maritime contrasts, with the amplitudes of the diurnal cycles being weaker over ocean than over land (Yang et al., 2013; Delgado-Bonal et al., 2022).

Sun-synchronous satellites provide fixed-time views of the planet, while geostationary satellite (GEO) systems provide a combination of fixed spatial coverage and high-frequency sampling, addressing some limitations of the sun-synchronous satellites with restricted viewing areas. A middle-ground compromise for this is offered by the Earth Polychromatic Imaging Camera (EPIC) aboard DSCOVR, which provides a unique perspective from a distance of 1.5 million km, with the global and detailed views of the planetary cloud properties of the illuminated part of the planet being resolved in either one (boreal winter) or two hours (boreal summer) (Marshak et al., 2018).

Another approach to studying clouds exploits the combination of different types of satellite observations to achieve broader observational coverage and better temporal sampling rates. Characteristic cloud products of this kind include those obtained from the ISCCP (Rossow and Schiffer, 1991) and CERES project (Wielicki et al., 1996). Cloud property aggregations in these satellite products create almost seamless datasets by combining information from the polar and GEO satellites with measurements acquired at different temporal spans and viewing angles. DSCOVR cloud products, on the other hand, provide lower-resolution views of only the sunlit part of the planet even though they have the advantage of consistency from being based on a single sensor (EPIC); this avoids errors arising from stitching together the views from multiple sensors with different viewing angles, capabilities, and characteristics.

Clouds play a major role in controlling the global radiation budget and influencing climate on a global scale as they cover about two-thirds of the Earth’s surface (Stubenrauch et al., 2013; King et al., 2013). The diurnal variations in cloud properties have important influences on cloud radiative impacts (Bergman and Salby, 1997); through their close association with precipitation, they also influence the hydrological cycle of the planet. Since the diurnal cycles of cloud properties predicted by general circulation models are highly inconsistent (Yin and Porporato, 2017), improving our understanding of these cycles may help improve cloud simulations and their impacts on climate models (Nakajima and King, 1990).

With future clouds being the largest single source of uncertainties in climate analysis and prediction (Stephens, 2005; Baker and Peter, 2008; Boucher et al., 2013), the manner in which cloud optical thickness (COT) may change in the future, including changes in the diurnal cycles, constitutes important aspects of the cloud radiative effect and feedback (Melnikova et al., 2000; Garrett and Zhao, 2006; Zelinka et al., 2016; Chen et al., 2018). Although there is increasing literature on the diurnal variations of clouds (Chen et al., 2018; Yang et al., 2020), these reports mostly focus on the cloud fraction (Cairns, 1995; Burleyson et al., 2013; Burleyson and Yuter, 2015). We have previously studied cloud fraction as well as cloud height diurnal variations based on EPIC data and now focus on the COT, which has not been investigated much (Li et al., 2022).

Regional diurnal cycles of cloud properties have been observed with GEO satellites, such as GOES (Greenwald and Christopher, 1999; Painemal et al., 2013) and Himawari-8 (Shang et al., 2018), but EPIC is the only sensor capable of covering the entire sunlit side of the planet at once. The importance of cloud property diurnal cycles becomes apparent when these results are compared with the fixed times of the polar orbiters. In this regard, Shang et al. (2018) showed that MODIS underestimates the daytime average cloud fraction in China by approximately 4% on an annual scale with respect to the Himawari-8 measurements. Owing to the fixed MODIS overpass times, the difference can increase to up to 21% between the surface-observed and MODIS space-observed cloud fractions over specific regions on the monthly scale (Wang and Zhao, 2017).

The present study aims to provide a global view of the EPIC COT diurnal changes, thus offering a complete view of the diurnal variations of one of the major cloud properties controlling reflected sunlight. The following section provides an overview of the algorithms used and outlines our methodology. The results provide detailed insights into the diurnal cycles of COT from EPIC on the global and regional scales, detailing the behaviors for different cloud types and analyzing the diurnal correlations among different cloud properties.

The charge-coupled device (CCD) array of EPIC comprising 2,048 × 2,048 pixels and corresponding to a spatial resolution of ∼8 km at nadir provides the reflectance values across 10 channels spanning from the ultraviolet (318, 325, 340, and 388 nm) and visible (443, 551, 680, and 688 nm) to near-infrared (764 and 780 nm) parts of the spectrum. The cloud product algorithm utilizes a surface-dependent threshold method for cloud masking by employing reflectances at 388, 680, and 780 nm along with the O₂ A- and B-band channels. These cloud fraction statistics match closely with those of GEO and low Earth orbit (LEO) systems when collocated, exhibiting global differences of only 1.5% (Yang et al., 2019).

The EPIC cloud mask algorithm includes confidence level flags for the cloud detection outcomes, with fewer than 3% of the pixels marked as low-confidence cloudy. EPIC derives the cloud heights from the O₂ A-band (780 and 764 nm) and B-band (680 and 688 nm) observations. Owing to photon penetration of the clouds, the radiance values measured by EPIC’s A and B bands depend on not only the cloud top height but also the cloud extinction profile. Therefore, EPIC’s retrievals represent the “effective” cloud height and pressure, which differ from both the geometrical cloud top heights of active observations such as lidar and effective heights inferred from thermal IR observations (Yang et al., 2013).

The low and high cloud classes are defined by taking into account the apparent predominant cloud thermodynamic phase (liquid or ice) in the Level 2 datasets. EPIC determines the cloud thermodynamic phase from the cloud effective temperature, which is inferred from the aforementioned cloud effective pressure (Yang et al., 2019). If the effective temperature is below 240 K, the cloudy pixels are labeled as ice, and if the temperature is above 260 K, the pixels are labeled as liquid; pixels with temperatures between these values are marked as unknown to prevent potential misclassifications. This methodology has been validated against MODIS (Meyer et al., 2016), showing thermodynamic phase agreement for approximately 77% of the pixels. Around 21% of the pixels are categorized as unknown, and the misclassifications are limited to about 2% compared to MODIS. We classify clouds below 3 km with a thermodynamic cloud phase of liquid as low clouds, and all clouds above 6 km with a thermodynamic phase of ice as high clouds. Although this filtering is not applicable to all the cloudy pixels in our dataset, it helps us avoid distortions of the daytime cycles due to misclassified pixels. About 10% of the EPIC cloud mask is cataloged as high clouds, and the diurnal cycles are studied separately for the low and high clouds.

Because the EPIC channels are not sensitive to particle sizes in the shortwave or midwave infrared range, the EPIC COT retrieval employs a single-channel approach. This method is similar to the approaches used by the International Satellite Cloud Climatology Project (ISCCP) (Rossow and Schiffer, 1991), the Multiangle Imaging SpectroRadiometer (MISR) instrument aboard the Terra satellite (Marchand et al., 2010), and the Geoscience Laser Altimeter System (GLAS) solar background application project (Yang et al., 2008). For retrievals over the ocean and land, the 780 nm and 680 nm channels are employed, respectively. These retrievals assume fixed particle sizes based on the MODIS global cloud effective radius modes of Collection 6 (C6) MODIS cloud products (14 µm for liquid clouds and 30 µm for ice clouds). Analysis of the MODIS data has indicated that the uncertainties for a single-channel retrieval, attributed to assuming a fixed cloud effective radius, are approximately 10% for liquid clouds and 2% for ice clouds (Meyer et al., 2016).

The EPIC COT algorithm utilizes a common code base and adopts the same forward model assumptions as the existing C6/C6.1 MODIS cloud optical/microphysical property retrievals (MOD06) reported by Platnick et al. (2017) after incorporating ancillary data wherever feasible. For every cloudy pixel, the algorithm retrieves and reports two COT values by assuming liquid and ice phases separately. This methodology is similar to the approach employed by Chiu et al. (2010).

In our analysis, the datasets are split into local time zones based on the longitude and UTC acquisition time. The observations are gridded into 1° × 1° grid cells, with the average cloud properties calculated for each cell. The data are further analyzed on the basis of local hourly bins, generating 24 local time maps for each EPIC granule.

EPIC’s location at the Lagrange L1 point provides optimal pixel resolution at nadir. Off nadir, these pixels become elliptical, with the long axis larger by a factor of about 1/cos (SZA) while the short axis remains unaffected. Depending on the orbital position, there are some limitations in the polar regions owing to the Earth’s axial tilt; hence, we limit our study of COT to regions within ±60°. EPIC acquires up to 13 (in boreal winter) and up to 22 (in boreal summer) images per day, with the image center always corresponding to local noon while the left and right edges correspond to sunrise and sunset, respectively. We analyzed 7 years of data from June 2015 to June 2022 (over 22,000 full-disk images). To prevent potential artifacts at the edges of the EPIC images during sunrise and sunset, we focused exclusively on pixels with local times ranging from early morning (8:00) to late afternoon (16:00 for liquid, 15:00 for ice) and containing sufficient amounts of data to provide robust statistics (Li et al., 2022; Delgado-Bonal et al., 2021).

Changing cloud properties are some of the main drivers of climate change; despite their importance, their diurnal variations have been poorly investigated thus far because of dataset limitations. The cloud diurnal contributions to the time-mean Shortwave flux at the surface and Top of the Atmosphere are as large as 20 W/m² locally, and the contributions to the time-mean Longwave flux at the surface and TOA are typically 1–5 W/m² (Bergman and Salby, 1997). Even with the recent and future increases in data availability from commercial satellites, the limitations remain since most of the satellites are polar orbiters. To analyze the full range of cloud variability properly, different vantage points in space are required.

Previous efforts to study cloud diurnal properties have used a combination of sensors, such as those included in the ISCCP, to provide a global view (Cairns, 1995) or have limited the research to specific regions using geostationary satellites (Chen et al., 2018; Delgado-Bonal et al., 2022). To this day, the only sensor capable of providing a global (albeit daytime) view is EPIC, whose capabilities are exploited in this and previous works.

In this work, we find that the COT does not exhibit different behaviors over the land and ocean for low clouds, unlike the cloud fraction and cloud height. The general diurnal evolution of the EPIC low-cloud COT is an increase from morning to noon, followed by an almost symmetric decrease in the afternoon. For high clouds, the COT over land exhibits a steady increase from morning to afternoon while that over the ocean shows a maximum around noon. The amplitudes of these cycles are generally smaller than those of the low clouds. Owing to EPIC’s daytime limitation, the nighttime changes in COT are still unknown, and further research with other instruments (including active sensors) will be required to unveil their patterns.

In our analysis of the COT, all seasons exhibit similar diurnal changes. The statistical distribution of the COT corresponds to a lognormal distribution; thus, we used the median values for comparisons between seasons rather than the means. Given their similarity, we have restricted the season to boreal spring (MAM) for this analysis. However, a zonal breakdown shows larger differences in the COT for both the absolute value and diurnal evolution. For example, low intermediate clouds over the ocean exhibit a completely different evolution over the tropics than over other regions of the planet.

We recognize the fact that clouds can be classified into different groups, and we have grouped them into ranges based on the optical thickness. By combining the diurnal cycles of COT and cloud fraction, we obtain the diurnal evolution of the cloud-fraction-weighted optical depths for low clouds, showing their contribution to the average and the distinct diurnal variability of each group.