Gary L. Russell1
Andrew A. Lacis
Wenying Su- 1NASA Goddard Institute for Space Studies, New York, NY, United States
- 2NASA Langley Research Center, Hampton, VA, United States
- 3Columbia University, New York, NY, United States
Detailed cloud information over the Earth’s sunlit hemisphere is an important EPIC-image biproduct stemming from reflected solar shortwave (SW) flux determination from EPIC-image backscattered radiances. Using MODIS and CERES satellite retrievals EPIC spectral radiances are transformed into pixel-level broadband radiances. Cloud property information gathered from low-Earth-orbit and geostationary retrievals coincident with EPIC-view geometry are selected. CERES angular distribution models (ADMs) are utilized to accomplish the EPIC radiances-to-flux conversion. Cloud, being the principal contributors to Earth’s planetary albedo, are also the controlling factor regulating the Earth’s global energy balance. The prime focus here is to study the global-scale variability of the terrestrial energy balance using global-scale EPIC-derived reflected fluxes and cloud property information obtained with daily time resolution, a unique capability specific only to DSCOVR Mission EPIC data acquired from the Lissajous orbital vantage point around the Lagrangian L1-point. One major sticking point in model/data comparisons is that climate GCMs and the real-world exhibit quasi-chaotic variability. Thus, cloud maps generated from climate GCM output, and satellite data retrievals, can only provide qualitative information in model/data comparisons. Global integration suppresses the ubiquitous weather noise, but issues with viewing geometry, diurnal cycle, and space-time resolution incompatibilities persist in model/data comparisons utilizing traditional monthly-mean GCM output formats and traditional monthly-mean satellite data displays. DSCOVR Mission EPIC data, coupled with DSCOVR satellite ephemeris-enabled GCM data aggregation provide a promising new approach. In this approach, integration over the sunlit hemisphere eliminates the quasi-chaotic weather noise, while assuring identical viewing geometry and consistent diurnal cycle sampling. The Earth’s rotation provides precise longitudinal alignment of the variability. Moreover, this approach also makes possible day-by-day model/data comparisons, and brings into model/data scrutiny relevant cloud process timescales that are otherwise excluded in traditional monthly-mean model/data comparisons. Results to-date show that DSCOVR Mission measurements from the Lagrangian L1 vantage point, including the use of ancillary and biproduct data assembled within this format, constitute a new and powerful capability for model/data variability profile comparisons operating with a 1-day time resolution.
1 Introduction
The EPIC imaging camera aboard NASA’s DSCOVR Mission satellite orbiting the Lagrangian-1 (L1) point, takes high resolution images of the Earth’s sunlit hemisphere. Depending on telemetry rate, some 13 to 22, 1024x1024 pixel EPIC images of the Earth are acquired per day (Marshak et al., 2018). Stemming from these 5 to 6 thousand EPIC images/year, EPIC data products have been tabulated since January 2017 in the form of broadband SW radiances and radiative fluxes, including global-scale cloud radiative properties. These EPIC data products provided important support in the calibration effort of the SW radiometry channels of the companion NISTAR instrument. They also played a key role in developing the methodology for converting the NISTAR SW radiances into radiative fluxes.
The hemispheric average of the diurnal cycle in the EPIC data is complex, but it is nevertheless precisely defined. We use the term ‘dayurnal’ to refer specifically to the daily 24-hour variability that is observed when viewing the Earth’s sunlit-hemisphere from an external perspective (i.e., the Lagrangian L1 point) as the Earth rotates. We want to avoid confusion with the well-established meaning of the term “diurnal,” which refers to the daily 24-hour variability that is ground-surface location-specific. In the present context, a dayurnal cycle data point represents a precisely defined sunlit-hemisphere average over many contributing local diurnal cycles. Specifically, all of the sub-satellite longitude ground-surface points are being sampled at high-noon of the diurnal cycle, and for each 15° of longitude in the eastward direction, those ground-surface points are being sampled at 1-hour intervals of the afternoon diurnal cycle until the evening terminator is reached. Similarly, for each 15° of longitude in the westward direction, the morning diurnal cycle is sampled at 1-hour intervals until the morning terminator is reached.
In the EPIC data context, precise accounting for diurnal cycle sampling is a key data feature. Based on EPIC and GOES-R/ABI observations, Delgado-Bonal et al. (2022) find that low clouds exhibit minimum altitude near local noon, while high-altitude clouds show a steady decrease from morning until evening, and that the diurnal cycle is weaker over oceans than on land. In view of that, it is therefore important that climate GCMs utilize the DSCOVR satellite ephemeris so that the sampling of their output data over the GCM sunlit-hemisphere is precisely aligned with the EPIC view of the Earth.
Another key feature of EPIC data is the extreme viewing geometry that is not typically encountered in other satellite data. The DSCOVR satellite is in a ∼6-month Lissajous orbit around the Lagrangian L1 point a million miles in the direction of the Sun (Marshak et al., 2018). Viewed from Earth, the angular size of the Lissajous orbit is about 12°. Because of the large inclination of the Lissajous orbit, the planetary-view phase angle (180° minus the scattering angle), also frequently described as the Satellite-Earth-Vehicle, or SEV angle, can become as small as 2°. At such extreme phase angles, converting backscattered radiances into radiative fluxes can be problematic. Also, encountered at extreme backscattering angles, is “the hotspot effect” with the reduced shadowing at near-zero phase angles in vegetated canopies (e.g., Ross and Marshak, 1988; Marshak et al., 2021). That is the price for being able to view all at once, the entire sunlit hemisphere of the Earth.
Those prior studies result employed the commonly used monthly-mean data format, but were still able to show the significant GCM overestimate of clouds over the oceans, along with a corresponding cloud underestimate over continental land areas. While these, and other ModelE2 cloud deficiencies had generally been known, or at least suspected, there had not been this type of direct and quantitative model/data comparison between satellite data and climate GCM output data to describe and quantify the specific nature of these ModelE2 cloud distribution shortcomings. Despite the monthly-mean format, this prior study is, nevertheless, an effective example of a longitudinal variability mismatch of a physical quantity where agreement is good at the global annual-mean level in that physical quantity.
The common practice of conducting model/data comparisons using monthly-mean tabulations has been the simplest means for averaging out the quasi-chaotic weather noise in both satellite and climate GCM data. But this practice also eliminates the timescales that are most directly associated with the life cycle processes of clouds and the formation, evolution, and the ultimate decay of storm systems. Accordingly, performing the EPIC vs. ModelE2 variability comparisons at the 1-day resolution range makes more informative model/data comparisons possible.
Variability in itself may be somewhat nebulous as a specific climate GCM diagnostic concept, in that there is no single physical process being targeted. There is also no single measurement that would convey, or quantify, the magnitude of variability, or even prove its presence, or absence. Its relevance to effective climate modeling becomes evident in comparing 1D and 3D approaches to model climate. A ID climate model can be run to a precise equilibrium (to seven decimals, or more) with virtually zero variability, since the end result is guaranteed by the initial input assumptions of the model. The basic information to adequately define variability requires a sufficiently long, self-consistent dataset of systematic measurements with adequate space-time sampling. Remarkably, EPIC data are sufficient to define the variability profile of the global planetary albedo and the global cloud cover, and also to eliminate the weather noise.
In that regard, DSCOVR Mission EPIC data are exemplary in their ability to address questions of climate system variability more effectively than has been possible using conventional satellite data. This is because in the DSCOVR/EPIC data analysis approach, systematic averaging over the sunlit hemisphere suppresses the quasi-chaotic weather noise in both the EPIC and GCM data aggregations while preserving the global-scale variability information. The DSCOVR data approach also eliminates the viewing geometry and diurnal sampling differences when the GCM data aggregation utilizes the DSCOVR satellite ephemeris information. Further, in this approach the global SW energy component of the Earth’s climate system in its entirety can be readily compared between EPIC observational data and GCM modeling data on a daily basis with precise longitudinal resolution. Since clouds account for 80%–90% of the reflected radiation by Earth (Stephens et al., 2015), the global variability of reflected solar SW radiation is also a direct measure of the global-scale cloud variability.
There are two separate objectives being pursued in this study. The first and primary objective is to demonstrate and explore the unique aspects and capabilities of EPIC data that could serve as a new source of global -scale 1-day resolution observational data that is optimized to assess climate GCM performance at time scales shorter than is currently afforded by the available monthly-mean satellite data. The secondary objective is testing the mechanics of the climate GCM replication of the EPIC viewing geometry and data aggregation methodology, for which we utilize a coarse-grid vintage version of the GISS ModelE2 for its simplicity of implementation. This is a model that was used by Hansen et al. (2005) to study the efficacy of radiative forcings, and to reproduce the annual global-mean surface temperatures over the past century. Given its globally uniform relative humidity criteria for cloud condensation onset, it was suspected to overestimate clouds over the oceans and underestimate clouds over land, but still able to accurately reproduce most aspects of global climate change over the past century. It is therefore of interest to quantify these suspected shortcomings of cloud distribution in the GISS ModelE2 relative to the new 1-day resolution EPIC data format in preparation for a more informed comparison of the EPIC data with the improved cloud physics scheme of the GISS ModeleE3.
2 Global variability at 1-day resolution
The key point in our present study relative to our prior EPIC-data analysis is depicted in the comparison of Figures 1, 2, which paints a much more dynamic view of the global planetary albedo variability compared to the subdued, smoothly shifting seasonal changes that are displayed in Figure 3 of Carlson et al. (2022). Explicit use of 1-day data resolution preserves the short-period cloud process variability that gets obliterated by the monthly-mean averaging in that prior study. Because of the daily rotation of Earth 1-day data resolution that is automatic for the DSCOVR Mission EPIC and NISTAR instruments, and because of the precise space/time sampling and viewing geometry coordination that is possible between EPIC measurements and GCM modeling results that permit identical averaging over the Earth’s sunlit hemisphere to zero out the chaotic weather noise, and thus effectively address the global-scale climate system variability on timescales shorter than a month.
Figure 1. Hovmöller map comparison of EPIC vs GISS ModeE2, of the Earth’s global-scale planetary albedo for year 2018. Left panel: Planetary albedo based on EPIC-derived solar SW radiative flux. Right panel: GISS ModelE2 GCM simulated planetary albedo. Y-scale depicts the 1-day resolution time-evolution expressed as Julian day-of-year at panel right, and calendar month-of-year at panel left. X-coordinate longitude scale is set by the Earth’s rotation. The global map at panel bottom is for longitudinal orientation. Superimposed sub-Sun latitude track depicts solar declination (top-of-panel scale, with dashed Date Line as zero point). Also superimposed are the DSCOVR sub-Satellite latitude track, and the Sun-Satellite phase angle track.
Figure 2. EPIC vs GISS ModeE2, Hovmöller map comparison of the Earth’s global-scale planetary albedo. Left panel (A): Planetary albedo of the Earth based on EPIC-derived solar SW radiative flux for year 2018. Right panel (B): Planetary albedo of the Earth for year 2018 based on GISS ModelE2 climate GCM simulation. The Y-scale depicts the time-evolution from January 1 to December 31. The X-coordinate longitude scale is determined by the Earth’s rotation. The Figure 2 EPIC and ModelE2 input data are the same as the EPIC and ModelE2 data in Figure 1, except that the 1-day resolution data in Figure 1 have been averaged into monthly-means.
Figure 1 presents a Hovmöller contour map comparison of the Earth’s planetary albedo based on the EPIC year 2018 SW flux data (left panel), and the corresponding climate simulations with the GISS ModelE2 (right panel) with data are displayed with 1-day resolution, rather than monthly-mean format. The most basic takeaway from the Figure 1 EPIC vs. GCM comparison is the clear documentation of the long-suspected ModelE2 cloud overestimate over the Pacific Ocean during the Northern Hemisphere summer months, and the cloud underestimate over continental and polar regions, which had been previously described by Carlson et al. (2022). The new finding from Figure 1 with the 1-day resolution EPIC data, is that ModelE2 was underestimating the strength of longer-period variability, while overestimating the shorter-period cloud variability, particularly in the longitudinal direction.
The Figure 1 (left panel) Hovmöller map of the EPIC-derived planetary albedo has the time-scale of the Y-axis increasing upward, with the mid-month points of the year displayed on the left and the Julian day of the year displayed on the right of each panel. The X-axis displays the GMT time of the daily EPIC-image data points, expressed in terms of their longitude positions, have been interpolated to a uniform grid for the Hovmöller contour mapping with the International Date Line (vertical dashed line) at figure center. Each plotted pixel point represents a global sunlit-hemisphere averaged planetary albedo at its specified longitude. For geographic orientation, a global map has been superimposed at panel bottom. Also included is the time dependent track of the Solar Declination, sub-Satellite latitude, and the Sun-Sat phase angle with their X-axis scale at the top of each panel. Because the entire contour map was constructed in quadrants, the maximum, average, and minimum planetary albedo values of each quadrant are also included along the Y-axis surrounding the Feb, May, Aug, and Nov labels. The global annual-mean planetary albedo from the EPIC data for 2018 is 30.40%, with the maximum and minimum values of 35.11% and 25.50%, respectively. Altogether, there were 5,351 sunlit-hemisphere SW flux data points from the EPIC images for the year 2018 that were linearly interpolated to define the time-series for the global-scale variability from January 1 to December 31. For those Julian days with missing data (identified by the red dash lines along the Y-axis at panel right), the missing data were linearly interpolated at fixed longitude from the prior and following day’s data.
As an overall impression, the Hovmöller contour map shows the highest planetary albedo occurring during the December to January time period when the DSCOVR satellite is at its southern extreme with Antarctica in view. A secondary maximum occurs during the May to June time period when the DSCOVR satellite is at its northern extreme. Longitudinally, the highest albedos are seen to occur over the heavily continental land areas encompassing Africa and Asia, with the lowest planetary albedos occurring over the Pacific Ocean. A strong minimum in the planetary albedo occurs from September to October, centered over the International Date Line. The most striking feature that emerges with the 1-day time resolution is the persistent global-scale variability of the Earth’s planetary albedo occurring at a wide frequency range at virtually all longitudes, and virtually all days of the year. The narrowest features appear to be only a day or two wide. The most prominent periodicity for the planetary albedo variability seems to be the 10- to 15-day timescale. There is also evidence of differences in slope, with some features having slopes that increase to the right (signifying eastward movement with time), while other features have slopes that increase toward the left (signifying westward movement with time). Because the EPIC-image pixel aggregation is over the entire sunlit hemisphere, there is information as to the latitude of any planetary albedo feature.
The Hovmöller map in Figure 1 (right panel), is the corresponding GISS ModelE2 simulation of the global planetary albedo for the year 2018, calculated with monthly-mean prescribed sea surface temperatures, and aggregated over the sunlit hemisphere (4° × 5°) grid-boxes using the DSCOVR satellite ephemeris to precisely match the EPIC-image viewing geometry. There are similarities and clear differences with the EPIC results. The overall impression, contrary to EPIC, had the maximum planetary albedo occurring over the Pacific Ocean during the summer months June to July, with the secondary maxima occurring in the November to December time period, and with the minimum planetary albedo at the longitudes of the continental land areas. This is fully in accord with the prior findings of Carlson et al. (2022). What is new is that the GISS ModelE2 exhibits ubiquitous short-short period variability occurring at all days of the year and at all longitudes. The variability is more subdued, and not nearly as punctuated as that of EPIC, and the ModelE2 features do not have the pronounced longitudinal extent as those of EPIC. The global annual-mean planetary albedo of 29.86% is close to the EPIC value. Not surprisingly, the range of the ModelE2 planetary albedo, with a maximum of 34.84% and a minimum of 26.21%, is smaller than that of EPIC. Also, there is no deep minimum in the September to October time period, as there is in the EPIC data.
Figure 2 re-emphasizes the main point the is being advanced in Figure 1. The same basic input data for planetary albedo for the year 2018 are displayed in Hovmöller contour format in both figures. The only difference is that the full 1-day resolution that is intrinsic to the EPIC data, is being utilized in Figure 1, while monthly-mean averages of the same data are displayed in Figure 2, as they were in our prior study. The same broad-brush conclusions that were reached by Carlson et al. (2022) that the longitudinal cloud distribution in the GISS ModelE2 is at variance with observations are compelling in both Figures 1, 2, and may even be more clearly discernable in the monthly-mean version, which is not all that surprising, since controlled resolution broadening can be a useful tool in isolating and identifying specific sources, or causes, that may all be simultaneously contributing variability at different frequencies.
But it is indisputable that the pervasive variability of the Earth’s planetary albedo, occurring at all longitudes and days of year at a wide range of amplitudes and frequencies, is not even suspected from the monthly-mean results in Figure 2. The only question that matters is whether the complex variability that is seen in Figure 1 is real, or just some semi-random modeling artifact arising from the extensive modeling effort required in deriving the radiative fluxes from the backscattered EPIC-image radiances. That question, based on simultaneous, direct and totally independent silicon photodiode measurements by NISTAR, the companion instrument aboard the DSCOVR spacecraft, is answered resoundingly in our companion paper (Lacis et al., 2025). The global planetary albedo of the Earth exhibits continuous, readily measurable variability occurring at a broad range of characteristic frequencies and amplitudes at all days of the year, and at all longitudes, just as shown in Figure 1.
Accordingly, the seasonal and longitudinal variability of the planetary albedo is by inference a direct measure of how the Earth’s global solar energy input varies seasonally, and as a function of longitude. Furthermore, since clouds account for 80%–90% of the reflected radiation by the Earth (Stephens et al., 2015), the global variability of reflected solar SW radiation is also a direct measure of the global-scale, seasonal and longitudinal cloud variability. Hence, there is much new to be learned via variability comparisons about how well the climate GCM cloud modeling methodology can simulate the complex cloud related processes of the climate system.
3 EPIC all-cloud global variability
An extensive modeling and data processing effort is involved in generating the EPIC Composite data stream (Su et al., 2018; Su et al., 2021). This involves constructing a 5-km resolution database that includes selecting detailed cloud properties such as cloud fraction, cloud-top altitude, cloud water/ice phase, optical depth, and particle size, by utilizing cloud property retrievals from low Earth orbit (LEO) and geostationary (GEO) satellites (i.e., MODIS, VIIRS, AVHRR, and GOES-13, -15, -16, -17, METEOSAT-8, -9, -10, -11, MTSAT-2, Himawari-8), that use a common set of retrieval algorithms (Minnis et al., 2008; Minnis et al., 2011). The 5-km resolution global cloud properties in this EPIC Composite database are then projected onto the roughly 8 × 8 km EPIC-image pixels, and aggregated over the sunlit hemisphere to produce the 5 to 6 thousand global-scale data points per year, of cloud fraction, cloud-top altitude, cloud optical depth and cloud particle size, for all-cloud and ice-cloud categories. In the process, the quasi-chaotic weather noise gets averaged out, and thus define a climate-style dataset to characterize the global-scale variability of the reflected solar SW radiation as well as that of the corresponding cloud properties.
Though nominally designed to be acquired at fixed 1-hourly intervals, because of telemetry and other constrains, there is a maximum of 22 EPIC images that are acquired per day from May to August, otherwise, the maximum image count is 13 images per day. There are also data dropouts, spacecraft maneuvers, interference from having the moon in the field of view, and telemetry problems where significant data gaps, and even missing days, occur. Accordingly, to facilitate the data analysis, the first step is to construct a uniformly spaced grid in hourly GMT by carefully filling in the missing data by linear interpolation in GMT between the observed EPIC-image data points. For missing days and large data gaps, the missing data are filled in by linear interpolation at fixed GMT (or fixed longitude). In this way, a continuous, uniformly spaced data stream is obtained from 1 January 2017 to 31 December 2018, given that we have two contiguous years of EPIC cloud properties for years 2017 and 2018. Similar considerations apply for two other contiguous years of EPIC cloud properties that are available for years 2020 and 2021.
Also, as a small refinement, we also account for the seasonal change in the Earth’s orbital speed by using the DSCOVR satellite ephemeris to adjust the time-series data points (by linear interpolation) with the position of the sub-satellite longitude to ensure that the data points line up consistently over the same longitude. This amounts to converting the GMT standard time to GLT local time, where each GLT hour corresponds precisely to 15° of longitude throughout the year. From this uniformly-spaced continuous data stream over the two contiguous years of EPIC data, we formed the uniform grid with 1-day resolution in time and 15° resolution in longitude for the Hovmöller contour map in Figure 1. This same data stream when plotted linear format, shown below, then provides a quantitative measure of the amplitude and frequency of the global-scale variability.
There is too much variability in the EPIC data products to make line plots at the full 1-day resolution. To facilitate viewing the EPIC data in line format, we utilize a triangular broadening profile to pass day-by-day over the data stream. Thus, for example, data smoothing with a 30-day half-width at half-maximum is basically equivalent to a monthly-mean average, except that it provides a uniform day-by-day smoothness, rather than imposing the arbitrary commonly utilized calendar boundaries to accomplish the averaging. Using a triangular 30-day half-width profile will totally zero out a 30-day wavelength sine wave, including integer multiples that add to 30 (such as 15-day, 10-day and 5-day frequencies), while strongly suppressing the frequencies close to 30-day, especially those with shorter period than 30-day variability. Meanwhile, a sine wave having a 60-day wavelength will be preserved in place, but with its amplitude reduced by more than half.
3.1 EPIC all-cloud sky-fraction
Using a triangular 30-day half-width broadening profile, Figure 3A displays the EPIC Composite all-cloud sky-fraction for the contiguous years 2017 and 2018. The results closely resemble the monthly-mean results of Carlson et al. (2022), except that the color coding for the East Atlantic (orange) and NS America regions (light blue) is interchanged. 24 color-coded lines are used to depict the seasonal change in all-cloud sky-fraction at corresponding 24 equally spaced longitudes, each tagged with the nominal GLT (local) time of when the high-noon sun occurs as that particular EPIC image is acquired, and also with the corresponding geographic location that serves as a longitude locator. The vertical dotted lines and the corresponding numerical month-of-year indicators identify the mid-month positions of the season. Hence, the vertical dotted lines delineate 30-day intervals, and the four X-scale tick marks per month represent time intervals of slightly more than 1-week. The daily data points are plotted in accord with the internal Julian-day-of-the-year scale. Also included at the bottom of each panel is a bar graph showing the number of EPIC images that were acquired on each day of the year. Given the substantial amount of data interpolation and filling in of missing data points, including the bar graph was to check whether there might be some correlation between the variability and the location of the filled in data gaps, or the number of EPIC images that are acquired each day. Only from May through August is the maximum rate of 22 images per day achieved. With the careful interpolation and filling in of missing data, there appears to be minimal impact on the overall variability structure.
Figure 3. Top panel (A): EPIC-derived daily all-cloud sky-fraction for years 2017 and 2018 for 24 color-coded longitudes (30-day resolution data referenced at high-noon GMT with corresponding geographic location), grouped in 5 color-coded regions: EurAsia (magenta), East Asia (green), Central Pacific (blue), NS America (light-blue), and East Atlantic (orange). Heavy black line depicts the seasonal global-mean. X-scale lists numerical mid-month of year; data use the internal Julian-day scale. Panel bottom bar-graph shows number of EPIC images per day. Bottom panel (B): 30-day broadened longitudes for EurAsia (magenta), Central Pacific (blue), NS America (light-blue), and East Atlantic (orange) from Panel (A), but with 4-day broadened data.
The 24 longitudes are grouped into 5 contiguous color groups. Of the principal color groups, the (magenta) continental EurAsia group is consistently exhibiting the lowest cloud fraction, except for the May to September period when the East Atlantic group (orange) dips to its deepest minimum cloud fraction in July. These two are the more heavily endowed continental land regions and are represented by their respective principal members Iraq (GLT = 9, heavy magenta line with black dot mid-month indicators), and by W. Africa (GLT = 13, heavy orange black-dot line). The highest cloud fractions are associated with the (blue) Pacific Ocean group represented by its principal member Central Pacific (GLT = 23, heavy blue line, also with black dot mid-month indicators), and the (green) East Asia group represented by its principal member Taiwan (GMT = 4, with the heavy green black-dot line). The (light-blue) North and South America group, represented by its principal member St. Louis (GMT = 18, heavy light-blue black-dot line), meanders along the heavy black line, which is the global-mean averaged over all longitudes. Interestingly, this ordering of the all-cloud sky-fraction is practically the reverse of the planetary albedo longitudinal ordering in Figure 1 of Carlson et al. (2022), where the minimum in planetary albedo is clearly occurring over the Central Pacific Ocean region, while the maxima in planetary albedo are located over the continental EurAsia land areas.
Overall, the dayurnal amplitude of the all-cloud sky-fraction does not change appreciably over the two-year time period, with annual-mean cloud fractions of 61.05% in 2017, and 62.06% in 2018, as listed in the figure bottom-right corners. There are prominent 60-day MJO-type oscillations in cloud fraction occurring at the East Asia (heavy green line) and Western Pacific (blue dash line) longitudes from September 2017 until February 2018, then taking a strong 60-day decrease, perhaps a part of a 5-month long oscillation with relative minima in April and September of 2018. There are also prominent 60-day MJO-type oscillations in cloud fraction evident along the EurAsia longitudes (heavy magenta line) from June 2017 through February 2018. NOAA’s Oceanic Nino Index (ONI) shows a negative (La Niña) value setting in the latter part of 2017 and extending into the early part of 2018, perhaps coincident with the rise in the global cloud fraction value (heavy black line) over the same time period.
To focus more on the shorter period variability of the all-cloud sky fraction, a broadening profile with a 4-day half-width is passed over the 1-day resolution data stream as displayed in Figure 3B. For clarity, only four of the principal members of the 5 color groups are depicted in their 30-day broadened form in the heavier-set colored lines, including their black-dot mid-month indicators, with the 4-day broadened data stream (in the more lightly-set colored lines) superimposed. There is a wide variety of waveforms, frequencies and amplitudes, with most of the strongest amplitude oscillations occurring over the Central Pacific (blue) longitude, and with generally smaller amplitude oscillations happening at the continental land-rich longitudes.
Theory predicts well-documented 5-day waves in the lower atmosphere with strong signatures in pressure and geopotential height (Madden and Julian, 1972), as well as 2-, 5-, 10-, and 16-day westward propagating traveling planetary waves that have been observed (Yamazaki and Matthias, 2019; Zhang et al., 2022). Note also that 10-20, 20-30, and 30-60-day low frequency periodicity has been observed in northern China in the average minimum temperature, with the 10-20-day oscillations being the most significant. All of these frequencies can be identified as being present in Figure 3B. However, the application of the 4-day resolution broadening to the 1-day resolution data stream will have obliterated the 2-day and 1-day variability. Nevertheless, significant 2-day variability, and even occasionally large 1-day changes in the EPIC 1-day resolution data are seen to occur, as evident in Figure 2 of our companion paper (Lacis et al., 2025).
Perhaps the most prominent are the MJO-type 30-day oscillations (Madden and Julian, 1972) that appear over the Central Pacific Ocean (blue line) from April to October of year 2018, coinciding with the ONI negative La Niña conditions ending, and converting to a more positive ONI though the end of year 2018. These prominent MJO features appear also to be showing the evolving ratio of the active to inactive MJO phase (e.g., Jones and Carvalho, 2002; Zhang, 2005) from the broad active MJO phases from May to July, to the much narrower active MJO phases thereafter. Otherwise, the 10-15-day shorter period and smaller amplitude variability appears to be more prevalent at the more continental EurAsia (magenta) and W. Africa (orange) longitudes.
The Figure 4A display of the EPIC all-cloud sky-fraction for the contiguous years 2020 and 2021 has basic similarity to the 2017 and 2018 years. The longitudinal ordering and the dayurnal amplitude of the all-cloud sky-fraction does not change appreciably, with annual-mean cloud fractions of 61.89% for 2020, and 62.04% for 2021. The NOAA Oceanic Nino Index shows a strongly negative (La Niña) index for both years. There are some limited intervals of 60-day MJO-type oscillations, with the most sustained MJO activity occurring at the EurAsia longitudes (heavy magenta line) from February 2021 until October 2021. There are also what appear to be extra-long period 120-day MJO-type oscillations in the all-cloud sky-fraction that are prominent at the East Asia (heavy green line) and Central Pacific (heavy blue line) longitudes, with broad minima occurring in April and August of both years. Also, what were broad, but distinct, July minima in the 2017 and 2018 all-cloud sky-fraction at the W. Africa (orange) longitude, have broadened greatly in the years 2020 and 2021.
Figure 4. Top panel (A): EPIC-derived daily all-cloud sky-fraction for years 2020 and 2021 for 24 color-coded longitudes (30-day resolution data referenced at high-noon GMT with corresponding geographic location), grouped in 5 color-coded regions: EurAsia (magenta), East Asia (green), Central Pacific (blue), NS America (light-blue), and East Atlantic (orange). Heavy black line depicts the seasonal global-mean. X-scale lists numerical mid-month of year; data use the internal Julian-day scale. Panel bottom bar-graph shows number of EPIC images per day. Bottom panel (B): 30-day broadened longitudes for EurAsia (magenta), Central Pacific (blue), NS America (light-blue), and East Atlantic (orange) from Panel (A), but with 4-day broadened data.
Aa a further point to note, there are actually no observational data available for the first two months of 2020 because of the DSCOVR spacecraft gyroscope malfunction. What appears in the Figure 4 plots for the first two months of 2020 are “filler” data, copied from year 2021, and used for buffering the observed data edge boundaries to facilitate the resolution broadening of the 1-day resolution EPIC data. In this way, for extra-wide broadening profiles, when the broadening profile extends beyond the dataset edge, the plausible buffer values (instead of zeros) minimize what would be a larger error in the broadened data edge points.
The same approach is used in constructing Figure 4B, as previously used in Figure 3B for years 2017 and 2018. Similarity in variability is expected for years 2020 and 2021, but there are notable differences. The variability at the Central Pacific longitude (blue) appears to be less organized in structure and with a somewhat reduced amplitude. There is still what look like MJO-type oscillations, but now exhibiting a near-40-day periodicity with a double peaked active phase, that is followed by a broad inactive phase, especially in the latter half of 2021. Meanwhile the continental EurAsia longitude (magenta) is showing 30-day MJO variability beginning March 2020, until shifting to a more complex form in July of 2020. Year 2021 begins with what appear to be 30-day MJO-type oscillation with increasingly broader inactive phases, before switching to higher frequency 15-day oscillations in May that continue until the end of 2021. There appears to be little change in the variability structure at the W. Africa (orange) longitude, compared to 2017 and 2018.
3.2 ModelE2 all-cloud sky-faction
The climate model used in this study is a (4° x 5°) coarse-grid coupled atmosphere-ocean version of the GISS ModelE2 (Schmidt et al., 2014) that has been used in numerous climate trend and climate sensitivity simulations and CMIP climate data comparisons. It is also the version that was previously adapted with DSCOVR satellite ephemeris based online model output data aggregation to simulate the EPIC-image viewing geometry of the Earth’s sunlit hemisphere (Carlson et al., 2022). This version of the GISS ModeE2 was optimized to test the prospects and capabilities for more closely coaligned model/data comparisons where the space-time data sampling and target viewing geometry are closely matched. Utilizing a 1-hour timestep for the ModelE2 radiative sampling establishes a close match to the data-rate of EPIC image acquisition.
With the availability of the daily EPIC Composite cloud product available (Su et al., 2018), and with the development of DSCOVR satellite ephemeris assisted GCM aggregation of climate simulation output data operational (Carlson et al., 2022), the necessary 1-day resolution global-scale observational data, and the corresponding climate GCM generated output data with identical EPIC-view space-time sampling, are in place for conducting a detailed 1-day resolution model/data comparison that has not been feasible to perform with conventional satellite data and conventional climate GCM output data. Figure 5 displays the GISS ModelE2 component of this novel model/data comparison.
Figure 5. Same as in Figure 3, except that the data source is the GISS ModelE2. Top panel (A): ModelE2 daily all-cloud sky-fraction for years 2017 to 2018, for 24 color-coded longitudes with 30-day resolution. Longitudes are referenced by high-noon GMT, and geographic location, and grouped into 5 color-coded regions: EurAsia (magenta), East Asia (green), Central Pacific (blue), NS America (light-blue), and East Atlantic (orange), and tagged with black-dot symbols at midmonth intervals. Heavy black line is the seasonal global mean. Bottom panel (B): 30-day broadened all-cloud sky-fraction reference longitudes: EurAsia (magenta), Central Pacific (blue), NS America (light-blue), and East Atlantic (orange), with superimposed 4-day broadened data.
In comparing ModelE2 Figure 5A with EPIC Figure 3A, what stands out is the big difference in range of the seasonal variation of the dayurnal cycle amplitude of the ModelE2 global-scale all-cloud sky-fraction. The ModelE2 amplitude of the diurnal cycle cloud-fraction is more than twice that of EPIC during the Northern Hemisphere (NH) summer months, and then, drop to less than half in November. The global annual mean ModelE2 sky-fraction is 55.06% in 2017, and 55.37% in 2018, which is about 10% smaller than EPIC. However, both EPIC and ModelE2 have year 2018 with a slightly larger sky-fraction than 2017. In an absolute sense, the seasonal changes in the global annual-mean sky-fraction of both EPIC and ModelE2 is relatively small and not that different in magnitude.
It is interesting that the ModelE2 global-scale longitudinal distribution is able to consistently maintain the same longitudinal ordering over both years 2017 and 2018. Remarkably, in the ModelE2 results, the East Asia (green) and Central Pacific (blue) longitudinal regions consistently exhibit the highest sky-fractions, while the EurAsia (magenta) and W. Africa (orange) longitudes have the lowest cloud-fractions.
Also on the positive side, the ModelE2 variability of the short period sky-fraction in Figure 5B is remarkably similar to the EPIC short period sky-fraction variability shown in Figure 3B, implying that the modeling of the cloud and storm process life cycles in ModelE2 achieves an adequate resemblance to real-world cloud variability. While there is no sustained period of 30-day MJO variability as in year 2018 of Figure 3B, there are isolated regions where such variability is present. Otherwise, the dominant periodicity appears to be of the 10-15-day variety with a similar range of amplitudes as seen in the EPIC cloud-fraction variability. Compared to 2017 and 2018.
3.3 EPIC all-cloud cloud-top height
The methodology to determine the principal cloud properties for the EPIC Composite cloud product is described by Minnis et al. (2008), Minnis et al. (2011). The cloud-top altitude is determined by matching cloud-top temperature retrieved by remote sensing means from satellite orbit to the atmospheric temperature profile at the corresponding point geographic location. Accordingly, the cloud-top altitude determined in this fashion would typically correspond to the level of where the cloud optical depth reaches unity, as measured downward from the top.
Figure 6A shows the EPIC Composite all-cloud cloud-top altitude for years 2017 and 2018. The prominent feature is the broad maximum of the dayurnal amplitude peaking in July, with the dayurnal minimum amplitude also occurring in July. The longitudinal ordering has the maximum cloud-top altitudes occurring over the continental land areas with the New Delhi, India longitude (dash magenta line) at the top. Meanwhile, the minimum cloud-top altitudes are seen to occur over the Central Pacific (heavy blue black-dot line). This is similar to the planetary albedo longitudinal ordering, but opposite to the cloud fraction ordering in Figure 3. Clearly visible is the 60-day MJO-type variability during the 207-2018 NH winter months over the EurAsia (magenta) longitudes. The W. Africa longitude (heavy orange black-dot line) exhibits long period variability over the entire 2017 to 2018 period, beginning with what appears to be 90-day MJO-type variability, shifting to a 120-day periodicity, then at mid-year 2017 continuing with 60-day variability, before converting again to a 120-day, or even longer, periodicity beyond May of 2018. The Central Pacific longitudes (blue) exhibit a peculiar long-period variability in year 2017, then a rather brief 60-day cycle that converts into a 120-day MJO-type periodicity.
Figure 6. Top panel (A): EPIC-derived daily all-cloud cloud-top altitude for years 2017 and 2018 for 24 color-coded longitudes (30-day resolution data referenced at high-noon GMT with corresponding geographic location), grouped in 5 color-coded regions: EurAsia (magenta), East Asia (green), Central Pacific (blue), NS America (light-blue), and East Atlantic (orange). Heavy black line depicts the seasonal global-mean. X-scale lists numerical mid-month of year; data use the internal Julian-day scale. Panel bottom bar-graph shows number of EPIC images per day. Bottom panel (B): 30-day broadened longitudes for EurAsia (magenta), Central Pacific (blue), NS America (light-blue), and East Atlantic (orange) from Panel (A), but with 4-day broadened data.
The short-period variability of the cloud-top height, shown in Figure 6B, exhibits characteristically similar short-period variability as the cloud fraction variability in Figure 3B. The 15-day short-period variability appears to the most persistent periodicity for the all-cloud cloud-top altitude variability. There also appears to be a tele-connection type variability in the all-cloud cloud-top altitude occurring at diametrically distant longitudes. When the cloud-top altitude is increasing from April to July at the continental land longitudes, e.g., New Delhi, India (dash magenta line), it is simultaneously decreasing along the Central Pacific longitudes (blue). Then, as the cloud-top altitude decreases from July to October over the continental longitudes, the cloud-top altitude in increasing over the Central Pacific longitudes, albeit at a reduced rate, but continuing to decrease through December of 2017.
Figure 7 shows the EPIC Composite all-cloud cloud-top altitude for years 2020 and 2021. Here, the first two months of year 2020 are explicitly shown as missing, due to the 2019-2020 DISCOVR satellite gyroscope malfunction. Even though the years 2020 and 2021 are strong La Niña years, there is relatively little change in the all-cloud cloud-top altitude from years 2017 and 2018. Figure 7A shows the prominent 60-day MJO-type variability occurring for most of year 2020 over the continental land longitudes, with the highest cloud-top altitudes centered along the New Delhi, India (dash magenta line) longitude. Figure 7B, exhibits characteristics similar to Figure 6B, but 2020 and 2021 being La Niña years, the amplitudes of the short-period variability appear to be somewhat enhanced. Also, the 15-day short-period variability is the dominant periodicity for the cloud-top altitude variability, although there are patches of what appear to be 10-day variability and a few cases of 30-day MJO-type oscillations.
Figure 7. Top panel (A): EPIC-derived daily all-cloud cloud-top altitude for years 2020 and 2021 for 24 color-coded longitudes (30-day resolution data referenced at high-noon GMT with corresponding geographic location), grouped in 5 color-coded regions: EurAsia (magenta), East Asia (green), Central Pacific (blue), NS America (light-blue), and East Atlantic (orange). Heavy black line depicts the seasonal global-mean. X-scale lists numerical mid-month of year; data use the internal Julian-day scale. Panel bottom bar-graph shows number of EPIC images per day. Bottom panel (B): 30-day broadened longitudes for EurAsia (magenta), Central Pacific (blue), NS America (light-blue), and East Atlantic (orange) from Panel (A), but with 4-day broadened data.
4 EPIC ice-cloud global variability
Cloud property determination for the EPIC Composite cloud product is described by Minnis et al. (2008), Minnis et al. (2011). Remote sensing CO2 slicing methodology from satellite orbit determine the cloud-top temperature, and matching cloud-top temperature to the geolocated atmospheric temperature profile, defines the cloud-top altitude. Clouds with cloud-top temperatures less than 273 K are deemed to be ice clouds for the entire atmospheric column, with their radiative properties determined accordingly.
However, cloud-top altitude, cloud water/ice phase, and the cloud radiative property determination is complex, and thus could be the weak link in the EPIC Composite cloud product. Hu et al. (2010) report that more than 95% of low-level clouds at high latitudes with temperatures that are between 0 °C and −40 °C are water clouds. Similarly, Bodas-Salcedo et al. (2016) note that clouds with supercooled liquid water cloud-tops account for about 1/3 of the total reflected solar radiation in the 40° to 70° S latitude zone.
4.1 EPIC ice-cloud sky-fraction
As is only to be expected, the 2017 and 2018 EPIC ice-cloud sky-fraction in Figure 8A has general similarity to its counterpart all-cloud sky-fraction in Figure 3A, but the longitude ordering is different. In addition, the overall dayurnal amplitude of the ice-cloud sky-fraction is significantly smaller than is the dayurnal amplitude of the all-cloud sky-fraction. It is the East Asia (heavy green, black-dot line) that for the most part, exhibits the highest ice-cloud sky-fractions. The Central Pacific (blue, black-dot line) is also near the top, but during the summer months, the EurAsia (magenta, black-dot line) (most probably, the Indian Ocean) pops up higher. The lowest ice-cloud sky-fractions consistently occur along the NS America longitudes (light-blue, black-dot line), with the deepest minimum occurring in July of both 2017 and 2018. The global annual-mean ice-cloud sky-fraction is 21.02% for 2017, and 22.01% for 2018. The one curious feature is the sharp decrease in the dayurnal amplitude of the ice-cloud sky-fraction in October of 2018. 60-day MJO-type oscillations are evident during the NH winter months over the East Asia (green) longitudes, and also in the early part of 2018 along the NS America (light-blue, black-dot line) longitude.
Figure 8. Top panel (A): EPIC-derived daily ice-cloud cloud-cover fraction for years 2017 and 2018 for 24 color-coded longitudes (30-day resolution data referenced at high-noon GMT with corresponding geographic location), grouped in 5 color-coded regions: EurAsia (magenta), East Asia (green), Central Pacific (blue), NS America (light-blue), and East Atlantic (orange). Heavy black line depicts the seasonal global-mean. X-scale lists numerical mid-month of year; data use the internal Julian-day scale. Panel bottom bar-graph shows number of EPIC images per day. Bottom panel (B): 30-day broadened longitudes for EurAsia (magenta), Central Pacific (blue), NS America (light-blue), and East Atlantic (orange) from Panel (A), but with 4-day broadened data superimposed.
In comparing the EPIC ice-cloud sky-fraction in Figure 8B to its counterpart in Figure 3B, the biggest difference is the different longitudinal ordering, with the Central Pacific longitude (blue, black-dot line) still at the top, but with the NS America longitude (light-blue, black-dot line) at the bottom. The W. Africa longitude (orange, black-dot line) is also near the bottom, while the EurAsia longitude (magenta, black-dot line) is now much close to the top. The short period 10-15-day variability is largely uncorrelated relative to the all-cloud sky-fraction, but the 60-day MJO-type oscillations, especially along the Central Pacific (blue) longitude, show significant similarity during most of the year in 2018.
Figure 9A depicts the 30-day broadened EPIC ice-cloud sky-fraction for the years 2020 and 2021. The longitudinal ordering is the same as it is for the years 2017 and 2018, but the maximum ice-cloud sky-fraction, peaking in March, appears to be more enhanced during the 2020 to 2021 La Niña years. Also, there is no evidence that during the 2020–2021 years of a sharp narrowing of the dayurnal amplitude that occurred in November of 2018. The global annual-mean ice-cloud sky-fraction for the years 2020 and 2021 are 22.12% and 22.25%, respectively, and slightly higher than for years 2017 and 2018. In addition, the lowest sky-fraction NS America longitude (light-blue, black-dot line) shows what appears to look like ultra-long period variability, similar to that in Figures 6A, 8A.
Figure 9. Top panel (A): EPIC-derived daily ice-cloud cloud-cover fraction for years 2020 and 2021 for 24 color-coded longitudes (30-day resolution data referenced at high-noon GMT with corresponding geographic location), grouped in 5 color-coded regions: EurAsia (magenta), East Asia (green), Central Pacific (blue), NS America (light-blue), and East Atlantic (orange). Heavy black line depicts the seasonal global-mean. X-scale lists numerical mid-month of year; data use the internal Julian-day scale. Panel bottom bar-graph shows number of EPIC images per day. Bottom panel (B): 30-day broadened longitudes for EurAsia (magenta), Central Pacific (blue), NS America (light-blue), and East Atlantic (orange) from Panel (A), but with 4-day broadened data superimposed.
The EPIC dayurnal cycle variability of the short-period ice-cloud sky-fraction for the years 2020 and 2021 in Figure 9B is basically the same as described for the all-cloud sky-fraction in Figure 7B. The main difference is in the enhanced magnitude of the ice-cloud sky-fraction in March of 2020 along the NS America (light-blue) and W. Africa (orange) longitudes. Also, the October to November peak sky-fraction of year 2018 has no counterpart in the year 2021.
4.2 ModelE2 ice-cloud sky-fraction
Given that the cloud property determination for the EPIC Composite cloud product utilizes the CO2 slicing methodology (Minnis et al., 2008; 2011) to determine the cloud-top temperature, and matching the derived cloud-top temperature to geolocated atmospheric temperatures less than 273 K to identify ice clouds, this will result in some intrinsic differences with how ice clouds are determined in the GISS ModelE2. Hence, due to the uncertainties encountered in the remote sensing ice-cloud determination, the ModelE2 ice-cloud distribution will necessarily differ from the EPIC results. By comparison, the GCM ice-cloud information is precisely known within the ModelE2 output data, even if their calculated optical depth is less than a tenth of optical depth. So, in the ModelE2 water/ice phase classification, all of those cases with optically thin cirrus clouds are classified as ice-clouds, whereas in the satellite retrievals, all of those thin cirrus cases would likely be classified as water-clouds.
It is, therefore, not that surprising that the ModelE2 global annual-mean ice-cloud sky-fractions in Figure 10A for the years 2017 and 2018 are 31.82% and 32.20%, respectively, amounting to well over half of the ModelE2 all-cloud sky fraction being covered by ice-clouds. By comparison, for EPIC the ice-cloud sky fraction is only about 1/3 of the all-cloud sky-fraction. Otherwise, the ModelE2 ice-cloud sky-fraction differences are very similar to the ModelE2 all-cloud sky-fraction differences described in Figure 5A. Again, the most striking difference with the EPIC ice-cloud sky-fractions in Figure 8A is the magnitude of the dayurnal cycle amplitude of the ModelE2 global-scale ice-cloud sky-fraction being more than a factor of two larger than that of EPIC during the NH summer months, and then, dropping sharply to less than half in November of 2017. 60-day MJO-type variability if seen along the EUrAsia (magenta) and W. Africa (orange) longitudes from November 2017 to April 2018.
Figure 10. Top panel (A): same as Figure 5, the data source is the GISS ModelE2, except the plot is for daily ice-cloud sky-fraction for years 2017 to 2018, for 24 color-coded longitudes with 30-day resolution. Longitudes are referenced by high-noon GMT, and geographic location, and grouped into 5 color-coded regions: EurAsia (magenta), East Asia (green), Central Pacific (blue), NS America (light-blue), and East Atlantic (orange), and tagged with black-dot symbols at midmonth intervals. Heavy black line is the seasonal global mean. Bottom panel (B): 30-day broadened all-cloud sky-fraction reference longitudes: EurAsia (magenta), Central Pacific (blue), NS America (light-blue), and East Atlantic (orange), with superimposed 4-day broadened data.
On the positive side, the ModelE2 short-period variability in Figure 10B is similar in structure, but a with greater amplitude, particularly during year 2018 over the Central Pacific (blue) longitude, as compared to EPIC. The 15-day variability, and also longer, appear to be the dominant short-period frequencies of the ModelE2 ice-cloud sky-fraction. A series of MJO-type oscillations occur in the first half of year 2018 over the Central Pacific (blue) longitude, also seen with a smaller amplitude in the mid-part of 2018 in the EPIC Figure 8B ice-cloud results. But there is also long-period variability with rather large-amplitude oscillations present in the ModelE2 results that have no counterpart in the EPIC ice-cloud sky-fraction variability.
4.3 EPIC ice-cloud cloud-top height
Since the EPIC Composite cloud property determination utilizes the CO2 slicing methodology (Minnis et al., 2008; 2011) to determine the cloud-top temperature, and then rely on matching the derived cloud-top temperature to geolocated atmospheric temperatures that are colder than 273 K to establish these clouds as ice clouds, there is a systematic selectivity that might include supercooled water-clouds in the ice-cloud category, yet potentially also exclude optically thin cirrus clouds, that might be too difficult to reliably retrieve from satellite orbit. Since the cloud condensation level in atmosphere is highly sensitive to the atmospheric temperature distribution as well as to atmospheric relative humidity, it follows that the cloud-top altitude will also have a strong atmospheric temperature dependence. Also, since accurate cloud altitude determination is of critical importance in calculating LW atmospheric radiative transfer, cloud-top cloud altitude variability is an important climate system variable in both its water-cloud and ice-cloud categories.
Figure 11A shows the dayurnal cycle variability of the 1-day resolution EPIC Composite ice-cloud altitude data with 30-day broadening for the years 2017 and 2018. This yields the close resemblance (except for the orange-to-light-blue color-coding switch between the NS America and the W. Africa longitudes) to the monthly-mean ice-cloud altitude results in Figure 11 of Carlson et al. (2022). In both cases, the NH summer season highest ice-cloud altitudes occur over the continental EurAsia longitudes (magenta), followed by East Asia (green), Central Pacific (blue), W. Africa (orange), and NS America (light-blue), being at substantially lower altitude than the rest. For the October 2017 to April 2018, the ice cloud altitudes remain basically flat at their minimum altitudes, although there is a local maximum occurring in January of 2018. During this time period there appear MJO-type oscillations at different phases, at different latitudes, thus producing some changes in the longitudinal ordering, in particular, the W. Africa longitude (orange) in early 2017 and late 2018, and also the East Asia longitude (green) during the same time periods.
Figure 11. Top panel (A): EPIC-derived daily icel-cloud cloud-top altitude for years 2017 and 2018 for 24 color-coded longitudes (30-day resolution data referenced at high-noon GMT with corresponding geographic location), grouped in 5 color-coded regions: EurAsia (magenta), East Asia (green), Central Pacific (blue), NS America (light-blue), and East Atlantic (orange). Heavy black line depicts the seasonal global-mean. X-scale lists numerical mid-month of year; data use the internal Julian-day scale. Panel bottom bar-graph shows number of EPIC images per day. Bottom panel (B): 30-day broadened longitudes for EurAsia (magenta), Central Pacific (blue), NS America (light-blue), and East Atlantic (orange) from Panel (A), but with 4-day broadened data superimposed.
Figure 11B shows the 1-day resolution EPIC ice-cloud altitude data with 4-day broadening for the years 2017 and 2018. The 4-day broadening has eliminated the 4-day and shorter period variability. There is a short string of what appear to be sustained 10-day oscillations along the Central Pacific longitude (blue) from November to December of 2017. Otherwise there appear to be a variety of short period fluctuations repeating in the 10- to 30-day range, also with a broad range of amplitudes. The 2020 to 2021 La Niña years, except for a slightly enhanced amplitude of the short-period variability during the winter months, are otherwise very similar in their shape to years 2017 and 2018.
5 Hovmöller maps of global variability
Of the different ways to display 2-parameter space/time-data variability, the foregoing line plots provide a quantitative assessment of data variability in terms of the amplitude, frequency, spacing, or duration of the variability. The Hovmöller (1949) format, on the other hand, is designed to emphasize space/time patterns of the data variability in a contour format, where the Y-scale has the data time dependence increasing upward, with the X-scale depicting the longitudinal spatial dependence of the data variability. The EPIC input data to the Hovmöller format contour plots is identical to what was used in the foregoing line plots.
Like the EPIC planetary albedo input data for Figure 1, each plotted pixel in Figure 12 represents global sunlit-hemisphere averages of the EPIC Composite cloud property data points: (1) All-cloud sky-Cover, (2) All-cloud cloud-top Height, (3) All-cloud cloud Optical Depth, and (4) All-cloud cloud Particle Size, with all data points being displayed at their specified time and longitude position. For geographic orientation, a global map is superimposed at panel bottom. Also included is the time track of the Solar Declination, sub-Satellite latitude, and the Sun-Sat phase angle, utilizing the X-axis scale at the top of each panel. With the entire contour map being constructed in quadrants, the maximum, average, and minimum cloud property information for each quadrant is also listed along the Y-axis surrounding the Feb, May, Aug, and Nov labels.
Figure 12. All-cloud year-2017 EPIC Composite cloud properties: Cloud cover fraction, Cloud-top height, Cloud optical depth, Cloud particle size.
The global annual-means from the EPIC Composite cloud property data for 2017 are the following: (1) All-cloud Cover = 60.0%, (2) All-cloud Height = 3.9km, (3) All-cloud OpDepth = 4.9, (4) All-cloud PartSize = 19.4μm, all listed at the bottom-right corner of each figure panel, respectively. Altogether, there were 5,388 sunlit-hemisphere EPIC-image and EPIC Composite cloud property data points for the year 2017, with 13 or 22 images acquired per day, depending on the spacecraft telemetry rate. These were linearly interpolated to define the uniform GMT-based grid for each horizontal line, beginning at the bottom right of the image, then moving westward to the left, as the Earth rotates in the eastward direction, then in the upward direction with each new day of measurements. In this way, a continuous uniformly gridded time-series for the EPIC cloud property global-scale variability is constructed from January 1 to December 31. For those Julian days with missing data (which are identified by red dash lines along the Y-axis at each panel right), the missing data were linearly interpolated at fixed longitude from the prior and following day’s data. As a small refinement, the DSCOVR satellite ephemeris is used to adjust the time-series data points (by linear interpolation) to conform with the position of the sub-satellite longitude to ensure that the data points line up consistently over the same longitude, so that the hourly data point positions correspond precisely to 15° of longitude throughout the year.
While clouds are the principal contributors to the Earth’s planetary albedo, this does not mean that the different cloud parameters will therefore vary in lock-step with planetary albedo variations. All that, and other absorbing and scattering climate system constituents, including the solar zenith angle, are ultimately involved in determining the reflected solar radiation. Thus, as evident from Figure 12, the individual cloud parameters have independent patterns of variability, with occasional correlations as well as anti-correlations. And, regarding climate GCM calculations to match the outgoing solar SW radiation, it would be expected that the corresponding cloud property parameters thus involved would generate variability that resembles the EPIC cloud parameter variability displayed in Figure 12.
The all-cloud sky-fraction is seen to be more variable in the vertical (day-to-day) dimension, rather than longitudinally. The highest cloudy-sky fractions are centered over the Pacific Ocean, practically right along the International Date Line longitude. Roughly 30-day spaced longitudinal extensions of increasing cloud fraction can be discerned to be occurring from mid-January to mid-July. More closely spaced cloud-fraction increases representing shorter-period variability are evident throughout the year. Recounting the line plot results in Figure 3, the minimum cloud-fractions are seen to be distributed along the NS America and W. Africa longitudes, with the deepest minimum cloud-fraction occurring in July along the England and W. Africa longitudes.
With the broad mid-May to September maximum cloud height, centered over the continental EurAsia region, and with the minimum cloud heights occurring simultaneously over the NS America longitudes, depicted previously in Figure 6, the variability of the all-cloud cloud-top height has a very different pattern of variability compared to the other EPIC Composite cloud parameters. It is also a rather curious feature, if not also suspicious, that the all-cloud cloud-top minimum heights seem to be confined to a fairly narrow longitude strip extending from 0° to about 120° W longitude.
Figure 13 displays the EPIC Composite all-cloud property data for year 2018 in Hovmöller format. For the most part the 2018 results are quite similar to those for year 2017. The global annual-means from the EPIC Composite cloud property data for 2018 are the following: (1) All-cloud Cover = 62.1%, (2) All-cloud Height = 3.9km, (3) All-cloud OpDepth = 4.7, (4) All-cloud PartSize = 18.6μm, all listed at the bottom-right corner of each figure panel, respectively. There were 5,351 sunlit-hemisphere EPIC images and EPIC Composite cloud property data points for year 2018, with 13 or 22 images acquired per day, depending on the spacecraft telemetry rate.
Figure 13. All-cloud year-2018 EPIC Composite cloud properties: Cloud cover fraction, Cloud-top height, Cloud optical depth, Cloud particle size.
The most noticeable differences are the somewhat intensified longitudinal width of the all-cloud sky-fraction variability, along with the appearance of the 30-day-spaced MJO-type oscillations from early June through October as well as the 20-day-spaced oscillations appearing from early February through March of 2018. As for the all-cloud cloud-top Height, the broad maximum appears to have diminished in intensity. But there are some 15- to 20-day-spaced features appearing from late February to late March that appear to have counterpart in the all-cloud sky-fraction variability. There is some all-cloud Height intensification at the W. Africa 0° longitude during May and October to November.
Both the all-cloud cloud Optical Depth and Particle Size appear to have diminished in intensity in 2018 compared to 2017. Also, the all-cloud Particle Size maximum appears to have moved slightly to the west. The 2017 narrow longitudinal strip of the all-cloud cloud Particle Size secondary maximum from 45° E to 0°, has disappeared in 2018.
Figure 14 displays the EPIC Composite ice-cloud property data for year 2017 in Hovmöller format. The global annual-means from the EPIC Composite ice-cloud property data for 2017 are the following: (1) Ice-cloud Cover = 21.0%, (2) Ice-cloud Height = 7.23 km, (3) Ice-cloud OpDepth = 4.84, (4) Ice-cloud PartSize = 26.2μm, all listed at the bottom-right corner of each figure panel, respectively. There is general similarity of the 2017 EPIC ice-cloud property variability to the 2017 all-cloud results, but with an overall intensification of the features. The higher amplitude ice-cloud sky-fraction peaks appear to be more accentuated relative to the background level, and the entire cloud-fraction seasonal profile appears to have shifted westward by about 30° in longitude. Overall, a number of the features appear to have a distinct upward slope to the right, indicating an eastward drift of that cloud-fraction feature.
Figure 14. Ice-cloud year-2017 EPIC Composite cloud properties: Cloud cover fraction, Cloud-top height, Cloud optical depth, Cloud particle size.
The broad May to September maximum of the 2017 EPIC ice-cloud cloud-top Height variability has expanded substantially in longitude, and basically eliminated the relative minimum that was present in the all-cloud sky-fraction variability map. Otherwise, the relative intensity of the features within the January to May, and the September to December time periods, has substantially diminished. There are also visible what appear to be a pair of fairly narrow, low intensity ice-cloud cloud-top Height tracks that run vertically along the 90° E and 135° W longitudes.
The 2017 EPIC ice-cloud cloud Optical Depth and cloud Particle Size panels show similar intensity enhancement of their higher amplitude features in the 2017 all-cloud variability panels in Figure 12, with decreased amplitudes for the lesser intensity features. Interestingly, for the ice-cloud optical depth, the relative width of the minimum level ice-cloud optical depths has decreased somewhat, while that of the ice-cloud particle size has substantially widened.
Figure 15 displays the EPIC Composite ice-cloud results for year 2018 in Hovmöller format. For the most part year 2018 results are similar to the year 2017. The global annual-means from the EPIC Composite ice-cloud properties for 2018 are the following: (1) Ice-cloud Cover = 22.0%, (2) Ice-cloud Height = 7.28 km, (3) Ice-cloud OpDepth = 4.5, (4) Ice-cloud PartSize = 25.0 μm, listed at bottom-right corner of each figure panel, respectively. There were 5,351 sunlit-hemisphere EPIC images in 2018. Compared to year 2017, there are more stronger amplitude ice-cloud sky-fraction features in year 2018. There are longitudinally brief 30-day MJO-type oscillation occurring from May to September along the International Date Line. Higher frequency 15-day oscillations are prevalent from January to May. As the case for ice-cloud sky fraction in year 2017, the ice-cloud sky fraction for year 20178 appears to be shifted westward by about 30° in longitude.
Figure 15. Ice-cloud year-2018 EPIC Composite cloud properties: Cloud cover fraction, Cloud-top height, Cloud optical depth, Cloud particle size.
Except for a small enhancement in strength, the broad maximum of the ice-cloud cloud-top Height in 2018 is basically unchanged from 2017. The two parallel pair of low intensity ice-cloud cloud-top Height tracks that run vertically along the 90° E and 135° W longitudes, still remain discernable. But for the 2018 ice-cloud Optical Depth and Particle Size panels, there is a virtually uniform decrease in the absolute magnitude of the variability across both panels.
6 Discussion
The chief motivation for this study was to assess the prospects of using 1-day resolution variability comparisons as the means for providing more direct model/data diagnostic inter-comparisons. Because of the need to average out the uncorrelated quasi-chaotic behavior of both the real-world and the GCM generated output data, while simultaneously coordinating the space-time data resolution, diurnal cycle timing, and viewing geometry compatibility, such model/data comparisons are difficult to perform. With the common practice of using monthly-mean averaging to eliminate the quasi-chaotic weather noise, all climate system variability at frequencies shorter than 30 days is lost.
This is where DSCOVR mission EPIC and NISTAR measurements of the Earth’s sunlit hemisphere acquired from the Lagrangian L1 vantage point, provide the best means currently available to resolve this dilemma in model/data diagnostic comparisons. Averaging over the sunlit hemisphere of the EPIC data, along with the similar aggregation of climate GCM diagnostic data using the DSCOVR satellite ephemeris to account for the viewing geometry, effectively eliminate the meteorological weather noise, while maintaining precise diurnal cycle timing and viewing geometry coordination.
Variability is a pervasive characteristic of the terrestrial climate system. All measurable climate system parameters and climate system phenomena vary constantly and interact with each other, being both the cause and the result of these interactions in a changing climate. These ongoing climate system causes and effects can be classified in terms of climate forcings and feedbacks, some known precisely, others less so. For example, the daily and seasonal changes in solar irradiance are known accurately at any given location on the Earth, but the responses of the different climate system constituents to the everchanging radiative forcing are far more diversified, and thus, far more difficult to isolate. And they may also involve significant time delays, thresholds, and discontinuous transformations. Some exhibit characteristic variability signatures in terms of their frequency of occurrence, duration, amplitude, and/or their size and geographic location.
The climate system’s variability persists because feedback responses occur at a characteristic size, magnitude, and/or frequency, which does not diminish in direct proportion to how close the climate system may be in approaching to its equilibrium point. Thus, as the climate system continues to move toward global energy balance, the magnitude of the feedback reactions is always large enough to overshoot the equilibrium point of global energy balance. This inability to incrementally converge to equilibrium, produces a continuing stream of quasi-chaotic fluctuations about the equilibrium point, thus constituting the ‘natural’ variability of the climate system.
Obtaining similar variability in climate GCM output data is therefore a strong indicator measuring the success of the GCM parameterization of the physical processes to reproduce the complex behavior of the Earth’s climate system. The uncorrelated quasi-chaotic variability (and the additional potential differences in viewing geometry, diurnal cycle timing, and spatial collocation) make quantitative model/data comparisons problematic. Nevertheless, the characteristics of this variability also have the potential to serve as a potent climate-model performance diagnostic.
Successive matching of more demanding comparisons to observations has been the indispensable driving force behind climate model development. In our early days of climate model development, it was no small feat to get the January and July climate simulations to resemble observations (Somerville et al., 1974). A decade later, the GISS Model II (Hansen et al., 1983), made it possible to simulate the global climate variability of the Earth’s climate system with more realism–sufficient to obtain a climate sensitivity of 4° C for doubled CO2, and to identify the principal feedback contributions due to water vapor, clouds, and snow/ice albedo (Hansen et al., 1984). Since climate is primarily a boundary value problem in physics, it is not all that surprising that a reasonable climate sensitivity would be obtained, even with the course spatial resolution of the GISS Model II, by enforcing conservation of energy, and by modeling the physical and radiative processes with adequate accuracy.
Different ModelE versions of the GISS climate GCM have had a wide range of applications such as reproducing the documented changes in the global surface temperature of the Earth over the past century (Hansen et al., 2002), and assessing the relative radiative forcing efficacy of contributing climate system constituents (Hansen et al., 2005). Further refinements and capabilities of the GISS climate GCM are described by Schmidt et al. (2006) for the reference ModelE version, and by Schmidt et al. (2014) for upgraded ModelE2 version. Climate GCMs are ‘tuned’ to reproduce the annual mean planetary albedo (Kopp and Lean, 2011; Loeb et al., 2018), by adjusting the critical relative humidity threshold for the onset of cloud condensation (Del Genio and Yao, 1993; Del Genio et al., 1996). A critical relative humidity threshold is used on the rationale that a temperature-humidity distribution exists within climate GCM grid boxes, and that cloud condensation will therefore occur well before the grid-box-mean relative humidity reaches the 100% level.
The basically good agreement between model results and observational data for global quantities is not all that surprising or remarkable. This is because horizonal transports and conversions of energy are all expected to average to zero in the real world and also in climate GCMs. Thus, there is reason to expect that changes in quantities like the global-mean ground surface temperature will be rendered with reasonable accuracy if the modeled radiative global energy balance of the Earth is in reasonable agreement with observations. How good the modeled surface temperatures are on a regional or local basis still remains as an open question. In an early doubled CO2 experiment with the GISS GCM, the magnitude of the regional advective feedback changes due to latent heat, sensible heat, and geopotential energy transports, including their latitudinal dependence, were seen to be an order of magnitude larger than those for the radiative feedback and forcing changes due to water vapor, clouds, and greenhouse gas changes (Lacis, 2018), with no feasible options for direct observational verification.
Achieving accurate climate sensitivity has always been the leading key objective in climate model development. And in this regard, treatment of cloud processes in climate GCMs is the leading candidate for controlling the climate sensitivity of a GCM. A recent study by Hansen et al. (2023) suggests that the inferred climate sensitivity, based on a detailed analysis of the geological record, is substantially higher than that proclaimed by most of the current climate models. Hansen et al. note that the variability of the early versions of ModelE2, because of insufficient ENSO activity, was substantially too low compared to observations, but that in a more recent version with enhanced ENSO treatment, the variability was now substantially higher than observations. As a result, climate GCM cloud treatment and variability have become leading candidates for observational scrutiny.
Our prior effort in model/data variability comparison (Carlson et al., 2022), using monthly-mean compilations of global EPIC fluxes demonstrated convincingly that the GISS ModelE2, significantly overestimated cloud cover over the oceans, while underestimating clouds over continental land areas. It was concluded that using a globally uniform relativity criteria for cloud condensation onset was the principal cause for the longitudinal planetary albedo mismatch. This longitudinal variability mismatch in planetary albedo may already have been resolved by the recent implementation of a more physics-based cloud scheme in the new version of the GISS ModelE3.
There was also a curious “standing wave pattern” in that earlier study, located in the Pacific Ocean near the International Date Line, in a Hovmöller ratio plot. When examined more closely using 1-day resolution EPIC data, it turned out to be a data artifact, missing data in the monthly-mean averaging being the cause. It also became clear that the monthly-mean averaging was obliterating the short-period variability in both the observational and in the GCM generated data. Missing data needed to be filled in carefully, based on their closest space-time neighboring data points. That alone was sufficient to make the data artifact disappear. This analysis led directly to displaying the EPIC and GCM data using 1-day resolution, and thus making the short-period variability, ongoing in the climate system, available for more informative model/data comparisons.
This also led us to include the NISTAR full-disk 1-Hetrz silicon photodiode measurements in these 1-day resolution model/data comparisons. The remarkably close agreement between the EPIC and the NISTAR space-time variability over 7 years of data, has established beyond doubt, as reported in our companion paper (Lacis et al., 2025), that the variability displayed in Figure 1 is actually real, and not some kind of data artifact.
This establishes a new paradigm in diagnostic model/data comparisons that make it possible to involve variability of the climate system on a broad range of frequencies that has not been feasible with conventional monthly-mean satellite and GCM generated output data. The short-period variability is specifically related to cloud process variability involving cloud and storm system life cycles, and due to the precise longitudinal slicing by the Earth’s rotation, to effectively analyze the cloud and storm system longitudinal distribution and space-time variability.
But there are some limitations and problem areas. The planetary albedo data are the most robust of the EPIC data products, since the EPIC-derived SW reflected radiative fluxes are wholistic variables, and are in good agreement with CERES global radiative flux determinations. The all-cloud sky-fraction is the next most robust climate variable of the EPIC Composite data product, since it only depends on a simple minimum cloud optical depth that is retrievable from satellite measurements, and perhaps also some differentiation from aerosol scattering. The empirical orthogonal function spectral analysis by Li et al. (2015) found that only the all-cloud sky-fraction and cloud-top altitude variability was robust among the different cloud property determinations.
Directly relevant to the study here in regard to the EPIC-derived cloud-fraction changes, the study by Delgado-Bonal et al. (2021) of the diurnal cycle cloud fraction variability, as opposed to the present dayurnal cycle global-scale cloud fraction variability of this study. The name distinction is important. The diurnal cycle refers the cloud-fraction change at a specified location during the course of the day. Based on EPIC image analysis of the daytime cloud-fraction evolution compiled at 1-hour intervals, Delgado-Bonal et al. show that for the case of liquid water-clouds, the diurnal cloud-fraction changes go in opposite directions for clouds over land, and for clouds over the ocean. Over land, water-clouds reach their maximum cloud-fraction at local noon, while over the oceans, the water-clouds reach their minimum cloud-fraction also at local noon, and that ice-clouds show no dependence on land or ocean. In the case of the dayurnal cycle, cloud-fractions are always averaged over the entire sunlit hemisphere to average out the chaotic variability. In this way, a single climate-style data point is obtained for that sub-satellite longitude, with the rotation of the Earth establishing the dayurnal cycle collecting such data points at hourly intervals to cover all longitudes, which implies a complex diurnal cycle average at each dayurnal cycle data point. Thus, over land, for the sub-satellite longitude it would be noon-time with its maximum cloud-fraction, and with neighboring longitude cloud-fractions decreasing in both the eastward and westward directions, while over ocean, the sub-satellite longitude would be at its minimum cloud-fraction, with neighboring longitude cloud-fractions increasing in both eastward and westward directions.
In the ModelE2 diagnostics package, detailed information on the GCM cloud properties is known precisely. But the GCM cloud optical depth limit of 10−3 is far below the satellite cloud detection limit. It is not known how frequently, or how many, such super-optically-thin cirrus clouds might occur as the only cloud in the column. Even with the possibility of such contributions to the ModelE2 all-cloud sky-fraction, the fact that the ModelE2 global annual-mean all-cloud sky-fraction is 55%, compared to the EPIC over 60% value, is a clear indication that ModelE2 underestimates the all-cloud sky-fraction by about 10%. This could just be a result from the coarse-grid (4° x 5°) resolution, but in any case, a closer estimate of this difference could be easily determined by including a minimum cloud optical depth threshold in the GCM aggregation process that is more in line with that of the satellite retrievals. It is not that optically thin clouds do not exist in the real world, just because they are not being detected by satellite measurements. Similarly, the optically thin cirrus clouds, generated by the ModelE2 cloud modeling, do not dominate the global radiation budget, but have their radiative effects appropriately evaluated and accounted for by the ModelE2 radiation model.
The cloud water/ice phase determination is another key area that is of interest that impacts climate sensitivity because of the differing water-cloud and ice-cloud radiative properties. Here there is an even greater difference between EPIC and ModelE2 results. The ModelE2 ice-cloud sky-fraction comprises 58% of the GCM all-cloud sky-fraction, compared to only 34% of the EPIC all-cloud sky-fraction. No doubt a significant portion of the EPIC-ModelE2 difference is due to optically thin high-altitude cirrus clouds putting the entire cloud column in the ice-cloud category. The ModelE2 radiation calculations take the individual atmospheric cloud layer radiative parameter fully into account, so it is not the case that an optically thin cirrus cloud would redefine the radiative properties of the entire cloud column. But for the sake of a clearer understanding of the effective ice-cloud-to-all-cloud ratio, here again, using the information available to the GCM diagnostics package, the ice-to-water fraction could be determined for the cloud optical depth of unity, measured from the top, to categorize the entire column as being water, or ice, in the GCM aggregation process of computing the sunlit hemisphere average for the purpose of cloudy sky-fraction determination.
Also relevant to this study is the cloud diurnal cycle daytime variability based on EPIC and GOES-R/ABI observations (Delgado-Bonal et al., 2022). Delgado-Bonal et al. find that low clouds exhibit their minimum altitude around local noon-time, while high-altitude clouds show a steady increase in height from morning until evening, while also finding that the diurnal cycle to be weaker over oceans compared to land. In our dayurnal cycle-based study, the diurnal cycle cloud variations are averaged instantaneously over the sunlit hemisphere, with identical sampling of the diurnal cycle dependence in the EPIC-image averaging and in the GCM diagnostic data aggregation using the DSCOVR satellite ephemeris information to account for the EPIC viewing geometry.
Even more problematic for effective self-consistency in model/data comparisons than definition of the precise cloud-top altitude, are the definition issues for the EPIC and ModelE2 cloud optical depth and cloud particle size comparisons. Ultimately, the problem is that for any tractable radiative transfer computation, plane-parallel geometry with homogeneous cloud particle distributions is required. Then for input to the Mie scattering calculations a cloud particle size distribution for effective radius and effective variance is needed to define the effective scattering matrix, or asymmetry parameter that then goes with the optical depth to compute the radiative properties of the cloud layer. Since the real world does not conform to all those constraints, parameterized “effective” values need to be assumed for the relevant parameters, and the assumptions that are made the different satellite retrievals, are different from the parameterizations used in GCM radiative modeling.
It would appear that it is simply the extreme heterogeneity of terrestrial clouds, their viewing geometry dependence, and modeling difficulty that could explain the Li et al. (2015) EOF analysis result of agreement between the different satellite retrievals for cloud sky-fraction and cloud-top altitude, but not for cloud optical depth or cloud particle size. A more recent comprehensive reanalysis of cloud spatiotemporal characteristics by Yao et al. (2020) tends to confirm that assessment.
Since the respective definitions for cloud optical depth or cloud particle size between the EPIC Composite cloud parameters and their ModelE2 counterparts are not yet fully reconciled, these GCM cloud parameters have not been included for explicit variability inter-comparisons. Nevertheless, for completeness, the full set of the EPIC Composite cloud parameters are displayed in the Hovmöller contour maps for the all-cloud and ice-cloud cases for the years 2017 and 2018. Not only do these Hovmöller maps show the different space-time variability patterns for the different cloud parameters, they may also provide a kind of data quality check of the EPIC cloud parameters in that there is a suspiciously persistent vertical alignment in longitude (or actually GMT time) of the cloud-top height, as well as the cloud optical depth and cloud particle size variability.
In any case, the EPIC Composite cloud data product represents a very effective synthesis of satellite retrieved cloud properties, collected in an ideal format for precise model/data comparison to similarly selected GCM diagnostic output data, and available for precise longitudinal slicing comparisons of the global cloud-property variability with 1-day resolution. The short-period cloud variability that was hidden from view by monthly-mean data averaging is now available for direct model/data comparison with climate GCM diagnostic output data similarly aggregated over the sunlit hemisphere to suppress the quasi-chaotic meteorological weather noise.
But there are also some obvious improvements that could be made. Data aggregation over the sunlit hemisphere, while effective in suppressing the chaotic weather noise, it also eliminates the latitudinal information content. From the GCM perspective, it would be a relatively simple task to tabulate the sub-totals of the data aggregation for specified latitude intervals (e.g., equal area quadrants), at the price of somewhat less efficient weather noise suppression, but would provide latitudinal information of data variability on hemispheric, or quadrant basis. This is of some significant importance, since the inter-hemispheric variability in ModeE2 simulations is substantially larger than observations indicate (e.g., Stephens et al., 2015). Also, the study by Bender et al. (2017), based on 13 years of CERES and MODIS data, finds substantial latitudinal differences in the variability of the tropical, subtropical, and mid-latitude cloud fraction and planetary albedo.
There are many further comparisons that could be made. There are literally hundreds of climate system variables that are routinely tabulated by the GCM diagnostics package for comparison to observations, and all of these variables could be included in the DSCOVR satellite ephemeris-based data aggregation over the sunlit hemisphere in support of each other for their global-scale variability comparisons. Thus, for example, 1-day resolution Hovmöller maps of the cloud-top temperature and of the atmospheric temperature pressure levels could readily confirm the likelihood that large summer season changes in the atmospheric temperature profile over continental land areas explain the broad summer-season maximum of the ice-cloud cloud-top variability. Likewise, cloud properties and the surface temperature could be sampled separately, or together, over land and ocean surfaces for further differentiation of the circumstances of the short-period and long-period surface temperature and cloud parameter variability.
Perhaps requiring a somewhat greater programming effort due to format differences for integrating over the visible disk area of EPIC images, latitude zone sub-totals and surface-type differentiation could also be implemented in the EPIC Composite parameter integration over the sunlit hemisphere. The ancillary data used in the EPIC Composite analysis (surface types, skin temperature, precipitable water, perhaps also CERES LW fluxes) could also be tabulated along with the SW flux and cloud property data for more comprehensive variability comparisons with similarly tabulated GCM climate system variables.
On a rotating planet, classical wave theory predicts Rossby waves with 2-, 5-, 10-, and 16-day periodicity (e.g., Yamazaki and Matthias, 2019). Zhang et al. (2022) note 10-20, 20-30, and 30-60-day low frequency variability observed in northern China. Though perhaps the most prominently discussed are the 30-day MJO oscillations (Madden and Julian, 1972). All of these frequencies, and also others, are present on a global-scale at all longitudes and times of year in the 1-day resolution EPIC data. The short-period variability is related cloud process, and was effectively hidden from view by the common utilization of monthly-mean averaging in typical model/data comparisons. It is encouraging that in the EPIC/ModelE2 comparisons, there was good agreement for the short-period cloud-fraction variability, suggesting the ModelE2 local-scale cloud treatment was operating properly, even though much of the other cloud related variability differences were significantly than for the EPIC results.
7 Summary
NASA’s DSCOVR Mission EPIC and NISTAR measurements are uniquely positioned to resolve a long-standing inadequacy in model/data comparisons to evaluate climate GCM performance against observational data. It has been long-standing practice to use monthly-mean averages to suppress the uncorrelated quasi-chaotic weather noise that affects both observations and the GCM output data. The main consequence of using this approach is that the short-period variability of the climate system is also eliminated from consideration. Cloud processes and storm-system life-cycles fall in this category of short-period variability, and thus beyond reach of being effectively considered in such model/data diagnostic inter-comparisons.
What makes the DSCOVR Mission EPIC and NISTAR data unique, is that the vantage point of their measurements is from the Lissajous orbit around the Lagrangian L1 point, which gives access to view the entire sunlit hemisphere of the Earth, and enable the unique ability to monitor the entire solar energy input to the Earth’s climate system by measuring the globally reflected solar SW radiation, with the rotation of the Earth providing precise registration of the longitudinal dependence of the Earth’s global energy input.
From a technical perspective, the key advantage of the DSCOVR Mission data is that by integrating the reflected solar radiation over the sunlit hemisphere, and converting the backscattered radiance to a radiative flux (which was done to help calibrate the NISTAR measurements), a climate-style data point of the total global energy reflected by Earth is created at that particular sub-satellite longitude of the EPIC image, with the chaotic weather noise systematically suppressed. The very same approach is sell-designed to work in the GCM setting. By using DSCOVR satellite ephemeris information to weight the contributions from each GCM grid-box to the sunlit-hemisphere aggregate to be in accord with the DSCOVR satellite viewing geometry, a climate-style data point of the total reflected global energy is obtained. The potential problem of high spatial resolution observational data vs. low spatial resolution model data is virtually eliminated since both climate-style data points represent the total reflected solar energy by the sunlit hemisphere for precisely the same viewing geometry, and precisely the averaging over the diurnal cycle, and both with the chaotic weather noise suppressed in the same fashion.
All that points to a successful demonstration showing that it makes good sense to replace the conventional practice of monthly-mean data averaging, which while suppressing the chaotic weather noise, also eliminates the short-period variability from consideration. In its place, the 1-day resolution DSCOVR Mission style sunlit-hemisphere averaged longitudinal slicing methodology, can be the new paradigm for a quantitatively improved climate GCM diagnostic capability of model/data comparison. But this is not the end product. This is only the first step toward expanding this methodology to retain well-defined latitudinal information by simply utilizing latitudinally spaced subtotals within the sunlit hemispheric data aggregation process. Furthermore, there are hundreds of climate GCM diagnostic variables that can also be sampled and aggregated in the same way as the radiative flux and the different cloud properties analyzed in this study.
However, there is no particular assessment of the accuracy of satellite data products here. The EPIC Composite cloud data product is at best a consensus that is assembled from all contributing LEO and GEO satellite retrievals, and in that sense, EPIC cloud properties are neither more, nor less, accurate than those of its contributors. And in that regard, the all-cloud sky-fraction is the basically only cloud property that is directly comparable to climate GCM data. All of the other cloud products require significant improvement in coordinating their satellite retrieval assumptions versus the GCM parameterization assumptions.
The key point of the EPIC cloud data is their sunlit hemisphere averaged 1-day resolution, which together with the precise longitudinal dependence due to the Earth’s rotation, enables continuous monitoring of the global-scale variability with the chaotic weather noise suppressed, and with precise viewing geometry and diurnal cycle sampling possible with similarly aggregated GCM data. Although variability is not a direct measure of cloud feedback, there is a close relationship since clouds are the principal tracers of atmospheric dynamics. And in that regard, it may be both a concern and perhaps also a reassurance, that the large longitudinal cloud distribution errors identified in the GISS ModelE2 did not appear to adversely impact the model’s ability to accurately simulate global climate change, although on an annual global-mean basis.
Going forward, climate GCMs will need to correctly simulate the regional, seasonal, and local variability of a changing climate system. Cloud feedbacks, cloud radiative forcings, and atmospheric dynamics are mutually manifest on multiple timescales and a wide range of spatial scales. Variability may be a somewhat nebulous concept, but all of the different physical processes in the climate system exhibit characteristic time and spatial scales over which they operate. Accordingly, the well-defined global-scale space-time variability of the EPIC reflected SW radiation and associated cloud property data are an important, and currently unique source of observational constraints for continued climate GCM development.
Basically, if a climate GCM were to accurately simulate the regional space-time variability of the climate system for multiple data products, that would imply that the GCM programming of the different physical processes has good accuracy, and that the likelihood of accurate prediction of future climate change consequences might likewise be greatly enhanced. Though slow and tedious, this remains the only option for a reliable assessment and understanding of the cloud and climate feedback sensitivity.
Data availability statement
The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author.
Author contributions
GR: Writing – original draft, Writing – review and editing, Software, Formal Analysis, Methodology. AL: Writing – review and editing, Writing – original draft, Conceptualization, Visualization. BC: Conceptualization, Methodology, Writing – review and editing, Writing – original draft. WS: Writing – review and editing, Writing – original draft, Data curation. JP: Writing – original draft, Methodology, Writing – review and editing.
Funding
The authors declare that financial support was received for the research and/or publication of this article. Funding for this research investigation has been provided by the NASA DSCOVR Project through WBS 510937.05.80.01.02.
Acknowledgements
The authors express their gratitude to the NASA Science Research Division managed originally by Jack Kaye, and to the NASA DSCOVR Program managed originally by Richard Eckman, and currently by Emma Knowland, for their continued encouragement and support.
Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The authors declare that no Generative AI was 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
Bender, F. A.-M., Engstrom, A., Wood, R., and Charlson, R. J. (2017). Evaluation of hemispheric asymmetries in marine cloud radiative properties. J. Clim. 30, 4131–4137. doi:10.1175/JCLI-D-16-0263.1
Bodas-Salcedo, A., Hill, P. G., Furtado, K., Williams, K. D., Field, P. R., Manners, J. C., et al. (2016). Large contribution of supercooled liquid clouds to the solar radiation budget of the Southern Ocean. J. Clim. 29, 4213–4228. doi:10.1175/JCLI-D-15-0564.1
Carlson, B., Lacis, A. A., Russell, G., Marshak, A., and Su, W. (2022). Unique observational constraints on the seasonal and longitudinal variability of the Earth’s planetary albedo and cloud distribution inferred from EPIC measurements. Front. Remote Sens. 2, 788525. doi:10.3389/frsen.2021.788525
Del Genio, A. D., and Yao, M.-S. (1993). “Efficient cumulus parameterization for long-term climate studies: the GISS scheme,” in The representation of cumulus convection in numerical models. AMS meteorological monograph 46. Editors K. A. Emanuel, and D. A. Raymond (American Meteorological Society), 181–184.
Del Genio, A. D., Yao, M.-S., Kovari, W., and Lo, K. (1996). A prognostic cloud water parameterization for global climate models. J. Clim. 9, 270–304. doi:10.1175/1520-0442(1996)009<0270:APCWPF>2.0.CO;2
Delgado-Bonal, A., Marshak, A., Yang, Y., and Oreopoulos, L. (2021). Global daytime variability of clouds from DSCOVR/EPIC observations. Geophys. Res. Lett. 48, e2020GL091511. doi:10.1029/2020GL091511
Delgado-Bonal, A., Marshak, A., Yang, Y., and Oreopoulos, L. (2022). Cloud height daytime variability from DSCOVR/EPIC and GOES-R/ABI observations. Front. Remote Sens. 3, 780243. doi:10.3389/frsen.2022.780243
Hansen, J., Russell, G., Rind, D., Stone, P., Lacis, A., Lebedeff, S., et al. (1983). Efficient three-dimensional global models for climate studies: models I and II. Mon. Weather Rev. 111, 609–662. doi:10.1175/1520-0493(1983)111<0609:etdgmf>2.0.co;2
Hansen, J., Lacis, A. R., Russell, G., Stone, P., and Fung, I. (1984). “Climate sensitivity: analysis of feedback mechanisms,” in Climate processes and climate sensitivity. AGU geophysical monograph 29, maurice ewing. Editors J. E. Hansen, and T. TakahaNielsen (Washington, DC: American Geophysical Union), 5, 130–163.
Hansen, J., Sato, M., Nazarenko, L., Ruedy, R., Lacis, A., Koch, D., et al. (2002). Climate forcings in goddard institute for space studies SI2000 simulations. J. Geophys. Res. 107 (D18), 4347. doi:10.1029/2001JD001143
Hansen, J., Sato, M. R., Nazarenko, L., Lacis, A., and Schmidt, G. A. (2005). Efficacy of climate forcings. J. Geophys. Res. 110, D18104. doi:10.1029/2005JD005776
Hansen, J. E., Sato, M. S. , L., Nazarenko, L. S., Sangha, I., and Kharecha, P. (2023). Global warming in the pipeline. Oxf. Open Clim. Change 3 (1), kgad008. doi:10.1093/oxfclm/kgad008
Hovmöller, E. (1949). The trough-and-ridge diagram. Tellus 1 (2), 62–66. doi:10.1111/j.2153-3490.1949.tb01260
Hu, Y., Rodier, S., Xu, K., Sun, W., Huang, J., Lin, B., et al. (2010). Occurrence, liquid water content, and fraction of supercooled water clouds from combined CALIOP/IIR/MODIS measurements. J. Geophys. Res. 115, D00H34. doi:10.1029/2009JD012384
Jones, C., and Carvalho, L. M. V. (2002). Active and break phases in the South American monsoon system. J. Clim. 15, 905–914. doi:10.1175/1520-0442(2002)015<0905:aabpit>2.0.co;2
Kopp, G., and Lean, J. (2011). A new, lower value of total solar irradiance: evidence and climate significance. Geophys. Res. Lett. 38, L01706. doi:10.1029/2010gl045777
Lacis, A. A. (2018).“Explaining climate. In our warming planet: topics in climate dynamics,”. Lectures in climate change. Editors C. Rosenzweig, D. Rind, A. Lacis, and D. Manley (World Scientific #EP 27), 1, 3–27. doi:10.1142/9789813148796_0001
Lacis, A. A., Russell, G., Carlson, B., Su, W., and Yu, Y. (2025). NISTAR photodiode measurements confirm EPIC-derived seasonal global-scale dayurnal variability of Earth’s reflected radiation. Front. Remote Sens.
Li, J., Carlson, B. E., Rossow, W. B., Lacis, A. A., and Zhang, Y.-C. (2015). An intercomparison of the spatio-temporal variability of satellite- and ground-based cloud datasets using spectral analysis techniques. J. Clim. 28 (14), 5716–5736. doi:10.1175/JCLI-D-14-00537.1
Loeb, N. G., Doelling, D. R., Wang, H., Su, W., Nguyen, C., Corbett, J. G., et al. (2018). Clouds and the Earth’s radiant energy system (CERES) energy balanced and filled (EBAF) top-of-atmosphere (TOA) data product. J. Clim. 31, 895–918. doi:10.1175/jcli-d-17-0208.1
Madden, R. A., and Julian, P. R. (1972). Description of global-scale circulation cells in the tropical Pacific. J. Atmos. Sci. 29, 1109–1123. doi:10.1002/cmdc.202500847
Marshak, A., Herman, J., Szabo, A., Blank, K., Simon, C., Cede, A., et al. (2018). Earth observations from DSCOVR EPIC instrument. Bull. Am. Meteorological Soc. 99, 1829–1850. doi:10.1175/BAMS-D-17-0223.1
Marshak, A., Delgado-Bonal, A., and Knyazikhin, Y. (2021). Effect of scattering angle on Earth reflectance. Front. Remote Sens. 2, 2719610. doi:10.3389/frsen.2021.719610
Minnis, P., Nguyen, L., Palikonda, R., Heck, P. W., Spangenberg, D. A., Doelling, D. R., et al. (2008). “Near-real time cloud retrievals from operational and research meteorological satellites,” in Proc. SPIE 7108, remote sens. Clouds atmos. XIII (Cardiff, Wales, UK), 710703–710708. doi:10.1117/12.800344
Minnis, P., Sun-Mack, D. F., Young, S., Heck, P. W., Garber, D. P., Chen, Y., et al. (2011). CERES Edition-2 cloud property retrievals using TRMM VIRS and TERRA and AQUA MODIS data, part I: algorithms. IEEE Trans. Geoscience Remote Sens. 49 (11), 4374–4400. doi:10.1109/tgrs.2011.2144601
Ross, J. K., and Marshak, A. L. (1988). Calculation of canopy bidirectional reflectance using the monte Carlos method. Remote Sens. Environ. 24, 213–225. doi:10.1016/0034-4257(88)90026-0
Schmidt, G. A., Ruedy, R., Hansen, J. E., Aleinov, I., Bell, N., Bauer, M., et al. (2006). Present day atmospheric simulations using GISS ModelE: Comparison to in-situ,, satellite and reanalysis data. J. Clim. 19, 153–192. doi:10.1175/JCLI3612.1
Schmidt, G. A., Kelley, M., Nazarenko, L., Ruedy, R., Cook, B. I., Russell, G. L., et al. (2014). Configuration and assessment of the GISS ModelE2 contributions to the CMIP5 archive. J. Adv. Model. Earth Syst. 6, 141–184. doi:10.1002/2013MS000265
Somerville, R. C. J., Stone, P. H., Halem, M., Hansen, J. E., Hogan, J. S., Druyan, L. M., et al. (1974). The GISS model of the global atmosphere. J. Atmos. Sci. 31, 84–117. doi:10.1175/1520-0469(1974)031<0084:TGMOTG>2.0.CO;2
Stephens, G. L., O'Brien, D., Webster, P. J., Pilewski, P., Kato, S., and Li, J. (2015). The albedo of Earth. Rev. Geophys. 53, 141–163. doi:10.1002/2014RG000449
Su, W., Liang, L., Doelling, D. R., Minnins, P., Duda, D. P., Khopenkov, K., et al. (2018). Determining the shortwave radiative flux from Earth Polychromatic Imaging Camera. J. Geophys. Res. 123, 11479–11491. doi:10.1029/2018JD029390
Su, W., Liang, L., Duda, D. P., Khlopenkov, K., and Thieman, M. (2021). Global daytime mean shortwave flux consistency under varying EPIC viewing geometries. Front. Remote Sens. 2, 747859. doi:10.3389/frsen.2021.747859
Yamazaki, Y., and Matthias, V. (2019). Large-amplitude quasi-10-day waves in the middle atmosphere during final warmings. J. Geophys. Res. 124, 9874–9892. doi:10.1029/2019JD030634
Yao, B., Teng, S., Lai, R., Xu, X., Yin, Y., Shi, C., et al. (2020). Can atmospheric reanalyses (CRA and ERA5) represent cloud spatiotemporal characteristics? Atmos. Res. 244 (1 November 2020), 105091. doi:10.1016/j.atmosres.2020.105091
Keywords: DSCOVR/EPIC, cloud fraction, cloud radiative properties, longitudinal slicing, variability profile, dayurnal-cycle, climate-GCM constraints, global-scale variability
Citation: Russell GL, Lacis AA, Carlson BE, Su W and Pilewskie JA (2025) Global-scale seasonal variability profiles of EPIC-derived vs. GISS ModelE-simulated all-cloud and ice-cloud fraction distributions. Front. Remote Sens. 6:1691948. doi: 10.3389/frsen.2025.1691948
Received: 24 August 2025; Accepted: 19 November 2025;
Published: 18 December 2025.
Edited by:
Alexei Lyapustin, National Aeronautics and Space Administration, United StatesReviewed by:
Chao Liu, Nanjing University of Information Science and Technology, ChinaAlfonso Delgado Bonal, Universities Space Research Association (USRA), United States
Copyright © 2025 Russell, Lacis, Carlson, Su and Pilewskie. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Andrew A. Lacis, YW5kcmV3LmEubGFjaXNAbmFzYS5nb3Y=
†Present address: Juliet A. Pilewskie, Colorado State University, Fort Collins, CO, United States
‡Retired