the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 18 May 2026
Acceleration After 2020 of Decreasing Earth Albedo from DSCOVR-EPIC 2015–2025 Compared with CERES 2000–2025
Abstract. The 10-year Earth albedo time series, its seasonal and wavelength dependence (10 wavelengths from 317.5 to 779.5 nm) of the illuminated Earth are estimated from radiance measurements by the Earth Polychromatic Imaging Camera EPIC located in an orbit around the Earth-Sun Lagrange-1 point L1. The measurements were made at a backscattered observing phase angle ranging from 2° to 12°. The maximum albedo occurs at 340 nm, ranging from 0.33 in June to 0.38 in December. For shorter wavelengths, 325 and 317.3 nm, the albedo decreases sharply because of ozone absorption. For wavelengths longer than 340 nm (388, 443, and 551 nm) the albedo decreases gradually until about 680 nm where reflectivity from vegetation increases (680 and 779.5 nm). The maximum albedo occurs in December when the illuminated Antarctic ice sheet is observed. The EPIC solar flux weighted annual mean albedo in 2016 is estimated to be 0.298 (317.5 to 779.5 nm) and 0.293 in 2025, (-1.68% change) compared to the Clouds and the Earth's Radiant Energy System (CERES) 2016 values of 0.289 and 0.284 in 2025 (-1.73% change) for the spectral range 300 to 5000 nm. The long-term trend of CERES albedo RAC (2000–2025.7) = -0.094±0.0007% yr-1 for a total linear change of 2.35% over 25 years, However, the albedo shows an accelerating decline after 2020, RAC (2020–2025.7) = 0.27 ± 0.001% yr-1 compared to RAC (2000–2020) -0.09 ± 0.0006% yr-1. Because the decline is non-linear, the total 25-year change is -2.7%. EPIC observes similar rates of albedo decline, RAE (2016–2019.5) = -0.18 ± 0.0006% yr-1 and RAE (2020.3–2025) = 0.25 ± 0.001% yr-1 validating the increased rate of albedo decline after 2020. This suggests that the accelerating reduced Earth reflection of sunlight back to space is enhancing global warming.
- Preprint
(7816 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2026-1631', Vincent Kofman, 17 Jun 2026
-
AC1: 'Reply on RC1', Jay Herman, 18 Jun 2026
- Dark side of the moon typically refers to the unseen portion of the moon, not the part that is lit by Earthshine. Revised to read “Photometric measurements are made of sunlight reflected from Earth to the dark portion of the crescent Moon and back to ground‑based telescopes”
Fig 7. Would suggest to keep the you axis scale the same throughout the panels to facilitate easier inter comparison between the brightness at the different times of year. Done
Sentence added: The EPIC measured Earth albedo includes the changing Sun-Earth distance.
- Could you elaborate a bit on why the albedo between CERES and EPIC are different? Is this mostly due to different viewing orientations or lower reflectivity in the near-to-mid infrared?
There are several differences in the two measurements that lead to different absolute albedo magnitudes. First, CERES albedo is measured in broadband from 300 to 5000 nm, while EPIC albedo is a sampling of 8 narrow band channels from 317.5 nm to 780 nm. The Earth albedo is dominated by the wavelengths corresponding to green (EPIC’s 551nm) and red (EPIC’s 680 nm). Second, CERES uses bi-directional reflectivity functions to stitch together diverse scenes. EPIC measures the total number of photons reaching the detector and assumes a Lambertian reflectivity. This paper does not address the absolute magnitude of the albedo needed for radiative balance calculations, but only the shape of the albedo time series.
-
AC1: 'Reply on RC1', Jay Herman, 18 Jun 2026
-
RC2: 'Comment on egusphere-2026-1631', Anonymous Referee #2, 28 Sep 2026
-
AC2: 'Reply on RC2', Jay Herman, 16 Oct 2026
General Comments: This study examined the trend of Earth’s albedo using DSCOVR EPIC measurements. The trends derived from EPIC are comparable with those inferred from CERES measurements. Both datasets indicate that Earth’s albedo has decreased since 2016, with the rate of decrease becoming larger after 2022. Because the CERES mission has been the sole satellite mission monitoring Earth’s energy imbalance, an additional independent mission such as EPIC provides an important means to validate the existing CERES observations. The study addressed the stability of the radiometric calibration of EPIC. However, the EPIC instrument measures only narrowband radiances from a single viewing direction, at wavelengths shorter than 780 nm. Deriving broadband albedo therefore requires angular and spectral integration, which involve assumptions about atmospheric profiles, surface types, and cloud properties.
The purpose of this paper is NOT to validate the numerical values of CERES albedo measurements, but rather to validate the shape of the trend decrease in albedo since 2016. Most of the contribution to Earth’s albedo comes in the visible wavelength channels. Decreases in EPIC’ visible channel albedos should be similar in shape to CERES changes. EPIC’s eight channel narrow band, 317.5 to 780 nm, backscatter irradiance data are not sufficient to be numerically compared to CERES continuous wavelength albedo 300 – 5000 nm at all angles.
The angular integration refers to integration over the viewing angles. The study does not discuss uncertainties associated with these angular or spectral integration steps. In addition, the impact of changing phase angles is not discussed, nor is the methodology for applying phase-angle corrections described in the albedo time series. The effect of the portion of Earth not observed by EPIC also warrants consideration.
EPIC trend estimates in percent per year are based on data from EPIC measured irradiance expressed in counts per second converted into albedo by multiplication by a single constant for each wavelength channel. It is just a change in units. The albedo time series are corrected for the changing phase angle. The purpose of this paper is to estimate the change in % per year in the 317 to 780 nm wavelength channels. To compare trends with CERES, the EPIC channels were added together after being weighted in proportion to the solar spectrum. For estimating trends, the small cyclic variation, 3-6 month periodicity, caused by the small missing portion of the illuminated Earth has no significant effect since it is included in the deseasonlity analysis. The trends are now estimated by two independent methods, MLR and Lowess, and shown to agree. One of the methods, MLR, Multivariate Linear Regression, also gives error estimates, Fig. 16 and Table 3.
Estimating these uncertainties is crucial for determining whether the detected trends are statistically significant. Given the likely magnitude of these methodological uncertainties relative to the inferred albedo trend, I recommend a major revision before publication in EGUsphere.
Specific Comments: 1. Are there uncertainties in the derived albedo trend associated with the phase-angle shift around 2020? Because the phase angle directly affects the measured bidirectional reflectance, it should influence the estimated Earth albedo. Notably, the timing of the observed change in the albedo trend coincides with this phase-angle transition.
The correction for the observing phase angle, SEV angle, has now been explicitly described in Figs. 4 and 5 along with Table 2.
- This study evaluates calibration reliability based on two components: (i) the stability of EPIC’s narrowband radiometric calibration and (ii) the stability of the lunar albedo under viewing geometries consistent with previous studies. These assessments support the stability of EPIC’s narrowband radiance accuracy, but they do not validate the accuracy of the angularly and spectrally integrated fluxes or albedo. The uncertainties associated with the methods used for angular and spectral integration should be discussed.
As mentioned earlier, this paper is not trying to validate the magnitude of the CERES albedo measurement, but to show that both data sets indicate a reduction in Earth reflectivity. The EPIC measurement relies on the decrease in the global average of observed counts per second (backscattered irradiance) for the illuminated Earth in a 24-hour period. The change mostly comes from decreases in cloud reflectivity as confirmed by the larger decreases in the O2 B- and A-bands.
- This study cites Su et al. (2018) and Penttilä et al. (2022); however, it appears that a different approach to the angular and spectral integration was used here. If so, please clarify the differences and justify the choice of the new method. (See comment above)
- Line 179: “EPIC multiplicative albedo calibration coefficients (Table 1) are used to convert counts per second into top-of-atmosphere Lambertian albedo Aλ = πIλ/Sλ where Iλ = the outgoing radiance at the top of the atmosphere and Sλ is the incoming corresponding solar flux.” The radiance measurements in this study were converted to albedo by assuming a Lambertian surface. However, this assumption is rarely realistic in practical applications. When anisotropy is not accounted for, the resulting Lambertian albedo becomes highly sensitive to the viewing geometry. The uncertainties associated with this approach should therefore be discussed.
The measurement in terms of Lambertian albedo is just a change in units from counts per second. The trends in percent per year are identical. The true anisotropic albedo changes that CERES attempts to model should be proportionate to the measured SEV corrected changes in counts per second from EPIC. EPIC measures the changes from cloud cover and surface reflectivity from sunrise to sunset, which is different than from CERES.
- Line 241-242: “The albedo estimates for these channels, 687.75 ± 0.2 and 768 ± 0.2 nm are much lower (Fig. 7) than for surrounding non-O2 absorbed wavelength channels. “ This result indicates that the spectral integration is highly sensitive to atmospheric and surface conditions, especially cloud conditions for O₂‑absorbing bands. Without characterizing atmospheric and surface properties, it is difficult to derive broadband quantities from narrowband measurements. How was this accounted for in the study? Simply weighting by the solar flux is not sufficient to obtain a physically meaningful broadband albedo.
The paper is not trying to determine the accurate broadband albedo, but just the summed narrow band albedo that behaves in a similar long-term manner to the CERES broadband albedo trend that is dominated by the visible wavelengths.
For example, trends in water vapor, cloud, and surface characteristics produce different spectral signatures in albedo. When TOA albedo changes in the visible and near‑infrared spectral ranges are not aligned, how can this be considered? With the current method, these spectral signatures would not be captured, potentially leading to biases in the inferred broadband albedo. By excluding the O₂‑absorbing band and averaging, the results may partially reflect albedo changes under cloud-free conditions. If the global trend is primarily driven by clear-sky factors, such as surface albedo, this approach might capture the corresponding global changes.
The albedo change is driven by long-term average decreases in cloud amount or cloud height as suggested by the larger decreases in O2 B- and A-band channels 688 and 764 nm, Fig. 16, compared to their adjacent channels 680 and 780 nm.
- Line 143: “corrected for a small missing portion of the Earth’s surface not seen in each EPIC image (3 to 8%) depending on EPIC’s orbital position.” How was the unseen portion of Earth accounted for in calculating the albedo? If the extent of this unseen portion varies, would such variations affect the albedo time series? Please provide the uncertainty ranges in the albedo estimates associated with changes in the missing portions.
The missing small portions of the Earth’s limb produce a small periodic perturbation in EPIC’s measured Counts/second that is corrected in the deseasonlity calculation and does not affect trends in albedo.
The referees’ thoughtful comments have caused us to rewrite most of the paper with some major changes. We now include an additional year’s worth of data for both EPIC and CERES, we show the time series, slopes of the Lowess fits, and the formal MLR trend analysis with error estimates for each EPIC wavelength band as well as the summation of the solar flux and FWHM weighted time series. The trend analysis is now also performed on deseasonalized time series. We have added explicit corrections for the changing observing angle of EPIC from the Earth 12.36O to 2O. Major text changes are marked in green.
-
AC2: 'Reply on RC2', Jay Herman, 16 Oct 2026
-
RC3: 'Comment on egusphere-2026-1631', Anonymous Referee #3, 29 Sep 2026
This manuscript performs a detailed intercomparison between top-of-atmosphere albedo products derived from two different data sources: the EPIC instrument aboard the DSCOVR platform, and the CERES mission. The paper looks at long-term trends in albedo from these two products, finding that there are statistically-significant decreases in albedo in both quantities, and that, in recent years, albedo decreases are accelerating. Additionally, the paper analyzes albedo at different wavelengths and explores seasonal and diel features of albedo, which is enabled by the location (around L1) and viewing capabilities (mostly daylit portion of Earth) of EPIC from the DSCOVR platform.
Overall, there is substantial information about albedo intercomparisons between products from two different missions, as well as the spectral and temporal breakdowns of albedo to make this manuscript compelling and potentially appropriate for publication. It is important for the scientific community to be aware that top-of-atmosphere, planetary-averaged albedo observations show that there are negative, and statistically-significant trends in this quantity. Furthermore, the independent findings from both EPIC and CERES that this quantity has been decreasing provide the Earth Radiation Budget community a high degree of confidence that albedo has been, in fact, decreasing. While previous assessments have intercompared different CERES flight models, the independent validation of CERES with EPIC presents far stronger independent validation that a validation across different CERES flight models. The spectral and temporal features from EPIC are also key for the scientific community that studies albedo.
At the same time, major revisions should be performed to address several issues in the current manuscript.
First, the trend analysis needs to include confidence intervals and the figures need to display such information. Performing trend analysis with end-points at different months, or over just a few years, can be quite problematic unless large confidence intervals are shown that reflect what is and is not significant.Second, the authors need to analyze the extent to which trends in atmospheric quantities in the near-infrared may be impacting the albedo products derived from EPIC in ways that differ from how those quantities impact albedo products derived from CERES. For example, water vapor absorption in the NIR can impact albedo, and increasing water vapor would lead to decreasing albedo, though EPIC measurements over spectral intervals mostly in the visible, are insensitive to these effects. The CERES instrument does measure across the near-infrared water vapor absorption bands, however. The authors need to show that water vapor trends either can or cannot explain the differences between EPIC and CERES albedo recovery since 2024. The authors also need to discuss what else in the near-infrared that are missing from EPIC measurements that could impact their EPIC time-series.
Third, the authors need to discuss the shared product development infrastructure between EPIC and CERES reflected shortwave radiation products. The authors cite various papers that discuss how CERES data and product processing pipelines are used to derive EPIC reflected shortwave radiation products, so the readers need to be aware of just how independent the EPIC and CERES products are from each otherFinally, the authors need to either discuss about the implications of these albedo trends for the Earth’s radiation budget and energy imbalance, which they indicate in their abstract as important, or remove that language from the manuscript.
On a minor note, the paper’s plain language summary needs to describe what CERES actually is to a reader who has not heard of this mission.
Citation: https://doi.org/10.5194/egusphere-2026-1631-RC3 -
AC3: 'Reply on RC3', Jay Herman, 16 Oct 2026
This manuscript performs a detailed intercomparison between top-of-atmosphere albedo products derived from two different data sources: the EPIC instrument aboard the DSCOVR platform, and the CERES mission. The paper looks at long-term trends in albedo from these two products, finding that there are statistically-significant decreases in albedo in both quantities, and that, in recent years, albedo decreases are accelerating. Additionally, the paper analyzes albedo at different wavelengths and explores seasonal and diel features of albedo, which is enabled by the location (around L1) and viewing capabilities (mostly daylit portion of Earth) of EPIC from the DSCOVR platform.
Overall, there is substantial information about albedo intercomparisons between products from two different missions, as well as the spectral and temporal breakdowns of albedo to make this manuscript compelling and potentially appropriate for publication. It is important for the scientific community to be aware that top-of-atmosphere, planetary-averaged albedo observations show that there are negative, and statistically-significant trends in this quantity. Furthermore, the independent findings from both EPIC and CERES that this quantity has been decreasing provide the Earth Radiation Budget community a high degree of confidence that albedo has been, in fact, decreasing. While previous assessments have intercompared different CERES flight models, the independent validation of CERES with EPIC presents far stronger independent validation that a validation across different CERES flight models. The spectral and temporal features from EPIC are also key for the scientific community that studies albedo.
At the same time, major revisions should be performed to address several issues in the current manuscript.
First, the trend analysis needs to include confidence intervals, and the figures need to display such information. Performing trend analysis with end-points at different months, or over just a few years, can be quite problematic unless large confidence intervals are shown that reflect what is and is not significant.An additional method of trend analysis has now been included, MLR (Multivariate Linear Regression) that deseasonalized the counts per second data and derives the linear trend and error estimates (see Fig.16). The larger decreases in and O2 B- and A-band channels 688 and 764 nm, Fig. 16, compared to their adjacent channels 680 and 780 nm suggests decreasing cloud cover or long-term decreases in average cloud height.
Second, the authors need to analyze the extent to which trends in atmospheric quantities in the near-infrared may be impacting the albedo products derived from EPIC in ways that differ from how those quantities impact albedo products derived from CERES. For example, water vapor absorption in the NIR can impact albedo, and increasing water vapor would lead to decreasing albedo, though EPIC measurements over spectral intervals mostly in the visible, are insensitive to these effects. The CERES instrument does measure across the near-infrared water vapor absorption bands, however. The authors need to show that water vapor trends either can or cannot explain the differences between EPIC and CERES albedo recovery since 2024. The authors also need to discuss what else in the near-infrared that are missing from EPIC measurements that could impact their EPIC time-series.
The broadband albedo measurement of CERES 300 to 5000 nm are not directly comparable in magnitude to EPIC narrow-band measurements 317.5 to 780 nm. Our trend analysis for the UV, visible, and near-IR channels is based on the observed reflected irradiance in counts per second converted by a constant multiplier into Lambertian albedo, just a change in units. Trends derived from SEV corrected counts per second and Lambertian albedo are identical. Since the reflected irradiance from the Earth is dominated by the visible channels, both CERES and EPIC should see a decrease in albedo, which they do. Because of the difference in measurements as detailed by the referee, the magnitude of the derived albedo will be different and the long-term changes will be different. But both measurements show an accelerating decrease in Earth reflectivity in the 2021 to 2026 period.
Third, the authors need to discuss the shared product development infrastructure between EPIC and CERES reflected shortwave radiation products. The authors cite various papers that discuss how CERES data and product processing pipelines are used to derive EPIC reflected shortwave radiation products, so the readers need to be aware of just how independent the EPIC and CERES products are from each other.A summary of the complicated multi-instrument CERES data is given on Page 4. This compares with the very simple direct measurements of the daily global average of EPIC counts per second of the illuminated Earth disk. The small missing portion of the Earth disk adds a cyclic component to the time series that is removed in the deseasonalized trend analysis. Two methods of trend analysis, Lowess and MLR, are compared and shown to be identical. The MLR method gives error estimates.
Finally, the authors need to either discuss about the implications of these albedo trends for the Earth’s radiation budget and energy imbalance, which they indicate in their abstract as important, or remove that language from the manuscript.
We have changed the abstract sentence related to energy balance by saying, “Both EPIC and CERES albedo data suggest that the accelerating reduced Earth reflection of sunlight back to space, mostly from clouds, is enhancing the amount of solar energy reaching the Earth’s surface.” Although not part of this paper, the CERES long wavelength IR data shows an increase in outgoing heat from the Earth that may approximately balance the increased solar energy input.
On a minor note, the paper’s plain language summary needs to describe what CERES actually is to a reader who has not heard of this mission.
Added: CERES data are obtained from multiple satellites over its 26-year observation period and use models of the Earth’s surface reflectivity. EPIC obtains a daily average of detector counts per second of the Sun illuminated Earth disk.
Thank you for your thoughtful comments. In summary, we now include an additional year’s worth of data for both EPIC and CERES, we show the deseasonalized time series, slopes of the Lowess fits, and the formal trend analysis with error estimates for each EPIC wavelength band as well as the summation of the solar flux and FWHM weighted time series. The trend analysis is now performed on deseasonalized time series that have been corrected for small changes in filter transmission. We have added explicit corrections for the changing observing angle of EPIC from the Earth 12.36O to 2O. Major text changes are marked in green.
-
AC3: 'Reply on RC3', Jay Herman, 16 Oct 2026
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 784 | 426 | 85 | 1,295 | 43 | 48 |
- HTML: 784
- PDF: 426
- XML: 85
- Total: 1,295
- BibTeX: 43
- EndNote: 48
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
98. Dark side of the moon typically refers to the unseen portion of the moon, not the part that is lit by Earthshine.
Fig 7. Would suggest to keep the you axis scale the same throughout the panels to facilitate easier inter comparison between the brightness at the different times of year.
Fig 08. Maybe relevant to confirm these are from the scaled solar flux due to Sun-Earth distance. Same in lines 245 and on.
302. Could you elaborate a bit on why the albedo between CERES and EPIC are different? Is this mostly due to different viewing orientations or lower reflectivity in the near-to-mid infrared?