Earth’s albedo time series reveals low radiative energy input in December 2020


 The Earth’s spherical albedo describes the ratio of light reflected from the Earth to that incident from the Sun, an important input variable for the Earth’s radiation balance. The spherical albedo has been previously estimated from satellites in low-Earth orbits, and from light reflected from the Moon. However, neither of these methods can produce continuous time series of the entire planet. We developed a global method to derive the Earth’s spherical albedo using the images from the Earth Polychromatic Imaging Camera (EPIC) on board NOAA’s Deep Space Climate Observatory (DSCOVR). The satellite is located in the Lagrange 1 point between the Earth and the Sun and observes the complete illuminated part of the Earth at once. The method allows us to provide continuously updated spherical albedo time series data starting from 2015. This time series shows a systematic seasonal variation with the mean annual albedo estimated as 0.295±0.008 and an exceptional albedo maximum in 2020, attributed to unusually abundant cloudiness over the Southern Oceans.


Introduction 26
Solar radiation is the primary energy source of the Earth and largely determines Earth's climate. 27 The proportion of the incoming solar radiation reflected back to space by the Earth is described by 28 the spherical albedo. It depends on the reflective properties of the Earth and thus it is affected by the 29 proportion of the highly reflective areas relative to darker areas. For example, the melting of the 30 Antarctic and Greenland ice sheets results in increased absorption and decreased albedo. Hence, the 31 Earth's spherical albedo is a major factor behind the global weather and climate processes. 32 However, up to date there has not been a global, day-to-day estimate of the spherical albedo. 33 In the first half of the 20th century, estimates of the spherical albedo were based on an indirect 34 method of observing the Earth-lit Moon 1 . These so-called earthshine observations depend on correct 35 estimations of the Moon's light-scattering properties, and thus they can be inaccurate and vary 36 considerably. The earliest satellite measurement of the spherical albedo was made in 1959 by the 37 Explorer 7 satellite and its value has remained approximately 0.3 ever since 2 . Since 1997 the albedo 38 is being overseen by the Clouds and the Earth's Radiant Energy System (CERES), which includes 39 five satellites and seven CERES instruments 3 . As of 2017, only five instruments are operational. As 40 it takes numerous hours for the CERES to scan the entire Earth while the cloud cover of the Earth 41 evolves in a matter of minutes, the spherical albedo evaluation method by the CERES instruments 42 produces noticeable uncertainties in the measured albedo value 2,3 . 43 To measure the spherical albedo directly one needs to simultaneously detect radiation reflected by 44 the Earth from all directions, which renders such measurements impossible. To circumvent this, we 45 use the Earth Polychromatic Imaging Camera (EPIC) on board the Deep Space Climate 46 Observatory (DSCOVR) spacecraft combined with angular distribution models provided by the 47 CERES based on many years of dedicated measurements. The DSCOVR is a spacecraft orbiting in 48 the Lagrange point 1 around 1.5 million kilometers from the Earth, which allows the EPIC to 49 always view practically the entire sunlit hemisphere of the Earth. The DSCOVR was launched in 50 2015, and the EPIC has been operational ever since apart from one six-month maintenance break in 51 2019 4 . 52 Data from the instruments allows us to propose an algorithm that automatically translates 53 directional reflectance obtained from the EPIC images into estimated value of short-wave spherical 54 albedo. We have launched a web service that collects the computed spherical albedo of the Earth 55 from the whole operational period and updates the data daily with the latest observations 5 . This is 56 the first global daily time series of the Earth's albedo and it spans already now over seven years in 57 time. This enables us to analyze the pattern of temporal variation in albedo over a year which 58 demonstrates anomalies in albedo behavior. The accurate estimate of the short-wave spherical 59 albedo is important in evaluating the energy balance of the Earth system. 60

61
The EPIC images constitute a time series of the sunlit part of the Earth, including atmosphere, 62 starting from June 2015 and still running. There are usually about 22 images per day during 63 Northern Hemisphere summer, and 13 during winter. Each multispectral image has 10 wavelength 64 channels between 317 and 780 nm. The channels have full-width-at-half-maximum values between 65 1 and 3 nm 6-9 . 66 Each image pixel represents the radiance reflected by the corresponding surface area on the Earth. 67 This radiance is a function of the reflective properties of the surface, and the solar and satellite 68 angles. The DSCOVR satellite is located in the first Lagrange point between the Earth and the Sun, 69 therefore the radiance is observed close to backscattering geometry with satellite and solar angles 70 being almost equal. To convert measured radiances at backscattering into integrated albedo values 71 at top-of-atmosphere (ToA), we apply the angular distribution models (ADMs) provided by the 72 CERES project 10,11 . The ADMs are provided for several surface types and cloud fractions, and we 73 combine these into four types for our analysis: clear and cloudy land and ocean. The reason for this 74 is that the angular resolution of the tabulated ADMs is only 10 degrees, and we want to interpolate 75 with a finer resolution in the backscattering direction and combining similar surface types gives us 76 more robust estimation (see Methods). Additionally, estimating temporary cloud fractions from 77 EPIC images is challenging, a more robust method can be developed for simple clear versus cloud-78 covered surface estimation. Land or ocean surface classification is available from the International 79 Geosphere-Biosphere Programme, and temporal cloud coverage we estimate from the EPIC images 80 using logistic regression model with input from EPIC channels at 325, 551, and 780 nm. 81 The ToA albedo values for each pixel over the Earth's sunlit disk are averaged for albedo of the 82 Earth at each wavelength channel. Finally, the per-wavelength albedos are summed with weights 83 from incident solar spectra at each channel, obtained from National Oceanic and Atmospheric 84 Administration (NOAA) climate data record of solar spectral irradiance 12 . The solar spectrum is 85 taken as constant in this study with only the distance between the Earth and the Sun influencing the 86 total flux level. The channels at 688 and 764 nm in EPIC images are left out since they are targeted 87 to measure atmospheric O2 at narrow absorption bands. The resulting time series of Earth's daily 88 albedo is shown in Figure 1.

91
The daily averaged time series has been filtered to remove outliers. For some EPIC images, not all 92 the wavelength channels are present. If there is not a proper coverage of usable images over the day, 93 some parts of the Earth are not present, and this would introduce a bias to the mean albedo of that 94 day. Therefore, these days are left out from our time series data. There is a period between late 2019 95 and early 2020 when the EPIC camera was not operational. 96 We can derive the mean yearly spherical albedo of the Earth, at visual wavelengths, by grouping the 97 daily time series values by the day in the year, averaging per day, and finally averaging over days in 98 a year. From the currently available data, this value is 29.5 ± 0.8 %. The value agrees well with 99 earlier estimates of 28.6-30.1 % by satellites in low-Earth orbits, see Kandel

110
We can verify the anomaly in December 2020 statistically by studying the monthly averages for 111 each year. The combined monthly averages are shown in Figure 3, plotted as variations from the 112 overall average albedo in a box-and-whiskers chart. The one-way analysis-of-variance test of the 113 differences in the mean values over several sets of observations (different years over a month) can 114 find significant (p-value less than 0.1 %) differences for all months. However, by far the smallest p-115 value, and therefore the largest difference, is for December, formally less than 10 -29 . 116 When conducting t-tests for year pairs inside a month, we find that the only year which differs from 117 all the other years when using the Bonferroni-corrected p-value limit of 0.001/mi, where mi is the 118 number of year data sets on month i, is the year 2020 and months August and December. From 119 these analyses we conclude that the largest yearly variations in a month are on December, and that 120 is mainly due to year 2020. 121 The Earth contains regions of low albedo (e.g., cloud-free ocean areas, vegetated land areas) and 142 high albedo (e.g., clouds, ice and snow surfaces, sandy deserts). Diurnal and annual variations in the 143 albedo time series (Figure 1, Figure 3) result from a modulation of these two aspects, depending on 144 the apparent longitude (diurnal variations) and latitude (annual variations). Especially, the cloud-145 covered ocean areas increase the albedo. The daily cloud-covered ocean fraction, estimated from 146 the EPIC images, has the correlation coefficient value of 86 % with the daily albedo value (see 147 Figure 6). 148 In annual albedo variations, the main albedo maximum occurs in December when the Antarctic ice 149 sheet, sea ice, and snow cover are visible entirely. At this time, cloud formations of the mid-latitude 150 cyclones over the Southern Ocean are well pronounced. Albedo is further enhanced by the shallow 151 convective cloud cover over the subtropical oceans and the relatively small areas of cloud-free 152 ocean areas visible, on average, at this time of the year. 153 The secondary albedo maximum occurs in June when the Greenland ice sheet and sea ice in the 154 Arctic Ocean are well exposed. Cloud formations of the mid-latitude cyclones of the North Atlantic 155 and Pacific storm tracks are active, although reduced from their winter maxima. Subtropical desert 156 areas are mostly cloud free exposing also these high surface albedo areas. Deep convective clouds 157 of the inter-tropical convergence zone are on the Northern Hemisphere at this time of t he year. 158 Again, relatively little cloud-free ocean areas are visible. The solar energy input at the Earth's average distance from the Sun is 1361 W m 2 ⁄ for the sunlit 166 disk of the Earth over all wavelengths, and the EPIC filters range from 317 nm to 780 nm in 167 wavelength 12 . The portion of the solar input between these wavelengths is 52.7 %. In ultraviolet 168 below 317 nm there is only 0.5 % of the total energy. We assume that the albedo derived here for 169 the range 317-780 nm is valid for somewhat longer wavelengths and estimate roughly that our 170 shortwave albedo is valid for at least 60 % of the reflection or absorption of the total solar input.

171
With this estimate, the input energy that is absorbed into the area of sunlit Earth's disk and 172 atmosphere is about 4.9 W m 2 ⁄ more during the albedo minimum in September than on average. 173 During the albedo maximum in December, about 8.7 W m 2 ⁄ less energy is absorbed.  A logistic regression model was separately fitted into data with clear land and cloud-covered  212 surfaces, and data with clear ocean and cloud-covered surfaces. A logistic regression model has a 213 linear function modeling the log-odds of the probability of an event. In our case, the event is that the 214 surface is covered with clouds. With some algebra the model can be written as 215 where is the probability of cloud-cover, and 0 + 1 1 + ⋯ + is the linear function with 217 unknown coefficients and known radiances from channels 1, … , . 218 We executed forward-selection stepwise regression to optimize the model with the best value of the 219 Bayesian Information Criterion (BIC) statistics of the model. We ended up with models with a 220 constant coefficient 0 and three coefficients for the EPIC wavelength channels at 325, 551, and 221 780 nm. For land vs. clouds model, all the four coefficients were tested significant with p-values 222 less than 0.1 %. For ocean vs. clouds model, all other but the coefficient for the channel 780 nm 223 were significant with the same p-value limit. 224 Scattering geometry and interpolation of angular distribution models 225 We found it realistic to successfully classify pixels of clear or cloudy land or ocean surface from the 226 EPIC images, but not different cloud types (water or ice) or levels of cloud cover that are present in 227 CERES angular distribution models (ADMs). ADMs are also divided into different wind speeds for 228 ocean and into different vegetation types for land. To exploit these, wind speeds on ocean pixels 229 would need to be connected from weather models into our analysis together with up-to-date land 230 cover information. This is a bit cumbersome but could be done. However, we decided to emphasize 231 more the interpolation of the scattering geometry inside the ADM models over having many 232 ocean/surface/cloud subtype ADMs. 233 The scattering geometry in the EPIC observations is such that the observation is always done close 234 to backscattering. The phase angle between the Sun and the DSCOVR spacecraft, as seen from the 235 Earth, is below 12° except for few rare cases in years 2020-21. On average, phase angle has been 236 8.2°. After the break in the DSCOVR operations in 2019, the minimum phase angle has gradually 237 started to decrease (see Figure 6). It was never below 4° before 2020, but now the minimum value 238 is 1.8°. 239 While in terms of phase angle EPIC is observing close to backscattering, the local Sun zenith angle 240 on each Earth pixel varies between 0° and 90°. Effect of phase angle on albedo 255 The phase angle of EPIC observations varies between 2° and 12°. After DSCOVR operations break 256 in 2019 the phase angle variation has increases and the minimum phase angle has decreased from 257 previous 4° to 1.8°. These decreasing phase angles introduce possible source of bias in our albedo 258 estimation. The resolution in the binned data of the ADM models is 10° in all three angles, 259 indicating that we do not have exact information on how possible backscattering effects (self-260 shadowing, coherent backscattering) are behaving on small phase angles inside this bin. 261 Currently, we can only estimate the possible bias resulting from small phase angles. We

281
The exceptional albedo values in December 2020 were at most 0.023 larger than the average 282 December value. This exceptional value was received at the time of local phase angle minima of the 283 DSCOVR spacecraft, 2.1°. We conclude that the effect of the phase angle can explain, at maximum, 284 about half of the difference, and that December 2020 is exceptional in our albedo time series even if 285 taking the possible phase angle effect into account. 286