1.1 EPIC Instrument

The EPIC instrument consists of a 30 cm aperture, 283.642 cm focal length Cassegrain telescope containing a multi-element field lens group focusing light onto a UV-sensitive 2048 × 2048 hafnium-coated charge-coupled device (CCD) detector with 12 bit readout electronics. Images are made through 10 narrowband filters, four in the ultraviolet, four in the visible, and two in the near infrared. The 10 filter transmission functions are shown in Fig. 1. Observations are made as light passes sequentially through each of 10 narrowband filters mounted in two moveable filter wheels and through an exposure control three-slot rotating shutter. The exposure times for each wavelength were adjusted during the first week of in-flight operation to achieve an approximately 80 % CCD electron well fill in the brightest scenes to avoid saturation and leaking from one pixel to another (blooming). Earth exposure times range from about 654 ms at 317.5 nm to 22 ms at 551 nm, which have not changed during the current life of the mission. Another set of exposure times was determined for viewing the full moon as seen from the earth (Table 1). The CCD has a well depth of approximately 8.5 × 10⁴ electrons (a maximum signal-to-noise ratio (SNR) of 290:1) before a small dark current correction that is a function of its in-flight operating temperature of −20^∘C. The 12 bit readout means that there are 2¹¹ (2048) readout steps or counts (42 electrons/count). The counts divided by the exposure time (counts per second) are converted to radiances or albedos using in-flight scene matching calibration from low earth orbit satellites (see Sect. 1.2 and Table 2). The maximum SNR applies to the brightest of scenes over high clouds or fresh snow over ice. Cloud-free and snow-free scenes have much lower SNR, which affects the visible channels more than the UV channels because of the lower scene contrasts with clouds caused by enhanced UV Rayleigh scattering. There are occasional bright flashes caused by ice crystals in high clouds that saturate a few pixels (see Fig. 2 and Marshak et al., 2017) in the equatorial and midlatitude regions.

Table 1Exposure Times for viewing the earth and full moon (earth side view)

Download Print Version | Download XLSX

Table 2πK_M on 1 January 2016.

Download Print Version | Download XLSX

Figure 2Normalized calibration functions referenced to its value at 4 January 2016 when D_E=1 AU. Average rate of increase is 0.016 per year.

Download

The filters of interest for calculating ozone amounts, aerosol index (AI), and cloud reflectivity have passbands centered on the vacuum wavelengths 317.5, 325, 340, and 388 nm with full widths at half maximum (FWHM) 1.0, 1.0, 2.7, and 2.6 nm, respectively. For the UV channels, 2 × 2 individual pixels are averaged on board the spacecraft to yield an effective 1024 × 1024 pixel image corresponding to an 18 × 18 km² resolution at the observed center of the earth's sunlit disk. The effective spatial resolution decreases as the secant of the angle between EPIC's sub-earth point and the normal to the earth's surface. Only the 443 nm channel is retrieved at full resolution to help with resolving cloud cover and obtaining improved color images. The sampling resolution of a single pixel is about 8 × 8 km² (about 1 arcsec), but including the effect of the optical point-spread function, the effective 443 nm channel resolution is about 10 km. The effective resolution at 443 nm has been verified by looking at clear scenes over the Nile River in Egypt and, occasionally, the cloud-free Amazon River in Brazil.

EPIC data have been obtained since 15 June 2015 at a rate of one set of 10 wavelengths every 68 min during Northern Hemisphere (NH) summer and one set every 110 min in the winter. The difference between summer and winter rates is caused by the reduced number of hours in the winter when the single antenna (located at Wallops Island, Virginia) is in view of the spacecraft and by limitations in the spacecraft memory technology from the late 1990s, when the spacecraft was designed and constructed. The EPIC instrument was refurbished in 2011 with improved filters to reduce out-of-band light, new filters for the O₂ A- and B-bands, and a new field lens group to reduce stray light.

Each wavelength of the 10-wavelength set is obtained at slightly different times. The first filter in the sequence is 443 nm, which takes about 2 min to complete a measurement (28 ms exposure time (Table 1) plus CCD readout and onboard processing time that includes 12 bit JPEG compression of a 2048 × 2048 pixel image). The remaining nine filter measurements take a total of about 5 min (exposure times plus CCD readout into memory) and then another 13 min to process the data for the nine filters (this includes 12 bit JPEG compression of 1024 × 1024 images that have been averaged on board in groups of 2 × 2 pixels before compression). Adjacent pairs of wavelengths are measured at 30 s intervals before the onboard processing is started. This means the individual channel images are not co-located at the pixel level because of earth rotation (15.03^∘ h⁻¹ or about 1670 km h⁻¹ at the Equator), the slow rotation of the spacecraft (0.082^∘ h⁻¹), and a small amount of spacecraft jitter. Each pixel views about 1 arcsec or 2.78 × 10⁻⁴ degrees. Data from an onboard star tracker and feedback from the earth's image on the CCD keep the images approximately centered on the CCD. The lack of native channel-to-channel co-location requires an elaborate spherical geometry geolocation analysis to adjust the data to a common latitude × longitude grid with an accuracy of one-fourths of a pixel.

This paper presents examples of the ozone and scene reflectivity retrievals that are used to obtain unique estimates of erythemal UV irradiance (or UV index, UVI) as a function of latitude, longitude, local solar time, and altitude above sea level. Since this is the first paper on EPIC-retrieved ozone, Sect. 1 contains a brief description of the calibration of the four UV channels and the ozone retrieval algorithm. Section 2 shows examples of natural-color images, Sect. 3 gives an example of retrieved ozone and the corresponding 388 nm Lambert equivalent reflectivity (LER; Herman et al., 2009), Sect. 4 presents a validation of EPIC-retrieved ozone compared to ozone from ground-based and satellite data, Sect. 5 shows details of the latitudinal and longitudinal synoptic variability of ozone, and Sect. 6 presents new results showing the sunrise-to-sunset variability of UV erythemal radiation reaching the earth's surface including the reduction by clouds.

1.2 Calibration

Before the raw EPIC data (counts per second) can be used, a number of pre-processing steps must be accomplished. The major steps are (1) measuring and subtracting the dark current signal, (2) “flat-fielding” the CCD so that the sensitivity differences between all 4 million pixels are determined and corrected, (3) correcting for stray-light effects to account for light that should be going to a particular pixel but instead is scattered to different pixels, and (4) determining the radiometric calibration for each wavelength channel in terms of EPIC counts per second to be converted to earth normalized radiances or reflectances (backscattered at approximately 172^∘). The earth upwelling radiance I_M ($\mathrm{W}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{nm}}^{-\mathrm{1}}{\mathrm{sr}}^{-\mathrm{1}}$) at the top of the atmosphere (TOA) is defined in terms of the albedo A_M given by Eq. (1):

For wavelength bands M=1 to 4, S_M is the incident solar irradiance ($\mathrm{W}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{nm}}^{-\mathrm{1}}$) weighted with the filter function for band M at 1 AU and D_E is the sun–earth distance in AU (astronomical units). Since EPIC does not measure solar irradiance, we use a high-resolution solar irradiance spectrum, S(λ) (Dobber et al., 2008), as a reference solar spectrum. The reference spectrum is convolved with EPIC's filter transmission functions T_M(λ) (Fig. 1) to obtain each EPIC channel's weighted solar irradiance S_M at sun–earth distance at 1 astronomical unit (Eqs. 1 and 2).

In-flight radiometric calibration is accomplished by comparison with albedo values measured by current well-calibrated LEO (low earth orbiting; e.g., Aura/OMI and Suomi-NPP/OMPS, National Polar-orbiting Partnership/Ozone Mapping and Profiler Suite) satellite instruments observing scenes that match in time and observing angles with those from EPIC. For albedo measurements, OMPS has a calibration accuracy of 2 %, while its wavelength dependence (precision) in the calibration is estimated to be better than 1 % (Jaross et al., 2014). The OMPS Nadir Mapper on Suomi-NPP has a 50 × 50 km² footprint in its normal operating mode with 36 cross-track views (±55^∘ satellite view angle or strip of about ±12^∘ equatorial longitude). It has a spectral resolution of 1 nm, which is close to EPIC's 317.5 and 325 nm channels' FWHM but narrower than EPIC's 340 and 388 nm channels. To perform in-flight calibration, OMPS' albedo spectra were either interpolated (for 317.5 and 325 nm channels) or convolved (at 340 and 388 nm) with each EPIC filter transmission function T_M (Fig. 1). Because the albedo spectra A_M(λ) (Eq. 1) removes the Fraunhofer line structure contained in both the solar irradiance S_M and the reflected earth radiance I_M(λ), the interpolation and convolution of A_M(λ) has better accuracy than directly using I_M(λ). OMI on Aura has 13 × 24 km² spatial resolution and about ±56^∘ cross-track views (a strip of ±1300 km or ±13^∘ equatorial longitude) with a spectral resolution of 0.42 nm. To match measurements with DSCOVR, OMI's albedo spectra were convolved with EPIC's T_M(λ). Then, the results in every two adjacent cross-track views and four consecutive along-track scans were combined to form 50 × 50 km² footprints for comparison with EPIC-measured counts per second obtained from 7 × 7 EPIC pixels.

EPIC raw counts per second inside each coincident footprint are preprocessed by the steps stated in a previous paragraph. Then, the counts-per-second average and variance in each coincident footprint are computed to obtain the EPIC albedo calibration coefficients K_M (Eq. 3). Misalignment between EPIC and OMPS or OMI footprints can result in large scene noise unless uniform scenes are selected and less uniform scenes discarded. This is achieved by weighting each coincident data point with the reciprocal of the percent EPIC counts-per-second variance inside the coincident footprint. All of the coincident points between LEO satellites and EPIC observations occur within ±40^∘ of the earth's Equator. Selected LEO footprints have viewing angles nearly identical to EPIC's (within 1^∘ in backscatter angle and 2^∘ degrees in solar zenith angle). EPIC's backscatter angle varies with latitude and longitude by less than 0.25^∘, since the angular size of the earth cycles from 0.45^∘ to 0.53^∘ to 0.45^∘ every 6 months depending on the location of DSCOVR in its orbit (an irregular Lissajous orbit about L₁ that is tilted relative to the ecliptic plane and perturbed by the earth's moon). These small differences in observing geometry are corrected in the atmospheric radiative transfer model calculations α(λ) (Eq. 4), resulting in corrections less than 2 %. EPIC albedo calibration coefficients are derived from Eqs. (3) and (4):

where M is the EPIC channel number, M = 1,2,3,4, A_M(OMPS) = OMPS albedo measurement in the EPIC channel-M wavelength band, α_M(EPIC) and α_M(OMPS) are computed albedo values for EPIC and OMPS coincident geometry, C_M(EPIC) is the average count rate over the pixels matching OMPS; D_E is the sun–earth distance in AU, α(λ) is the computed high-resolution normalized radiance spectrum, S(λ) is the referenced high-resolution solar irradiance spectrum, and T_M(λ) is the EPIC filter transmission profile or the OMPS slit function.

All of the coincidence points with LEO satellite instruments were measured using the area of the EPIC CCD within 600 pixels of its center. There are about 15 000 coincidence data points accumulated by the end of 2016. Because of the large number of data points, statistical averaging errors are small. An atmospheric radiative transfer model, RTM, takes total column ozone (TCO) and surface reflectivity from LEO retrievals to obtain both α_M(EPIC) and α_M(LEO). Although uncertainties in the RTM can propagate into the computed albedos, the resulting uncertainties in α_M(EPIC) and α_M(LEO) are approximately identical, and they approximately cancel in Eq. (3). The resulting EPIC albedo calibration uncertainty is mostly inherited from the OMPS albedo calibration uncertainty, which has an accuracy of 2 % and a precision of 1 % in relative (wavelength-dependent) values. For the UV channels, the calibration factors K_M are not constants but slowly increasing functions of time (on average 0.016 per year; see K_M(t) in Fig. 2), which is normalized to one on 1 January 2016. Table 2 shows the reference values of K_M multiplied by π.

Using Tables 1 and 2 and Fig. 2, EPIC albedo measurements are derived with

Note that the factor D${}_{\mathrm{E}}^{\mathrm{2}}$ for solar irradiance at 1 AU is contained in the albedo calibration coefficient K_M. Since solar activity changes (e.g., 27.5 day cycle) are negligible for EPIC UV channel wavelengths, daily solar irradiance changes are only adjusted with the sun–earth distance D_E. Users of EPIC data may also be interested in radiance measurements. The radiance calibration coefficients can be derived with Eq. (6),

and the radiance measurements can be obtained with Eq. (7):

The uncertainty in the radiance calibration can increase significantly due to errors in estimating the absolute solar irradiance. Uncertainty in estimated S_M for EPIC UV channels in Table 1 is about 3 %.

1.3 Ozone algorithm

Once the albedo calibration factors are applied to EPIC's measured counts per second, the calculated albedos can be combined to retrieve TCO, LER, and AI. The TOA directional albedo calculation uses the TOMRAD radiative transfer calculation code, which has a spherical geometry correction for large solar zenith angles (SZAs) and satellite looking angles (SLAs) (Caudill et al., 1997). The calculation uses the same climatological ozone profiles used in OMI retrievals, altitude-weighted average effective ozone temperatures, ground reflectivities, terrain height, and climatological cloud heights. Spectrally resolved O₃ absorption cross sections are from Brion et al. (1993, 1998); Daumont et al. (1992); and Malicet et al. (1995). The resulting spectra are convolved with the EPIC filter transmission functions (Fig. 1) and with the reference solar irradiance spectra (see Eq. 4).

The resulting computed α_M (Eq. 4) are compiled into a finely stepped look-up table as functions of ozone profiles and solar-view angles. EPIC ozone retrieval uses the 388 nm channel for computing the surface reflectivity with a formula similar (except for choice of wavelengths) to that used in cloud reflectivity studies (Herman et al., 2009). Then, the retrieval is based on two ozone absorption channels, 317.5 and 340 nm for low-optical-depth conditions or 325 and 340 nm for high-optical-depth conditions, together with the 388 nm measurement to form triplet equations. The ozone retrieval algorithm assumes a linear wavelength dependence in the surface reflectivity (Eq. 8),

where λ₀ is given wavelength 388 nm. The total column ozone (TCO) is given by Eq. (9),

where Ω₀ is an initial climatology estimate of TCO or the TCO from the previous iteration step; λ₁ and λ₂ are the selected ozone absorption wavelengths; N_λ is the N value defined as the logarithm of the albedo values by Eq. (10),

and ΔN_λ is the N-value residue (difference between the measured N value and the computed N value; $\frac{\partial N_{{\mathit{\lambda }}_x}}{\partial Z}=\text{measurement sensitivity}$ with respect to the total column ozone, Z=Ω, or the surface reflectivity, Z=R) for wavelengths λ₁ or λ₂.

Figure 3Natural-color EPIC earth images from 6 June and 6 December 2016 showing the field of view during the respective hemispheric summers. In both of these images, 6 months apart, the EPIC orbit is to the west of the earth–sun line, causing the west side of the globe (sunrise) to appear brighter than the east side (sunset). Notice the bright specular reflection over Argentina, South America, embedded within a cloud feature. This is thought to be from ice crystals in high clouds (Marshak et al., 2017).

Download

If one assumes the sensitivities to the surface reflectivity, $\partial N_{\mathit{\lambda }}/\partial R$, are wavelength independent, Eq. (5) for the triplet algorithm is similar to the Version 8 TOMS algorithm (Rodriguez et al., 2003).

Since the algorithm for ozone (Eqs. 8 to 10) requires the use of two or more wavelength channels, the measured counts per second for each channel must be geolocated on a common latitude × longitude grid that is accurate to 0.25 of a single pixel size. When projected on the 3-D earth, the sampling size is about 8 km at nadir and effectively increases to 10 km when EPIC's point spread function is applied. The result for 2 × 2 pixel averaging is a spatial resolution at nadir of about 18 km, which gets larger as the secant of the SLA from the nadir point. SLA is measured relative to the normal to the earth's surface and is 0^∘ at nadir and almost 90^∘ at the earth's sunlit terminator. The radiative transfer spherical geometry correction is accurate until about 80^∘ in SZA and SLA, which means that retrieved ozone values near the earth's terminator are not accurate.