The core quantity at stake is aerosol optical depth, or AOD, a measure of how much sunlight or moonlight is attenuated by particles suspended in the atmosphere. AOD is a fundamental climate and air quality variable, and AERONET’s fleet of Cimel sun-sky radiometers has long been the gold standard for measuring it during daylight hours. But the atmosphere does not stop changing at sunset. Nighttime AOD can differ significantly from daytime values because the boundary layer shallows, relative humidity rises and drives hygroscopic particle growth, emission patterns shift, and chemical evolution proceeds without photochemistry. Capturing this 24-hour cycle matters because aerosols that cool the surface by day can actually warm it at night, and nocturnal radiative effects can even promote radiation fog that further processes aerosols.

Measuring AOD by moonlight, however, is a fundamentally harder problem than measuring it by sunlight. The Sun is an extraordinarily stable calibration source whose irradiance varies by only about 3.4 percent over a year, an effect that is almost entirely geometric and can be corrected with precision. The Moon is another story entirely. Lunar brightness at full moon is ten times or more that at quarter moon, and near full moon the irradiance can change by a percentage comparable to the annual solar variation in just two to three hours. This is partly due to the opposition effect, in which the porous lunar regolith produces constructive backscatter, and shadow hiding, a geometric effect that boosts lunar brightness by roughly 40 percent during the full moon window. Earth-moon distance adds another 12 percent swing within two weeks from apogee to perigee.

The network’s solution builds on the ROLO model, the Robotic Lunar Observatory model developed by the United States Geological Survey to predict lunar spectral irradiance. ROLO alone was never accurate enough for absolute calibration, and early attempts that relied on it produced unphysical results, including large negative AOD values during clean atmospheric conditions. The AERONET team instead developed an empirical correction. At the high-altitude calibration facilities at Mauna Loa in Hawaii and Izaña in the Canary Islands, reference instruments undergo both solar and lunar Langley calibrations, and the ratio of the resulting calibration coefficients is compared against the nominal sun-to-moon gain factor of 4096. Deviations from that ideal ratio, binned as a function of lunar phase angle, yield a correction factor for the ROLO model at each of six wavelengths from 440 to 1640 nanometers.

The updated assessment relies on a vastly larger set of calibration pairs than the original provisional lunar product released in 2016, with automated Langley pair counts ranging from 634 to 1537 depending on the channel. Each pair was retained only if the associated AOD fell below strict thresholds and the correlation coefficient of the fit was high enough. The corrections were then fitted with a nonparametric local regression, or LOESS, model to generate lookup tables at 0.8-degree resolution in lunar phase angle. Notably, the team chose to include data from earlier versions of the Cimel CE318-T Sun-Sky-Moon photometer despite their greater scatter, because restricting the dataset to newer instruments would have cut the number of Langley pairs by a factor of two to three, and the additional noise proved largely random.

To validate the corrected product, the researchers exploited the brief transition windows between day and night, comparing solar and lunar AOD measured by the same instrument within three-hour intervals under low and stable aerosol conditions. The analysis drew on 372 sites and 454 different instruments, yielding more than 280,000 comparison events. The result is striking: mean AOD differences at 440 nanometers were within plus or minus 0.005 for nearly all lunar phase angle bins, and at 675 nanometers and longer wavelengths the differences were within plus or minus 0.002. At low AOD, the mean solar-lunar difference of roughly 0.002 sits well inside the AERONET instrumental uncertainty of 0.01. The transition statistics were comparable to or smaller than lagged comparisons made entirely within day or entirely within night, confirming the absence of systematic calibration bias.

The assessment also probed what happens at high aerosol loading, where real atmospheric change over the transition interval dominates the comparison. Using both standard normal and bootstrap statistical methods, the team found a small but statistically significant tendency for daytime AOD to exceed nighttime AOD by about 0.013 to 0.016 in the 0.6 to 0.8 AOD bin. Rather than a calibration artifact, the researchers attribute this to genuine diurnal aerosol production, including daytime photochemistry and greater emissions from human activity and biomass burning. Scatter plots spanning all AOD values up to 1.0 showed nearly identical slopes of about 0.95 and concordance correlation coefficients of about 0.96 for both day-to-night and night-to-day transitions.

One subtlety of lunar photometry is that the Moon’s dimmer signal imposes a lower ceiling on measurable AOD. Under the network’s minimum signal threshold, solar photometry can theoretically capture total optical depth up to about 7.31 at an airmass of one, and observed solar maxima in the dataset, including extreme biomass burning events in Indonesia in 2015, came close to that limit. Lunar measurements under bright moon conditions reach roughly two to three times lower maximum AOD, and under minimum illumination the ceiling drops to about half the bright-moon value. The consequence is a disproportionate exclusion of high-AOD observations at shorter wavelengths: on average 3.2 percent of Level 2 lunar AOD at 440 nanometers is removed, three times the solar rate, with some dusty and smoky sites excluding more than 8 percent. At 675 and 870 nanometers the removal rates are 0.3 percent or less.

Cloud screening posed another question, since lunar measurements cannot perform the cross-scan curvature check that helps identify thin cirrus during solar operation. The team estimated the impact by comparing solar data processed with and without the curvature test across 259 sites. The differences proved negligible in most cases: aggregated multi-year monthly AOD differences ranged from only 0.0004 to 0.0014, and of nearly 15,000 individual months, just nine showed a mean AOD difference greater than 0.02. Even at Singapore, where the curvature check removes 5.1 percent more data points, the effect on monthly means was small. The researchers note that future versions may narrow remaining screening disparities, potentially with the help of machine learning algorithms.

The broader significance is considerable. Nighttime AOD from AERONET is already being used to validate satellite retrievals from the VIIRS Day-Night Band, which achieved correlations of roughly 0.8 over the United States, the Middle East, and the Indian Subcontinent. Lunar photometry is also the only option for monitoring Arctic haze during the polar night, which can last nearly four months at high-latitude sites such as Ny-Ålesund in Svalbard. With the new assessment demonstrating that the empirically corrected lunar AOD product agrees with solar measurements across diverse aerosol conditions worldwide, scientists now have a validated, continuous day-night record of atmospheric aerosols, opening the door to a truly complete picture of the 24-hour aerosol cycle and its consequences for air quality and climate.

Subject of Research: An assessment of lunar photometry in AERONET

Article Title: An assessment of lunar photometry in AERONET

Article References: Schafer, J., Eck, T. F., Slutsker, I., Gupta, P., Sorokin, M., Sinyuk, A., & Smirnov, A. (2026). An assessment of lunar photometry in AERONET. Atmospheric Measurement Techniques, 19(20), 6487-6502. https://doi.org/10.5194/amt-19-6487-2026

Image Credits: AI Generated

DOI: 10.5194/amt-19-6487-2026

Keywords: assessment, lunar, photometry, AERONET, scientific research, peer-reviewed research