SKYDUST

A guide to the method

A cloudless sky is
only the beginning.

Dust can dim the stars. Moonlight and city lights can wash them out. Unsteady air can blur a planet. Skydust brings these effects together to estimate what the night sky will be like to observe.

In the Gulf, dust, sea salt and other airborne particles can make a big difference even when the weather forecast shows no cloud. Understanding the view means asking what happens to light on its way through that air.

How Skydust works · Revised 6 September 2026 · Version 2.2 · Plain-language explanations, with optional technical notes

Transparency

How much light gets through?

Cloud and airborne particles weaken the light arriving from stars and other objects.

Sky brightness

How dark is the background?

Scattered moonlight and city glow make faint objects harder to distinguish.

Seeing

How steady is the image?

Turbulent air bends light unevenly, making fine detail shimmer or blur.

A night can be transparent but unsteady, or steady but hazy. These are different qualities, and they matter differently for a faint galaxy and a bright planet.

01 / THE METHOD

From atmospheric data to an observing estimate

Start with the air.Weather forecasts describe cloud, temperature, humidity and wind. An aerosol forecast estimates the combined effect of tiny particles suspended in the atmosphere.
Compare with observations.Aerosol instruments provide matched station checks; airport reports add local context. Satellite cloud ingestion is not active in this release. Each samples a different place or part of the atmosphere.
Follow the light.For the chosen time and site, the model calculates where objects and the Moon sit in the sky. It estimates light loss and the background glow in each direction.
Estimate the view.A star’s remaining brightness is compared with the background behind it. Weather, dew and seeing indicators add information about the observing conditions.

An observation is a measurement at a particular time and place. A forecast predicts later conditions. Skydust’s final view combines those inputs with physical calculations and simplifying assumptions.

02 / THE PHYSICS

Why stars fade in hazy air

Imagine looking at a distant light through a veil. Some light reaches you directly; some is redirected or absorbed along the way. Astronomers call that weakening extinction.

Aerosols are tiny airborne particles, including mineral dust, sea salt and pollution. Their combined effect on direct light is described by aerosol optical depth, or AOD. It measures the effect through a vertical column of air: a larger number means more light is removed from the direct beam.

AOD has no unit, and it is not a percentage. Skydust uses the value at 550 nanometres, a wavelength in the green part of visible light, as a reference. It is also different from a ground-level particle concentration or air-quality index.

Looking low means looking through more air

Light from a star overhead takes a relatively short route through the atmosphere. Near the horizon it takes a much longer, slanting route. The amount of atmosphere along that route, relative to looking straight up, is called airmass.

Illustration only. The real atmosphere has no sharp upper edge and curves around Earth. Skydust uses the Kasten–Young airmass approximation to account for the changing path near the horizon.

Try the physics

How much direct starlight survives?

Change the particle load and the star’s height. This is an illustration at 550 nm, not a live forecast. It isolates aerosol extinction and leaves out cloud, gas and molecular effects, and background glare.

Aerosol optical depth0.30

Fewer particles · 0More particles · 1.2

Star’s height above the horizon30°

Low · 5°Overhead · 90°

55%

of the direct light remains after aerosol extinction

The line of sight crosses about 2.0 times the overhead air column.

Same atmosphere, different directions: AOD = 0.30
Star’s height	Approximate airmass	Direct light remaining*
90° · overhead	1.0	74%
30°	2.0	55%
10° · low	5.6	19%

*Calculated for aerosols alone at 550 nm, assuming the same atmospheric profile in each direction. These are light-transmission examples, not star-visibility scores. Other sources of extinction remove additional light.

The equations behind the example

The Beer–Lambert law connects optical depth to the fraction of direct light transmitted:

T_aerosol = e^{−τ₅₅₀ X}
Δm_aerosol = (2.5 / ln 10) τ₅₅₀ X ≈ 1.086 τ₅₅₀ X

T: Fraction of direct light remaining; multiply by 100 for a percentage.
τ₅₅₀: Vertical aerosol optical depth above the observer at 550 nm.
X: Relative airmass, approximately 1 overhead.
Δm: Extra dimming in astronomical magnitudes. A larger magnitude means a fainter object; one magnitude is a factor of about 2.512 in brightness.

The reference-wavelength extinction estimate is:

k₅₅₀ = k_Rayleigh,550 + 1.086 τ₅₅₀ + k_ozone,550
m_observed ≈ m_{above atmosphere} + k₅₅₀ X

Here k is the extinction per unit airmass, in magnitudes. Rayleigh scattering is light redirected by air molecules; ozone absorbs part of the light. The sum is a reference-wavelength approximation. A camera filter or the eye responds to a range of wavelengths.

The interactive example uses the Kasten–Young (1989) approximation with apparent altitude h in degrees:

X = 1 / [sin(h) + 0.50572(h + 6.07995)^−1.6364]

A common airmass factor is useful when the atmospheric profile is reasonably similar along each line of sight. Near the horizon, uneven haze, refraction and terrain make real conditions less predictable.

Reference: Kasten & Young (1989). Table values are calculated from the equations above.

Particles can also make the background brighter

The light removed from a star’s direct beam is only part of the story. Particles can redirect light from the Moon and artificial lighting into the patch of sky you are looking at. That glow reduces contrast, like trying to read faint writing on an illuminated screen.

The Moon’s phase matters, but so do its height and the angle between it and your target. Dust and droplets often scatter strongly in the forward direction, helping create bright glare around the Moon.

Skydust therefore estimates brightness by direction. Two equally bright stars can be differently visible if one lies in a bright patch of sky and the other has a darker background. The predicted cutoff is called the limiting magnitude: the faintest star an observer is expected to distinguish under the assumed conditions.

Your eyesight, dark adaptation and nearby lights also affect that cutoff. A naked-eye estimate does not directly predict what a telescope or camera will record.

How the moonlight calculation works

The method builds on Krisciunas & Schaefer (1991) and Winkler (2022). It follows the incoming lunar beam to where scattering occurs, then the scattered light toward the observer. This is more realistic than applying the full incoming atmospheric loss before every scattering event.

A phase function describes which directions scattered light favours. The Henyey–Greenstein approximation uses an asymmetry parameter, g, to describe forward scattering. The single-scattering albedo, ω, is the fraction of extinction caused by scattering rather than absorption.

Kim et al. (2011) report mean dust values near g = 0.752 and ω = 0.944 at 550 nm. These are useful regional starting points, not measurements of the particles over a particular site tonight.

Normalised phase functions can be weighted by their scattering optical depths; the aerosol contribution is ωτ. This constrains the mixture but does not establish the absolute brightness. Local calibration remains necessary, and repeated scattering becomes more important in optically thick air. No single correction in magnitudes applies to every dusty night.

Krisciunas & Schaefer (1991) · Winkler (2022) · Kim et al. (2011)

Humidity changes some particles

Sea salt and some pollution particles take up water as humidity rises. They grow, and their effect on light changes. Mineral dust does not respond in exactly the same way. A humid coastal night and a dusty inland night can therefore have different glare even at similar AOD.

The CAMS aerosol model already represents humidity-dependent particle growth in its optical calculations. Applying an extra blanket multiplier to that AOD would risk counting the same effect twice. Skydust’s additional humidity adjustment concerns the assumed scattering pattern; it should be treated as a model assumption that needs local checks.

Background: Rémy et al. (2019), IFS aerosol scheme.

What the humidity adjustment assumes

Skydust shifts dust-like parameters toward more forward-scattering, weakly absorbing particles as the water-attracting aerosol fraction increases. The assumed values approach g = 0.85 and ω = 1.

These values are a simplified representation of a mixture. Their usefulness depends on knowing the particle types and the humidity where those particles are, which may differ from humidity at ground level. Stronger forward scattering can redistribute glare toward the Moon; it does not guarantee a fixed improvement elsewhere in the sky.

Why altitude, sharpness and dew need separate treatment

Being above some haze can help

A mountain site may sit above part of the aerosol layer. But particles can remain aloft after the ground cools at sunset. The benefit depends on the height of the haze, not just the height of the observer. This release retains the supplied aerosol column when the model-surface reference is unknown; it does not automatically subtract a sea-level layer at mountain sites.

Clear air can still give a blurry image

Temperature variations and turbulence distort incoming light. Skydust uses high-altitude wind as a rough seeing indicator; turbulence nearer the ground and around the telescope also matters.

Dew is different again: a lens or mirror can lose heat to a clear sky and become colder than the surrounding air. Condensation can begin when its surface reaches the dew point. Humidity, wind and the equipment’s cooling all affect the risk.

On seeing and turbulence: ESO observing-condition definitions.

03 / THE DATA

What each source contributes

The method draws on five types of external feed. They complement one another, but they do not all measure the same thing or provide continuous coverage at every site.

Open-Meteo

Model forecast

Cloud, humidity, dew point and wind provide the weather context. Aerosol optical depth comes from the Copernicus Atmosphere Monitoring Service (CAMS), distributed through Open-Meteo.

Resolution depends on the product. Open-Meteo lists the global CAMS composition feed at about 45 km; this must not be confused with finer weather grids. Interpolated values at a coordinate do not add local measurements. Aerosol product documentation.

NOAA Aviation Weather Center

Airport observations

METAR reports describe visibility, cloud base and weather at airports. Skydust uses airports serving Dubai, Abu Dhabi, Sharjah, Al Ain, Ras Al Khaimah and Fujairah.

They are valuable checks near the surface. Airport visibility alone cannot tell us how much aerosol lies above the whole observing site.

NASA AERONET

Column measurements

Specialised instruments infer AOD from how much direct sunlight, or suitable moonlight, reaches them. UAE reference sites include Masdar City, Mezaira and Khalifa University, subject to data availability.

Nighttime measurements require a suitable instrument, a sufficiently bright Moon above the horizon, and usable sky conditions. Availability and quality level must be checked for each record; an instrument’s uncertainty is not the forecast’s uncertainty.

EUMETSAT Meteosat

Satellite observations

Recent imagery can help locate cloud. The experimental reader in this release is disabled until its product, encoding and acquisition times are verified. Cloud values currently come from weather forecasts.

Cloud detection and position have uncertainty. Tall clouds can appear displaced because the satellite views them at an angle, an effect called parallax. A recent image helps assess current conditions; it is not itself a forecast for the whole night.

CelesTrak

Orbit data

Orbital elements describe satellite paths and support estimates of possible trails.

A predicted path is only part of the problem. Whether a trail appears in an image also depends on illumination, satellite brightness, exposure and the camera’s field of view.

Star catalogues and the observing calendar

The Yale Bright Star Catalogue through CDS supplies star data. NASA’s solar and lunar eclipse canons and the International Meteor Organization’s calendar provide event references.

Skydust calculates positions, rise and set times, phases and local observing geometry. An event being above the horizon does not mean the sky will be clear enough to see it.

What this version actually calculates

Every hour uses its matching weather and aerosol input. The headline sky brightness is for the zenith, with the Moon’s actual angular distance from that direction. The selected site’s sky dome and star field use the same time and location.

Natural background, artificial light and scattered moonlight are added as amounts of light before conversion to magnitudes. A faint artificial-light value is not an additional natural sky. The forecast site map shows a modelled moonless reference background, separate from the hourly weather estimate.

What does a zero calibration offset mean?

The Moon term is multiplied by 10^−0.4δ. At δ = 0 that multiplier is 1: the baseline moonlight remains. Zero does not turn moonlight off, and it does not mean the absolute normalisation has been validated.

The default repeated-scattering factor is a Rayleigh-based approximation. The earlier extension to all aerosol scattering is experimental and is not active by default. Absorption by gases is retained in the attenuation term.

AERONET comparisons use raw CAMS at the instrument location and matching hour. They do not automatically correct all sites or later forecasts. METAR visibility is not inverted into a trusted column AOD. Historical sky-quality percentiles are withheld until a consistent aerosol record is available.

SQM comparisons are recomputed after an observation. They are retrospective baselines, not saved forecasts from before the event. Compatible clear-sky readings first constrain city glow, then the moonlight offset. Independent data are still needed to test the fitted result.

From individual readings to a measured map

Forecasts answer “what might conditions be like?” Measurements answer “what was recorded here?” The SQM Map preserves that distinction. Each point retains its date, instrument and observing context; no modelled value is substituted for a missing reading.

For the standard comparison, select repeated SQM-L-family zenith readings with reported cloud-free skies, spread ≤0.10 mag, and stored geometry showing Sun ≤−18° and Moon ≤−2°. These collection rules support comparison; they do not make the sky or instruments identical.

Within a 0.01° coordinate cell and instrument model, first take the median of submissions per GST noon-to-noon night. Then take the median of those nightly values, so a heavily sampled night does not dominate simply because it has more submissions. For an even count, the median is the midpoint of the two middle magnitude values. This is a descriptive median in magnitude units, not an arithmetic mean of light flux.

Show the number of nights, dates and range of nightly medians. A range describes the observed spread, not uncertainty in the true long-term average. Haze, airglow, Milky Way position and instrument differences remain possible influences. Hold independent nights or sites aside before claiming that fitted predictions are validated.

Explore the measured map

04 / LIMITS & CHECKS

Where the estimates need caution

What the current method can and cannot establish
Output	Current limitation	How to read it
Star dimming	Based on established extinction physics, but sensitive to AOD and the aerosol height profile.	Useful for understanding trends; least reliable near the horizon or in uneven haze.
Moonlit sky brightness	Absolute brightness has not yet been established by local sky-meter calibration.	A guide to likely glare patterns, with no demonstrated absolute accuracy.
Light pollution	Estimated from population and an empirical distance model.	Broad site comparisons; actual lighting and atmospheric conditions can differ.
Seeing	A wind-based proxy, without a full turbulence profile.	A rough indicator, not a measurement of image sharpness in arcseconds.
Dew	Uses an assumed amount of equipment cooling.	A risk estimate, not the measured temperature of your optics.
Cloud	Forecast grids can miss local or fast-changing cloud; satellite data have their own limits.	Check recent conditions as well as the later forecast.

Three lessons from the model’s development

Sunset does not clear the aerosol column. The initial method linked the aerosol layer directly to the daytime mixed layer, which becomes shallower at night. Allowing for particles remaining aloft avoids an exaggerated improvement at mountain sites. The remaining uncertainty is the actual vertical aerosol profile.
Horizontal visibility is not vertical AOD. A surface visibility report and a column measurement describe different paths through the air. Combining them can estimate an effective haze depth under assumptions about uniformity; it does not uniquely measure the real layer structure.
A local fit needs an independent check. The supplied city-glow model contains a provisional distance rule and a Dubai brightness anchor. The claimed reference-site fit was not accompanied by reproducible measurements. Even a documented local fit would need testing at other sites and on new nights.

These changes address weaknesses in the assumptions. Their effect on forecast accuracy needs to be demonstrated with dated observations, reproducible calculations and independent comparisons.

How accuracy should be checked

The useful question is: how often does a forecast made in advance match a later observation? That is different from adjusting a model to measurements already available.

AOD forecasts can be checked against AERONET. A meaningful result needs the forecast lead time, observation quality, site, wavelength, dates and number of matched cases. It should report the typical error and whether the model tends to predict too much or too little aerosol.

A percentage such as “within 15%” is insufficient on its own: a small absolute error becomes a large percentage when AOD is very low. Absolute errors should also be shown. Measurements used to correct the forecast should be kept separate from the observations used to evaluate that correction.

Sky brightness needs its own check using a Sky Quality Meter (SQM) or calibrated photometry. Matching AOD alone does not validate the predicted sky glow, star count or seeing.

05 / FURTHER READING

The science behind the explanation

Kasten & Young (1989), Revised optical air mass tables and approximation formula — atmospheric path length.
Krisciunas & Schaefer (1991), A Model of the Brightness of Moonlight — estimating scattered lunar light.
Winkler (2022), A revised simplified scattering model for the moonlit sky brightness profile based on photometry at SAAO — revised scattering geometry and phase functions.
Kim et al. (2011), Dust optical properties over North Africa and Arabian Peninsula derived from the AERONET dataset — regional dust properties.
Rémy et al. (2019), Description and evaluation of the tropospheric aerosol scheme in the Integrated Forecasting System (IFS-AER, cycle 45R1) of ECMWF — aerosol modelling and humidity effects.

Data providers are linked in the section above. Open-Meteo data are provided under CC BY 4.0 with upstream attribution requirements, including CAMS for aerosol data. Thanks to the AERONET investigators and staff who establish and maintain the observing sites. Eclipse reference data: Fred Espenak and Jean Meeus, NASA’s Goddard Space Flight Center.