---
title: 'Otus 3: Night-Time Light Pollution Modeling'
url: https://www.emergentmind.com/topics/otus-3
type: topic
---

# Otus 3: Night-Time Light Pollution Modeling

Otus 3 is a light-pollution modelling software package developed by DarkSkyLab to convert satellite-measured ground radiance into physically interpretable maps of night-time artificial light. Its defining feature is a shift away from the traditional emphasis on zenith night-sky brightness toward modelling the illumination actually received on the ground. In this framework, Otus 3 computes both zenith luminance and horizontal ground illuminance, and it separates ground illumination into diffuse skyglow and direct lamp emission. The system was applied to metropolitan France using VIIRS-DNB ground radiance data and a large sky-brightness calibration dataset, yielding the first artificial illuminance maps for France for 2013–2018 and 2019–2024 [2510.02977].

## 1. Conceptual scope and motivation

Otus 3 was introduced in the context of artificial light at night (ALAN), where the dominant modelling tradition has focused on Night Sky Brightness (NSB), especially at the zenith. Otus 3 redefines the problem by treating ALAN not only as a question of sky brightness but also as a question of total illumination on the ground. The central rationale is that many biological and ecological effects are more directly linked to horizontal ground illuminance than to the brightness of the sky overhead. This motivates the software’s dual output structure: zenith luminance remains available, but it is complemented by ground-level metrics intended to better represent exposure for living organisms [2510.02977].

The modelling pipeline is organized around two stages: an emission model and an atmospheric propagation model. The inputs are VIIRS-DNB ground radiance measurements, and the outputs are maps of artificial skyglow luminance, artificial skyglow illuminance, artificial direct illuminance, and their totals after addition of natural dark-sky components. According to the authors’ description, this makes Otus 3 the first software in their formulation that can model total artificial illumination on the ground rather than only sky brightness.

The France application provides the practical context for the software’s design. France is presented as a useful test case because public lighting is often switched off during the middle of the night and regulations restrict upward emission and spectral characteristics. Otus 3 was therefore used not only to map light pollution but also to assess whether these policy changes are visible in large-scale ALAN trends.

## 2. Output variables and photometric decomposition

A fundamental distinction in Otus 3 is between zenith luminance and horizontal ground illuminance. Zenith luminance corresponds to the traditional skyglow metric, whereas horizontal ground illuminance measures the total light falling onto a horizontal surface at ground level. The ground quantity is decomposed into three parts: Artificial Skyglow Illuminance (ASI), Artificial Direct Illuminance (ADI), and Natural Ground Illuminance (NGI). The zenith quantity is decomposed into Artificial Skyglow Luminance (ASL) and Natural Zenith Luminance (NZL) [2510.02977].

The pipeline combines these components in a specific way. Artificial skyglow illuminance plus natural ground illuminance gives diffuse ground illuminance. Diffuse ground illuminance plus direct artificial illuminance gives total ground illuminance. Artificial skyglow luminance plus natural zenith luminance gives total zenith luminance. This decomposition is central to the software’s interpretation of light pollution because skyglow and direct lighting have different spatial structures: skyglow is smooth and long-range, whereas direct illumination is highly local and concentrated around urban infrastructure.

This distinction also addresses a recurrent misconception in ALAN assessment: sky brightness alone is not equivalent to ecological exposure. Otus 3 treats skyglow and direct emission as separate physical pathways and thereby makes it possible to distinguish environments that have similar zenith brightness but different ground-level exposure.

## 3. Emission modelling and atmospheric propagation

The emission model follows Garstang’s parameterization. The upward radiance from the ground is written as

$$
RAD(\theta) = k \times GLE \times \frac{\left(2G(1-F)\cos\theta + 0.554F\theta^4\right)}{2\pi},
$$

where \(RAD(\theta)\) is the ground radiance emitted toward zenith angle \(\theta\), \(GLE\) is the Ground Luminous Emittance, \(F\) is the fraction of light emitted above the horizon, \(G\) is the fraction of incoming light reflected by the ground, and \(k\) is a normalization parameter that captures spectral overlap between satellite and lamp emissions and phenomenological effects such as albedo, masking by buildings, and vegetation. Once \(GLE\) has been inferred from observed radiance, the direct artificial illuminance on the ground is computed as

$$
ADI = GLE \times (1-F).
$$

In this formulation, direct artificial illuminance is explicitly distinct from skyglow because it represents light reaching the ground without first being scattered through the atmosphere [2510.02977].

Skyglow is treated through convolution of the \(GLE\) map with diffusion kernels derived using the SkyGlow software. These kernels depend on distance, atmospheric properties, and emission geometry. They are used separately for artificial skyglow luminance at zenith and artificial skyglow illuminance on the ground. The paper notes that kernels can be asymmetric in principle, for example under anisotropic city emission or terrain effects, but the France study uses symmetric kernels.

The atmospheric treatment includes two explicit regimes: clear-sky conditions and low-cloud conditions with cloud base at \(1.0~\mathrm{km}\) and full cloud fraction. For the main France simulations, the atmospheric parameters were set to \(AOD_{550}=0.1\), aerosol gradient \(0.65~\mathrm{km}^{-1}\), and molecular scale height \(8~\mathrm{km}\). Clouds are not treated as a simple scalar correction. Instead, they define a separate propagation regime with its own kernels. The qualitative consequences are spatially heterogeneous: far from cities, clouds reduce skyglow by blocking distant ALAN; in dark rural areas they also reduce natural night brightness by blocking stars and airglow; over cities and nearby suburbs, they can strongly increase artificial skyglow illuminance by reflecting urban light downward.

## 4. VIIRS input data, cleaning, and model calibration

Otus 3 is driven by VIIRS-DNB data from the Suomi NPP satellite. VIIRS DNB provides global night-time radiance observations in the \(0.5\text{–}0.9~\mu\mathrm{m}\) range with roughly \(750~\mathrm{m}\) spatial resolution. For the France application, computations were performed on \(500~\mathrm{m}\) pixels in EPSG:2154 for metropolitan France. The authors used NASA Black Marble calibration products and built two long-term composites, 2013–2018 and 2019–2024, by averaging yearly composite images and retaining only radiance detected in at least two of the six years in each period. This was intended to suppress spurious ephemeral lights [2510.02977].

A specific preprocessing problem arises because VIIRS observations contain not only surface radiance but also diffuse halo emission around bright cities. Since that halo corresponds to skyglow seen from above rather than direct emission from the surface, Otus 3 removes it using a ratio-based cleaning criterion tied to modelled skyglow:

$$
RAD / ASI < 0.35 \frac{nW / (sr~cm^2)}{mlux}.
$$

The cleaning is iterative in principle, although the authors state that one pass is sufficient because the halo is dominated by the brightest city pixels.

The France implementation also uses a radiance asymmetry measure derived from near-nadir and off-nadir VIIRS observations,

$$
ASY = \frac{RAD_{near\text{-}nadir} - RAD_{off\text{-}nadir}}{RAD_{near\text{-}nadir} + RAD_{off\text{-}nadir}},
$$

to assign emission-function classes. Areas with \(ASY > -0.07\) are assigned Model ID 1 with \(F=0.05\), and areas with \(ASY < -0.07\) are assigned Model ID 2 with \(F=0.15\). In both cases, \(G=0.15\). The fitted normalization parameter \(k\) is 278 for Model 1 and 371 for Model 2.

Calibration was performed using a large independent sky-brightness dataset collected with the Ninox device, which combines a Unihedron SQM-LU with automatic minute-by-minute zenith sky brightness measurements after sunset. The broader campaigns since 2018 in France, Chile, and Arizona produced more than 25 million NSB measurements across over 400 sites, but the Otus 3 calibration for metropolitan France used 139 sites with sufficient NSB measurements and good VIIRS coverage. To extract a robust clear-sky reference, the authors introduced Clear Sky Brightness (CSB). Clear intervals were identified using a Night Sky Stability metric,

$$
NSS = 7 + \log_{10}(Var(NSB - \widehat{NSB})),
$$

where \(\widehat{NSB}\) is a 10th-degree polynomial fit to the NSB time series; two-hour segments with \(NSS < 2.7\) were treated as clear. Model calibration then used a maximum-likelihood fit of simulated zenith luminances to observed CSB values, varying \(k\) and \(NZL\). The best-fit natural zenith luminance was

$$
NZL = 0.214~mcd~m^{-2}.
$$

## 5. France application and empirical findings

Using the calibrated framework, Otus 3 produced the first French artificial illuminance maps for the periods 2013–2018 and 2019–2024. The maps represent middle-of-the-night conditions defined by the VIIRS overpass window of roughly 00:30–02:30 local time. In the 2019–2024 maps, diffuse skyglow is generally much weaker than direct illuminance, diffuse illumination is smooth and extended, and direct illumination is concentrated in cities and linear infrastructure such as highways and rivers. The darkest regions identified in the maps are the Massif Central, Brittany, Corsica, and Nièvre [2510.02977].

The headline quantitative result is that artificial ground illuminance in the middle of the night decreased by \(23\%\) between 2013–2018 and 2019–2024 across metropolitan France. The decline is described as especially pronounced in rural areas around small towns and cities, where public-lighting shutdowns are common. This result is notable because it contrasts with the broader global trend of increasing ALAN reported in satellite-based analyses.

Otus 3 also produces clear-sky and low-cloud scenarios. Under full low-cloud cover, urban areas become much brighter in skyglow, whereas rural areas darken because clouds block both natural night light and distant city light. This reinforces the software’s broader claim that cloud effects are strongly context-dependent and cannot be reduced to a uniform brightening or darkening factor.

A plausible implication is that Otus 3 changes the empirical basis on which national-scale light-pollution trends are assessed. The France result does not merely indicate a change in zenith sky brightness; it indicates a change in ground illuminance, which the authors argue is often the more relevant exposure metric for organisms.

## 6. Uncertainty, limitations, and prospective extensions

The paper gives explicit uncertainty estimates for major outputs. Zenith luminance uncertainty is about \(18\%\) for skyglow, with an additional natural-sky uncertainty of \(\Delta NZL = 0.04~mcd~m^{-2}\). Ground skyglow illuminance uncertainty is assumed similar to ASL, about \(18\%\). Natural ground illuminance uncertainty is given as \(\Delta NGI \approx 0.2~mlux\), with nominal \(NGI \approx 0.7~mlux\). Direct artificial illuminance has substantially larger uncertainty, estimated at \(49\%\), because it depends sensitively on the emission function and does not average out over many pixels [2510.02977].

The paper also identifies a VIIRS sensitivity floor: minimum detectable radiance is approximately \(0.5~nW~sr^{-1}~cm^{-2}\), corresponding to a minimum detectable \(ADI \approx 25~mlux\). This implies that very low-level rural direct lighting may be missed. Additional limitations include uncertainty in cloud and aerosol parameterization, the absence of independent ground illuminance measurements for direct validation, residual Milky Way contributions in the darkest CSB measurements, the use of symmetric rather than anisotropic kernels, and reliance on VIIRS observations that are temporally restricted to late-night overpasses.

The proposed next steps are correspondingly concrete: add more direct illuminance measurements using lux meters, integrate TESS photometer data, refine cloud and aerosol treatment, improve the Milky Way background model, extend analyses to the beginning and end of the night, and exploit higher-resolution satellite data such as SDGSAT-1. This suggests that Otus 3 is both a modelling framework and a photometric reorientation of ALAN studies: it preserves zenith luminance as a standard astronomical observable while elevating ground illuminance to a primary quantity for ecological and environmental analysis.

Source: https://www.emergentmind.com/topics/otus-3