ILLUMINA project - Luminaire inventories

Martin Aubé Ph.D.

Various methods based on geomatics and remote sensing were and will be explored to produce spatialized inventories of lighting devices (spectral luminosity, height, geometry) as well as properties of their immediate environment (reflectance, terrain elevation, etc). The current method used is the DMSP-OLS inversion but some other methods have been explored before.

Deconvolution of Astronauts images

This method use the images taken by the astronauts on board the International Space Station. We select the images taken with the 400mm lens so that the resolution is about 8 meters. After removing the natural light from the images, we deconvolve the image with Lucy-Richardson algorithm and then use Sobel filter to replace each small lighted patches into a single pixel. Each pixel correspond to a lighting device. We also use the color ratios to classify the lighting devices according to their spectral type.

Scan cities with the LANcube multiangle radiometer

This method allows to locate and identify spectral type of any lighting system located within about 15m from a road or parking lot.

VIIRS-DNB satellite radiance inversion

This is the default method used with Illumina v2.

DMSP-OLS inversion

This approach, which we think is the most promising is to use the DMSP-OLS 1km x 1km stable light product du derive the installed light luminosity. The idea rely on some assumptions.

We assume that the light received at the satellite level is dominated by direct light from the earth surface (we are neglecting contribution from atmospheric scattering).
We assume that the ground reflection is lambertian and that we can get a good estimation of the reflectance from the nearest modis band.
We assume that the user may provide the uplight value of the light output pattern toward zenith and the integration over all upward direction. The light output pattern allow the conversion from total luminisity of the lamp fixture to Watt per unit of steradian exiting the light fixture in a given direction.

These assumptions may be summarize by equation 1.

{# L_{OLS} \propto \Phi (\frac{1}{\pi} (1-F_{up})\rho cos(z)+ LOP(z)) #} (1)

Where
{#L_{OLS}#} is the relative OLS radiance,
{#\Phi#} is the radiant flux or luminosity of light fixtures,
{#\rho#} is the underlying reflectance.
{#LOP(z)#} is the angular light output pattern giving the emission per unit of solid angle at a given zenith angle z, and
{#F_{up}#} is the uplight fraction. This value is obtained from the integration of LOP(z) from z=0 to {#z=\pi/2#}

Equation 1 may be inverted in order to give a relative estimate of the light fixture luminosity or radiant flux. We assumed a zenith angle of z=0.

{# \Phi \propto \frac{L_{OLS}}{(\frac{1}{\pi} (1-F_{up})\rho + LOP(0))} #} (2)

Since OLS radiance are not calibrated to absolute units, we have to calibrate équation 2. This work is done by comparing the model result with a field measurement of the sky radiance. We suggest to use the zenith radiance.

The gridded ground reflectance is taken from MODIS satellite reflectance product.

Open gateway to light fixture inventories (not implemented yet)

The Open gateway called OpenSAND, aims to offer a research and education environment based on the concepts of the free software. Within the framework of the OpenSAND project, we will open and liberate the access:

to documentation
to scientific analysis and reduction methods
to raw and analysed data
to educational documents related to the project

The platform will be designed so that members could be users and/or contributors. Since the database to be setup is very vast, we will call upon competences of the public. We are convinced that the population can contribute significantly to the advance of sciences without any particular science education. The population will be invited to contribute by adding data to the system. Acquisition methods will be explained on the OpenSAND website. The scientific community, public organizations as well as the population will be able to use freely the system . Applications of such a system are obviously not limited to the problem of modeling of light pollution. Presentation of the Open science concept by Jean-Denis Giguère

Using remote sensing to estimate artificial luminosity

Comparison of the NDVI (Landsat TM) and the density of population (Statistics Canada) with artificial luminosity per unit of ground area.

Validation campaign: Sherbrooke (Qc, Canada) ** Geographical domain considered:
Attach:landsat-illum.jpg
Image Landsat-7 TM enhanced color composite.

Attach:domaine-illum.jpg
Topographic map.

Limits of the experiment domain

NS 45.50 deg to 45.25 deg.
E-o -72.00 deg to -71.75 deg.

Horizontal size of a cell:

0.00025 X 0.00025 deg (27.78 m N-S X 19.52 m E-W)

Size of the matrix:

1000 X 1000 X 50 vertical levels

The NDVI (Normalized difference vegetation index) exploits a specific feature of the vegetation spectrum to highlight its presence. The vegetation spectral reflectance shows a significant increase in the near-infrared, which is not observed for other usual surfaces such as the bare soil and water.

Attach:reflectance.gif
Spectral reflectance of vegetation, bare soil and water.

The NDVI is calculated as follows:

(CH2 - CH1)/(CH2 + CH1)

where CH1 = channel 1 (0.58-0.68 um) and CH2 = channel 2 (0.725-1.1 um)

We acquired Landsat-7 images ( R = infrared (band 4 (0.750-0.900 um)), V = red (band 3 (0,630-0.690um)), B = green (band 2 (0.525-0.605 um)), so that CH1 = V and CH2 = R correspond respectively to band 3 (0,630-0.690um) and band 4 (0.750-0.900 um). The image below corresponds to the calculation of the NDVI at a resolution of 0.001 deg (~100m). This resolution corresponds to the size of our validation parcels used to verify the relation mentioned above. The NDVI image was resampled to 100m resolution using an averaging algorithm from the original 15 m resolution.

Attach:ndvi-moyenne-100metre.jpg
Click to see the NDVI at 0.00025 deg (25 m) resolution 132 K ?

On this image, whitish gray levels correspond to high values of NDVI (i.e. denser vegetation cover). In fact NG=NDVI X 100 + 100, i.e. a level of gray of 100 is equivalent to a NDVI=0 whereas NG=200 is equivalent to NDVI=1. The levels of gray lower than 100 correspond to a negative NDVI. A negative NDVI, indicates an absence of vegetation (e.g. water, bare soil, asphalt, etc).

Attach:dpop2001.gif
2001 population density.Validation using luminaire distribution provided by Hydro-Sherbrooke. Dubé et al.. 2003

Attach:ndvi-lum_sm.jpg ?

InventaireEn