A CSO Frontier Series Output- What is this?
The report involves the use of satellite data and GIS methods to produce statistics. In this section, the methodology is discussed.
This report used the monthly datasets from the NOAA Satellite Visible Infrared Imaging Radiometer Suite Day-Night Band (VIIRS-DNB) for measuring light emissions.
Satellite data on artificial light emissions has been collected and used to measure artificial light levels for a considerable period of time. The US Defence Meteorological Satellite Program (DSMP), a series of high-resolution military weather satellites, began publishing night light images of Earth from the early 1990s. This eventually led to Cinzano and Falchi’s work in the early 2000s developing a World Atlas of artificial light measurements using satellite imagery and instrumentation, which began with a map of artificial light in Europe. Therefore, while satellite imagery isn’t the only way of measuring artificial light (ground measurements are also used), it is still a tried and tested method.
The NOAA’s VIIRS-DNB (Satellite Visible Infrared Imaging Radiometer Suite Day-Night Band) system superseded the DSMP data on artificial light emissions, which had been available from 1992 to 2013. The World Atlas was updated in 2016 to take account of improvements in satellite measurement of artificial light, including the VIIRS-DNB dataset used in this report. There are many academic examples of using the VIIRS-DNB dataset to measure artificial light. In the Irish context, the work of Espey and Power use ground measurements to study light pollution in specific locations but use the VIIRS-DNB datasets for benchmark purposes. In New Zealand, a detailed study was made of rebuilding activity after a major earthquake using monthly VIIRS-DNB datasets.
The VIIRS-DNB monthly composite data is provided from the NOAA website and from the Colorado University of Mines, a research partner of the NOAA in analysing the VIIRS-DNB dataset. The data is public domain (once acknowledgement is made to NOAA in any publication) and is produced by month. The widespread use of the VIIRS-DNB dataset is aided by the datasets being relatively easy to use and analyse, consisting of TIFF standard raster files (containing light intensivity levels in lumens and corresponding geographical coordinates). Indeed, the data collection process for analysis involves no more than downloading and processing of publicly available VIIRS-DNB files.
The monthly VIIRS-DNB datasets contains the following relevant variables:
Latitude and longitude of measurements at 1km x 1km grid resolution. This resolution means that extremely low geographical level analysis (such as street or Census small areas) is not feasible but given that it is not possible to identify buildings or individuals from this dataset, disclosure control concerns are also minimised.
The average monthly composite cloud-free radiance figure. The unit is nanoWatt per square centimetre per square radian (nW/cm2/sr). A typical value for a rural area may be less than 5, while a city centre may have values well above 50.
The number of cloud free days in each month.
In terms of technical specifications, the datasets are provided using the standard TIFF raster format which simplifies the analysis process associated with the data. Due to their standard format and structure, the datasets can be combined with other GIS sources to permit the production of detailed statistics both at a national and international level. The files are divided into global segments, with this report using the Western Europe segment.
This monthly average VIIRS-DNB dataset is produced by the Earth Observation Group (EOG) at the NOAA. Prior to producing the monthly average, the EOG apply a clould filtering process. This removes effects that could impact on the quality of the night light measurement (such as cloud cover and lightning). This means the number of cloud-free days for each data point may fluctuate. As cloud cover can affect the accuracy of measurements, it is important to identify (and remove) cloud-affected data points. These data points affected by cloud cover are identified using data from the VIIRS Cloud Mask product (VCM). It should be noted that the monthly data does not filter certain light effects such as fires and boats. The resulting monthly datasets contain average radiance values and the number of cloud-free observations.
Statistics based on visual light range collected by satellite data (such as light emissions and land use analysis) are most accurate when clouds are absent (“cloud-free”) from the sky. It is not an exaggeration to say that cloud cover is usually the greatest quality issue in affecting satellite data for this and many other purposes. In many cases, given the Irish climate, promising research areas using such data are relatively infrequent, given our cloud cover.
To address this issue, the NOAA VIIRS-DNB datasets use cloud-free days for measuring light levels at night but while this helps ensure high quality measurements, it also means that for many months during a year there were insufficient cloud free days for light pollution statistics to be produced. This means that data is not always available for every month.
When processing the VIIRS-DNB data, imputation and editing are not used since missing data indicates that the VIIRS-DNB QA methodology has concluded that there is not enough valid light measurements (without clouds) to produce an accurate monthly average.
This report uses the NOAA average radiance (brightness) values estimates for artificial light from cleaned datasets (processed for cloud cover) combined with CSO geospatial boundaries. All units used for monthly average light levels (radiance) in this report are nanoWatt per square centimetre per square radian (nW/cm2/sr). For brevity, these are simply referred to as units.
The VIIRS-DNB dataset has already been classified and coded as monthly cloud-free averages based on nightly VIIRS-DNB measurements. These datasets were integrated with OSI boundary files (e.g. counties, settlements, electoral divisions etc.) using the ‘SF’ R package to permit geographical analysis. Next, the R raster package was used to generate summary light emissions statistics.
By overlapping the light emissions data points on the OSI boundary files it is possible to produce average artificial light estimates for the areas in these boundary files. Mapping files are also created in R using R temperature maps. These are graphical representations of the overlaid OSI and VIIRS-DNB datasets. The use of ‘SF’ and the ‘Raster’ R package allows processing of raster images.
R was chosen as it is open source (unlike the CSO’s other commonly used remote sensing/GIS analysis package ArcGIS) and there are already widespread resources available on coding and analysing satellite imagery. It also allows for the use of an R server which makes it easier to process the VIIRS-DNB dataset, although this report was prepared on a desktop PC. The summary output files also exist in tabular CSV files, which also include the detailed EU light emissions statistics.
By only making data available for months where there is sufficient cloud free days to derive the light output measurement, VIIRS-DNB greatly simplifies the quality assurance process. For the CSO, the main concern was whether the derived output was a good measurement of light levels at night. To assess these concerns, we can see that, firstly , the VIIRS-DNB dataset is used by many academic researchers studying artificial light emissions and its use in this respect has been widely accepted. Secondly, Irish urban artificial light levels are compared with known urban areas to establish if the light levels make sense (e.g. the extremely high levels of lighting in the Pembroke Docks area detected makes sense due to the presence of office buildings). This was similarly the case for key geographical features such as airports and central business districts.
Another important test is to see if the EU-designated Deep Sky Reserve areas had very low average artificial lights and this was indeed the case. Similar comparisons were made at an international level – the high levels in the Algarve and Randstad areas of Portugal and Holland respectively can be clearly seen in the VIIRS-DNB. In all these cases it can be seen that the VIIRS-DNB does correspond to known artificial light intensity distributions.
These contain monthly average light emissions calculated at various NUTS levels for various EU countries and detailed Irish statistics. The VIIRS-DNB monthly cloud-free average dataset ensures that cloud-affected datapoints do not affect the calculations and ensures high quality estimates. However, the disadvantage of this approach means that data is not available for every month.
There are methods that may be applicable to estimate light emissions for the missing months. For example, the CSO could consider the daily VIIRS-DNB datasets (which are also available) and use methods such as mosaics (combining of multiple images to fill in the “gaps”) to come up with estimates for light emissions adjusted for clouds. Mosaic based approaches are often used by researchers but can be technically challenging to implement from first principles. Furthermore, for all the research carried out on the VIIRS-DNB datasets, it does not appear that researchers have examined this issue in detail. A further challenge with the mosaic approach is that it can be very difficult to produce consistent estimates using this approach, which can be problematic from an official statistics perspective. For this reason, it was decided to simply use the standard VIIRS-DNB monthly composites. However, the use of mosaics may be revisited in the future.
Go to: 6. Acknowledgements
Learn about our data and confidentiality safeguards, and the steps we take to produce statistics that can be trusted by all.