1. Introduction
Light pollution is increasingly recognized as a relevant threat to a sustainable world. The detrimental consequences of an inadequate use of artificial light at night are known to negatively affect the environment [
1,
2,
3], overall public expenditure and energy consumption [
4,
5], the intangible heritage associated with the contemplation of the starry skies [
6] and, potentially, human health [
7,
8,
9,
10,
11,
12,
13,
14].
Assessing the evolution of these unwanted effects is a pre-requisite for evaluating public lighting policies and making informed decisions on the social use of artificial light at night, particularly in regards to outdoor light sources. Two complementary approaches can be adopted, both of which provide relevant information on this issue. On the one hand, the evolution of the overall amount of artificial light emissions can be estimated from data provided by instruments operating on Earth orbiting platforms, such as the Visible Infrared Imaging Radiometer Suite Day/Night Band (VIIRS-DNB) [
15], or the radiometrically calibrated RGB images taken with off-the-shelf digital single-lens reflex cameras from the International Space Station [
16,
17]. On the other hand, some particular light pollution effects can be quantitatively measured from ground sites. One of these effects, probably the most conspicuous one for the majority of the population, is the increase in the night sky brightness (NSB) above its expected natural levels [
18,
19,
20]. The anthropogenic NSB is due to the scattering of artificial light by the air molecules and the aerosols present in the atmosphere, as well as by the enhanced reflections in the clouds. This NSB represents in itself a problem, since the artificially bright sky acts as a secondary light source that increases both the brightness within the visual field of nocturnal species and ground-level irradiance, modifying the rules of the struggle for survival, and reducing, for humans and non-humans alike, the visibility of the starry sky.
The NSB can be measured in different photometric bands, either visually [
21] or by using an ample variety of detectors [
22,
23,
24]. The widespread availability of these tools facilitates use by a high number of research institutions, public meteorological agencies, and social scientists alike, to perform routine measurements of the NSB, contributing to building large-size databases that can be used to monitor the evolution of the anthropogenic sky brightness [
25].
Given the extreme variability of the NSB signal, with characteristic multi-scale oscillation times ranging from seconds to years, the detection of small overall changes over time is a demanding task. Any such goal requires the choice of some numerical indicator with small intrinsic variance and acceptable robustness. In this work, we describe the use of one such indicators, the zenithal m
FWHM, based on the NSB data acquired under clear skies in the conditions of a moonless astronomical night (sun below −18°, and moon below −5°). This indicator allows one to overcome some drawbacks of the m
1/3 significant magnitude used in previous works [
20], which in dark-sky locations can be potentially biased toward darker values by the readings taken during extreme weather conditions (snow or dense fogs).
3. Results
The zenithal NSB data recorded by the 14 SQM detectors of the GNSBMN between 1 January 2015 and 31 December 2018 were used for this study (14 × 4 datasets). The yearly histograms of each station, both for the full set of NSB measurements and for the NSNM subset, show a reasonable constancy over the years, as illustrated in
Figure 5 with data from the Paramos station. Only in a few particular instances the histograms showed noticeable differences, due to changes experienced in the detectors proximity, for instance, before and after the installation of new light sources close to them, as in the station of O Cebreiro in February 2016.
The definite shape of the NSNM histogram is different for each station, according to the particular strength of the emissions of artificial light from its area of influence [
47] and the prevailing meteorological conditions, especially in regards to the statistics of the cloud cover.
Figure 6 displays some of these histograms, corresponding to the stations of Santiago de Compostela (urban), Guísamo (periurban), Illas Cíes (transition), and Labrada (rural and mountain), for year 2015. The reader may consult the full NSB data version of these histograms, for the same year, which is shown in
Figure 3 of [
20]. The location of the darker (clear skies) scattering peak in the magnitude axis, where the m
FWHM is calculated, is informative of the typical light pollution level at the site, and its area is indicative of the proportion of clear nights. The location of the brighter peak (cloud reflections), relative to the clear nights’ one, gives us a measure of the typical cloud amplification factor at the site [
20,
44]. Note that as we approach darker places, the light reflected from the clouds tends to be weaker and the peak corresponding to overcast nights approaches (until it is eventually confounded with, and can even surpass, as commented above) the one for clear nights.
The average number of NSB data per individual station and year, contained within the FWHM interval of the NSNM clear nights’ peak, was ~37,700, with a standard deviation of ~11,400. The largest number of records used to compute the yearly mFWHM was 66,562 (Labrada station, 2018), and the smallest one 17,710 (Illa de Sálvora station, 2015). To avoid artifacts, the control condition that the FWHM records span less than one magSQM/arcsec2 at either side of the peak maximum was checked before computing the mFWHM. To attenuate the effects of random histogram noise, the center of the clear nights’ peak on the magnitude axis can be estimated from a weighted average of the positions of its five largest NSB values, and the corresponding peak value can be estimated from the average of the three largest values. From these data, the FWHM interval can be defined, and the mFWHM is computed as the average of all individual NSNM records contained within this interval. Averaged over the four years and within the stations belonging to each area type (standard deviation in parenthesis), the FWHM was 0.61 (0.08) magSQM/arcsec2 in urban centers, 0.38 (0.08) magSQM/arcsec2 in periurban sites, 0.37 (0.04) magSQM/arcsec2 in transition regions excluding Illa de Ons due to the artifacts associated with the change of detector (see below), and 0.54 (0.19) magSQM/arcsec2. in rural and mountain sites.
Table 1 summarizes the results. The m
FWHM for each station and year, its overall change in the period 2015–2018, and its average change per year are indicated in columns 2 to 7 in units mag
SQM/arcsec
2. The uncertainty of the yearly m
FWHM results is 0.029 mag
SQM/arcsec
2 (one standard deviation). This uncertainty is basically due to the limited precision of the SQM readings, since the standard deviation of the mean of the individual readings is one to two orders of magnitude smaller. The temporal evolution is graphically depicted in
Figure 7.
Eleven out of the fourteen stations showed a net increase in the mFWHM at the end of the period, compared with its initial value (p < 0.001). This suggests that the skies, in the SQM photometric band, tended to become darker. In three stations, the overall evolution was toward brighter values (net magnitude decrease). In two of these stations, the brightening is significant to a p-value smaller than 0.001, and they correspond to two particular cases: O Cebreiro, where, as on commented above, new light sources strongly influenced the detector readings from February 2016 onwards; and Illa de Ons, where a faulty detector was replaced in 2018, causing a sudden 0.3-magSQM/arcsec2 drop in the records very likely due to an unintended change in the pointing direction of the device. The remaining station with a negative magSQM/arcsec2 net change is Santiago de Compostela, whose readings tended to be fairly constant across this period. This latter change is significant only to a p-value smaller than 0.25. If only the stations with positive darkening are considered, the inter-stations average overall change during the period 2015–2018 is 0.27 magSQM/arcsec2 (with an average standard deviation 0.13 magSQM/arcsec2; that is, the skies seem to be darkening in the SQM photometric band at a relevant rate of 0.09 magSQM/arcsec2 per year.
4. Discussion
The results presented in the section above suggest the existence of a significant trend toward darker readings in the majority of the stations of the GNSBMN. Darker readings may arise from causes intrinsic to detectors themselves (aging of the optics or some optoelectronic components), long-term changes in the average aerosol concentration profiles over the observation site and surrounding territories, and/or actual changes of the artificial light emissions toward the upper hemisphere within the spectral sensitivity band of the detector.
Regarding the first possibility, the readings of the SQM detectors have been shown to be stable over long periods of time, with small annual drifts [
54]. The glass windows of the GNSBMN detectors are periodically maintained according to the MeteoGalicia operating protocols, much like the remaining optical instruments installed in their stations. Visual inspection performed in some of them by the authors did not reveal any significant amount of deposits that could contribute to darkening the readings of the detectors. Besides dedicated maintenance, a positively contributing factor is the high amount of rainfall and the regime of moderate to strong winds prevailing in Galicia during extended periods of the year. The possibility of long-term changes in the aerosol concentration profiles, averaged over the clear and moonless astronomical nights of the year, cannot be ruled out, although it does not seem very likely that such a significant reduction in the aerosol content of the atmosphere took place during these years as to explain the recorded NSB values. The third possibility, actual reductions of the artificial light emissions toward the upper hemisphere within the spectral sensitivity band of the detector, is not implausible. The current trend of the replacement of gas discharge lamps by LED sources in Galicia is characterized by several features that give rise to contradictory NSB effects: (i) the substitution of low CCT high-pressure sodium lamps (~2000 K) by typically cold white light LEDs (4000 K) tends to increase the absolute amount of light scattered in the vicinity of the sources and hence to produce brighter skies in these places; (ii) this same substitution could allow a better delineation of the specific areas to be lit, reducing light spill and the overall amount of light sent to the upper hemisphere (strongly limited, in current ordinances), which affects the NSB in the opposite direction; and (iii) this substitution process takes place under a framework of lighting regulations that generally require achieving average lighting levels smaller than the pre-existing ones (albeit still greater than needed). This last factor, together with the fact that many outdoor public lighting installations are presently managed by private energy service companies (ESCO), whose net earnings depend on the savings on the overall energy consumption, may contribute a mild pressure toward smaller lighting levels. However, we cannot provide a closed conclusion for the time being. A detailed quantitative modeling of this transition process is currently underway and will be reported as soon as available.
It is worth mentioning that substantially similar results (not shown here) for the 2015–2018 evolution are obtained if the m1/3 significant magnitude is used as the indicator, instead of the mFHWM. As a matter of fact, only one station, Xares, shifts from positive to negative m1/3 change for this period. The net brightening observed in Xares using the m1/3 indicator is due to the episodic strong snowfalls experienced in year 2015 in this mountain region that gave rise to a non-negligible number of SQM readings in the extremely dark region of the histogram, thus biasing toward larger magnitudes the initial m1/3 value. During the remaining years, this phenomenon did not occur, and the Xares m1/3 show a trend of monotonic darkening. The average darkening for the ten stations with net positive change, estimated by the means of the m1/3, amounts to 0.09 magSQM/arcsec2 per year, the same value obtained with the mFWHM.
This study has been performed using the whole datasets acquired at the native sampling rate of the GNSBMN detectors (one reading per minute). However, the same results could be obtained with longer sampling periods. Reprocessing the datasets by taking one sample every 5 min, the maximum absolute difference of the mFWHM for all stations and all years, with respect to those obtained with the original one-minute readings, is smaller than 0.0009 magSQM/arcsec2, well below the uncertainty of the final results. If the sampling period is increased to one reading every 10 min, the maximum absolute difference is below 0.0017 magSQM/arcsec2.
The m
FWHM metric used in this work is affected by some limitations. First, its calculation depends on the selection of the FWHM region of the clear nights’ histogram peak. Our histograms were calculated from densitograms with 0.05-mag
SQM/arcsec
2 NSB resolution bins and ten-minute time resolution, each individual densitogram value being the ten-minute average of the one-minute detector readings. Coarser or finer bins may give rise to slightly different definitions of the FWHM interval. It is not expected, however, that these choices will give rise to systematic artifactual changes in the temporal evolution trends. Second, our measurements correspond to the NSB angularly averaged around the zenith, within the FOV of the SQM detectors, and the reported trends apply specifically to this photometric magnitude. Specific evolution indicators, following an approach similar to the one used here for the zenithal m
FWHM, could equally be defined for other NSB metrics such as the average magnitude of the celestial vault and the all-sky average light pollution ratio [
55], among others. Finally, let us also recall that the brightness evolution recorded in the SQM device-specific photometric band does not necessarily represent the evolution experienced in the human visual band. As recently shown by Sánchez de Miguel et al., changes in the spectral content of the light incident on the detectors may give rise to different evolution trends depending on the spectral band in which they are measured [
56].