Evolution of Brightness and Color of the Night Sky in Madrid

remote sensing

Article Evolution of Brightness and Color of the Night Sky in Madrid

José Robles 1,*, Jaime Zamorano 1,2 , Sergio Pascual 1,2 , Alejandro Sánchez de Miguel 3 , Jesús Gallego 1,2 and Kevin J. Gaston 3

1 Departamento de Física de la Tierra y Astrofísica, Facultad CC. Físicas, Universidad Complutense de Madrid, Plaza de las Ciencias 1, 28040 Madrid, Spain; jzamorano@ﬁs.ucm.es (J.Z.); sergiopr@ﬁs.ucm.es (S.P.); [email protected] (J.G.) 2 Instituto de Física de Partículas y del Cosmos, IPARCOS, 28040 Madrid, Spain 3 Environment and Sustainability Institute, University of Exeter, Penryn, Cornwall TR10 9FE, UK; [email protected] (A.S.d.M.); [email protected] (K.J.G.) * Correspondence: [email protected]; Tel.: +34-6-5520-4262

Abstract: Major schemes to replace other streetlight technologies with Light-Emitting Diode (LED) lamps are being undertaken across much of the world. This is predicted to have important consequences for nighttime sky brightness and color. Here, we report the results of a long-term study of these characteristics focused on the skies above Madrid. The sky brightness and color monitoring station at Universidad Complutense de Madrid (inside the city) collected Johnson B, V, and R sky brightness data, Sky Quality Meter (SQM), and Telescope Encoder Sky Sensor-WiFi (TESS-W) broadband photometry throughout the night, every night between 2010–2020. Our analysis includes a data ﬁltering process that can be used with other similar sky brightness monitoring data. Major changes in sky brightness and color took place during 2015–2016, when a sizable fraction of the streetlamps in Madrid changed from High-Pressure Sodium (HPS) to LEDs. The sky brightness detected in the Johnson B band darkened by 14% from 2011 to 2015 and brightened by 32% from 2015 to 2019.

Citation: Robles, J.; Zamorano, J.; Keywords: light pollution; photometry; sky brightness; street lighting retrofit Pascual, S.; Sánchez de Miguel, A.; Gallego, J.; Gaston, K.J. Evolution of Brightness and Color of the Night Sky in Madrid. Remote Sens. 2021, 13, 1511. 1. Introduction https://doi.org/10.3390/rs13081511 The nighttime environment is being altered across increasingly large areas of the Earth as a consequence of the introduction of artificial lighting. Indeed, the vast majority of Academic Editor: Miroslav Kocifaj the human population now lives under light-polluted skies [1]. Such light pollution is exacerbated by the widespread failure to limit the use of artificial lighting to the places, Received: 20 February 2021 forms, and times in which it is required to deliver the intended human benefits. Artificial Accepted: 7 April 2021 Published: 14 April 2021 lighting is, for example, often excessively bright, and is allowed to spread into areas (spatial trespass) and times (temporal trespass) beyond those in which it is actually needed. This

Publisher’s Note: MDPI stays neutral has a wide array of consequences for living organisms, including measurable effects on with regard to jurisdictional claims in human health [2,3] and on wild plants and animals [2–4], many of which arise from the published maps and institutional afﬁl- impacts of artiﬁcial nighttime lighting on the timings of biological activities [5]. iations. Given these environmental impacts, it is important to understand the effects of widespread and ongoing street lighting changes from legacy (mainly High-Pressure Sodium; HPS) to Light-Emitting Diode (LED) technology [6–8]. A common method for assessing the extent and impact of light pollution is using remote sensing and ground-based techniques to measure night sky brightness at the zenith as a proxy. Whilst the conse- Copyright: © 2021 by the authors. Licensee MDPI, Basel, Switzerland. quences of street lighting changes for direct light emissions are relatively straightforward to This article is an open access article determine, this is more challenging for the brightness and color of the night sky (skyglow). distributed under the terms and First, although skyglow can be detected from nighttime satellite imagery (e.g., Defense Me- conditions of the Creative Commons teorological Satellite Program/Operational Line-Scan (DMSP/OLS) and Suomi National Attribution (CC BY) license (https:// Polar Orbiting Partnership/Visible Infrared Imaging Radiometer Suite—Day-Night Band creativecommons.org/licenses/by/ (SNPP/VIIRS-DNB); [9,10]), the associated sensors are broadband and do not detect blue 4.0/). light. Thus, nighttime satellite imagery does not provide color information.

Remote Sens. 2021, 13, 1511. https://doi.org/10.3390/rs13081511 https://www.mdpi.com/journal/remotesensing Remote Sens. 2021, 13, 1511 2 of 25

This is important both because these colors can change with lighting technology and because emissions of different colors can influence skyglow in different ways. Indeed, concerns have arisen that temporal trends in the emissions measured using such imagery can be misleading if the consequences of shifts in the spectrum of artificial light emissions are not accounted for [11]. Second, whilst ground-based sensors are available for measuring skyglow (e.g., Sky Quality Meter (SQM); STARS4ALL Telescope Encoder Sky Sensor-WiFi (TESS-W) photometer [12]), and networks of such devices are being developed, these also tend to be broadband and, again, proper adjustments need to be made for spectral shifts to evaluate temporal trends [6,11,13]. SQMs, broadband photometers used for network monitoring stations, are an example of a device in need of such adjustments. Sánchez de Miguel et al. [11] showed that properly assessing night sky quality with SQMs needed adjustments via color indices. The contributions of both night sky brightness and color have been recognized by researchers studying light pollution. Kyba et al. [14] used color as an indicator of light pollution and pointed out the need for photometer measuring in several bands to monitor sky color evolution. The authors used standard SQMs with red, green, and blue filters to monitor night sky color evolution, however, the filters were not the standard ones used in astronomy. In particular, the red filter did not properly cover the HPS spectra produced by streetlights. More recently, Barentine et al. [7] used several different data collection instruments, including a Sky Quality Meter (SQM-L) and an unmodified Canon T5i DLSR, to measure night sky brightness changes in Tucson, AZ (USA) following a street lighting retrofit from a mixture of high- and low-pressure sodium (HPS/LPS) streetlamps to fully shielded 3000 K LED. They reported a decrease in skyglow, however, they did not carry out a color analysis. In this paper, we report how the brightness and color of the night sky changed following a major retrofit of the street lighting in the urban area of Madrid, Spain. Madrid is well-suited for such research, as it has been a focus for multiple previous studies of night sky brightness and light pollution [9,11,15–23]. Here, we used a professional astronomical device (All-Sky Transmission Monitoring Camera; AstMon) that provides information in the Johnson B, V, and R photometric bands along with their associated color, and was operated for a long period (a decade). Finally, we provide a framework that could be used for a collective analysis with data from the International Space Station (ISS) and the Spectrometer for Aerosol Night Detection (SAND) spectrometer to better understand the general problem of the spatial reach of artificial light scattering from cities [10].

2. Materials and Methods 2.1. Background Madrid is a densely populated urban center with approximately 3.3 million inhabitants and with about 5.4 million inhabitants within a 27 km radius from downtown [23]. The entire Madrid region includes 6.6 million inhabitants. It has four other province capitals (Avila, Segovia, Toledo, and Guadalajara), which act as satellite cities. The entire Madrid region retrofitted their streetlights. Therefore, the skyglow changes in the Madrid region cannot be explained by the skyglow changes within the city of Madrid alone, although the city is the primary source of light pollution. It is relatively straightforward to determine the impact of a streetlight retrofit on night sky brightness in a small, isolated town that changes its public street illumination over the course of, say, a week. Using monitoring photometers before, during, and after the streetlight retrofit, one can easily determine the resultant change in brightness. Large cities, however, present a more complicated case, as retrofits have inevitably to be performed in phases. Madrid has been upgrading its street lighting system from HPS to LED technology since 2011, and this process is ongoing. Like many other cities, it has also been implementing other street lighting changes (e.g., changing part of its street lighting system from high power HPS to lower power HPS, testing and implementing regulations to reduce the intensity of street lighting by some percentage after midnight). Remote Sens. 2021, 13, x FOR PEER REVIEW 3 of 26

Remote Sens. 2021, 13, 1511 implementing other street lighting changes (e.g., changing part of its street lighting system3 of 25 from high power HPS to lower power HPS, testing and implementing regulations to reduce the intensity of street lighting by some percentage after midnight). We will focus on events before, during, and after the major streetlight retrofit that occurredWe willin Madrid focus onin 2015. events This before, year during,was notable and afterin the the extent major of streetlightthe upgrade. retroﬁt Photo- that graphsoccurred taken in Madrid by astronauts in 2015. aboard This year the was International notable in theSpace extent Station of the (ISS) upgrade. before Photographs and after thetaken main by changes astronauts in street aboard illumination the International show Spacevisible Station variation (ISS) in beforecolor from and after2011, the when main streetchanges lighting in street in Madrid illumination was dominated show visible by HPS variation lamps, to in 2017, color when from street 2011, lighting when street in somelighting areas in changed Madrid wasto primarily dominated white by LEDs HPS lamps,, following to 2017, the 2015 when retrofit street (Figure lighting 1). in For some example,areas changed Rivas Vaciamadrid, to primarily white at the LEDs, lower followingright of Figure the 2015 1, completely retroﬁt (Figure changed1). For from example, HPS toRivas LED. Vaciamadrid, at the lower right of Figure1, completely changed from HPS to LED.

FigureFigure 1. Images 1. Images of Madrid of Madrid at night at night taken taken from from International International Space Space Station Station (ISS) (ISS in) 2011 in 2011 (left ()left and) and 2017 2017 (right (right). The). The area area marked in whitemarked on the 2011in white image on isthe reproduced 2011 image withis reproduced greater detail with ingreater the 2017 detail image. in the The 2017 red image. circle The marks red thecircle position marks ofthe the position Astronomical of the Astronomical Observatory of Universidad Complutense de Madrid. Images from NASA Gateway to Astronaut Observatory of Universidad Complutense de Madrid. Images from NASA Gateway to Astronaut Photography of Earth [24] https: Photography of Earth [24] https://citiesatnight.org/ (accessed on 9 April 2021). //citiesatnight.org/ (accessed on 9 April 2021).

ChangesChanges to to both, both, manufacturing manufacturing technology technology and and lamp lamp power power (Figure (Figure 2)2 occurred) occurred be- be- forefore and and after after the the major major streetlight streetlight retrofit retroﬁt in in 2015. 2015. Changes Changes included included replacing replacing HPS HPS lamps lamps bulbsbulbs with with lower lower power power LED LED bulbs bulbs (from (from 250W/150W 250 W/150 to W 150W/70W). to 150 W/70 Additionally, W). Additionally, a por- a tionportion of 150W of 150 HPS W globe HPS lamps globe were lamps changed were changed to 56W demi to 56-globe W demi-globe LED lamps. LED The lamps. diffusers The werediffusers different were between different these between demi-globe these lamps, demi-globe the 150W lamps, had thea 0% 150 Upward W had Light a 0% Output Upward RatioLight (ULOR), Output and Ratio the (ULOR), 56W LED and lamps the 56W had LEDa 5% lampsULOR. had The aISS 5% image ULOR. of The2017 ISS (Figure image 1, of right2017 panel) (Figure shows1, right an panel)apparent shows decrease an apparent in street decreasebrightness in due street to the brightness change in due power to the ofchange HPS lamps in power (Right of HPSpanel, lamps Figure (Right 2). It panel, should Figure be noted2). It that should there be was noted not that a complete there was changenot a complete from HPS change to LED from lamps. HPS The to LED streetlight lamps. retrofits The streetlight were ma retroﬁtsde in steps were throughout made in steps thethroughout year, and thethe year,fraction and of the technology fraction ofchange technology was inconsistent change was from inconsistent year to year from (Figure year to 2).year According (Figure2 ).to Accordingthe Madrid to C theity MadridHall, there City is Hall, a reduction there is in areduction light power in lightafter powermidnight after whenmidnight HPS whenstreet HPSlights street are dimmed lights are to dimmed80% power to 80%and L powerEDs to and 60%. LEDs to 60%.

RemoteRemote Sens. Sens. 20212021, ,1313, ,x 1511FOR PEER REVIEW 4 4of of 26 25 Remote Sens. 2021, 13, x FOR PEER REVIEW 4 of 26

FigureFigure 2. 2. Changes Changes in in streetlighting streetlighting in in the the City City of of Madrid, Madrid, in in terms terms ofof the total numbernumber ofofof street streetstreet lamps lampswithlamps differentwith with different different technologies technologies technologies (left; all ( left(left districts; ;all all districts districts included), included),included), and High-Pressure andand High-Pressure Sodium Sodium (HPS) street ((HPSHPS lights’)) streetmeanstreet lights powerlights’ mean’ mean (right power ).power Open ( right(right Source:).). Open Open Madrid Source: Source: City MadridMadrid Hall. http://www-2.munimadrid.es/CSE6/control/ CityCity HallHall.. http://www-2.munimadrid.es/CSE6/control/seleccionDatos?numSerie=08010200033seleccionDatos?numSerie=08010200033madrid.es/CSE6/control/seleccionDatos?numSerie=08010200033) (accessed on 9 April 2021).)) ((accessed on 9 April 2021))..

2.2.2.2. Available Available Data Data TheThe data data used used for for this this work work werewere were obtainedobtained obtained viavia via observationsobservations observations from from fixed ﬁxed monitoring monitoring instrumentationinstrumentation on on top top of of the the Physics Physics building building of of Universidad Universidad Complutense Complutense de de Madrid Madrid instrumentation on top of the Physics building of◦ Universidad0 00 ◦ Complutense0 00 de Madrid (UCM(UCM(UCM Observatorio Observatorio UCM, UCM, IAU IAU IAU-MPC--MPCMPC code code I86, I86, 4040 40°°272727′′040404″″NNN 03 03°4343′3434″WW)) in in Ciudad Ciudad Univer-Univer- Univer- sitariasitaria ( red( (redred circle circle circle in in in Figure Figure Figure 1), 1),1), northwest northwest northwest of of of MadridMadrid Madrid’s’’ss citycity city centercenter center (Figure (Figure 3).3 The). The locationlocation location ofof of this observatory within a metropolitan area makes it possible to study the high level thisthis observatory observatory within within a a metropolitan metropolitan areaarea makesmakes itit possiblepossible to study the highhigh levellevel ofof of light pollution of the surrounding area, thus, making it a unique place dedicated to lightlight pollution pollution of of the the surrounding surrounding area, area, thus thus,, makingmaking itit aa unique place dedicatedd toto mon-mon- monitoring night sky brightness with ﬁrst-class instrumentation. itoringitoring night night sky sky brightness brightness with with first first--classclass instrumentation.instrumentation.

Figure 3. Universidad Complutense de Madrid ( (UCM)UCM) Observatory with with labeled labeled devices devices to to moni- monitor Figure 3. Universidad Complutense de Madrid (UCM) Observatory with labeled devices to moni- lighttor light pollution pollution near near the the astronomical astronomical observatory observatory domes. domes. tor light pollution near the astronomical observatory domes.

Remote Sens. 2021, 13, x FOR PEER REVIEW 5 of 26

Remote Sens. 2021, 13, 1511 5 of 25

The UCM astronomical observatory uses two Sky Quality Meters (SQM-LE and SQM-LU), and more recently, two TESS-W photometers, continuously to monitor the nightThe sky UCM brightness astronomical of Madrid. observatory The SQM uses-LE twoconnects Sky Quality to a computer Meters (SQM-LEwith a Local and Area SQM- LU),Network and,more LAN recently,wire, while two the TESS-W SQM- photometers,LU connects with continuously a USB connection. to monitor TESS the night-W is skya brightnessphotometer of equipped Madrid. with The SQM-LEa dichroic connects filter that to aallows computer an extension with a Local of the Area spectral Network, re- LANsponse wire, to the while red theregion SQM-LU of the connectsspectrum. with Thus, a USB the TESS connection.-W photometer TESS-W is is better a photometer suited equippedthan the SQM with for a dichroic monitoring ﬁlter polluted that allows night an sky extension brightness of the variation. spectral responseWhile these to the pho- red regiontometers of have the spectrum. a broad single Thus, photometric the TESS-W band photometer (panchromatic is better detector), suited than the the UCM SQM All for- monitoringSky Transmission polluted Monitor night sky (AstMon) brightness provides variation. all-sky While maps these in photometersthe Johnson B, have V, and a broad R singleastronomical photometric photometric band (panchromatic bands [25]. The detector), UCM astronomicalthe UCM All-Sky observatory Transmission also hosts Monitor a (AstMon)Spectrometer provides for Aeroso all-skyl Night maps Detection in the Johnson (SAND) B, for V, monitoring and R astronomical the night sky photometric spectra bands[21]. [25]. The UCM astronomical observatory also hosts a Spectrometer for Aerosol NightFigure Detection 4 demonstrates (SAND) for the monitoring detection therange night of skyour spectrainstrumentation. [21]. It shows a com- parisonFigure of the4 demonstrates mean night sky the detectionspectra after range midnight of our instrumentation.in January 2014 (before It shows the a retrofit) compari- sonand offor the the mean same nightmonth sky in spectra2019 (after after the midnight retrofit). inIn Januarythe top panel, 2014 (before we plot the the retrofit) detection and forrange the available same month by TESS in 2019-W (note (after that the retrofit).the spectral In the response top panel, extends we plotto the the red detection region) rangeand availableSQM, along by TESS-Wwith the (notespectra that obtained the spectral by the response SAND extendsspectrometer. to the In red the region) bottom and panel, SQM, alongwe present with thethe spectraJohnson obtained B, V, and by R the photometric SAND spectrometer. bands, which In the are bottom within panel,AstMon’s we presentspec- thetral Johnsonrange. For B, V,reference, and R photometric we include bands,the prominent which are emission within AstMon’slines of typical spectral lamps range. (mid- For reference,dle panel). we include the prominent emission lines of typical lamps (middle panel).

Figure 4. SpSpectrometerectrometer for for Aerosol Aerosol Night Night Detection Detection (SAND (SAND)) representative representative night night sky skyspectra spectra for for Madrid after midnight in January of 2014 and 2019 2019 (blue (blue and and red red solid solid lines; lines; upper upper panel). panel). The The 2019 spectrum2019 spectrum shows shows the blue the blue component component of the of the LED LED spectrum spectrum that that peaks peaks at 450at 450 nm. nm. The The upper upper panel alsopanel shows also shows the TESS-W the TESS and-W Sky and Quality Sky Quality Meter Meter (SQM) (SQM spectral) spectral responses. responses. Some Some prominent prominent emission emission lines of typical lamps are labeled in the middle panel. The Johnson B, V, and R photomet- lines of typical lamps are labeled in the middle panel. The Johnson B, V, and R photometric bands are ric bands are plotted in the lower panel. plotted in the lower panel.

Data collection has been ongoing since the installation of each device (Table1;[ 18–20]). We have compiled long-term temporal time series with new high-quality data for SQM-LE and SQM-LU (available since 2012), TESS (since 2016), AstMon (since 2010), SAND (since

Remote Sens. 2021, 13, 1511 6 of 25

2013), and data from Aerosol Optical Depth (AOD) from AERONET (since 2012). To our knowledge, there is currently no other such systematic information accounting for the evolution of night sky brightness in Johnson B, V, and R astronomical photometric bands and its associated color for a city with a large population. A detailed analysis of the evolution of the spectra provided by SAND and the AOD data collected by AERONET is beyond the scope of this paper and will be presented elsewhere.

Table 1. Instruments used in this work.

Instrument Starting AstMon all-sky astronomical camera B V R images 2010/08 Johnson B, V, R Every 5 min Sky Quality Meter (SQM) 2012/10 SQM-LE serial number 1952 One measurement per minute 2013/10 SQM-LE serial number 2218 One spectrum SAND spectrograph 2013/07 Every 10 min STARS4ALL TESS-W photometer 2016/07 stars5 One measurement per minute 2017/11 stars85 (with blue-blocking ﬁlter)

2.3. Data Management and Pre-Processing The data from the SQMs were recorded with the freeware software PySQM [26], while TESS-W photometers use the Internet of Things (IoT) protocols to insert the data in real- time via STARS4ALL repositories. Both photometers point to the zenith. Their data were archived using the standard International Dark-Sky Association (IDA) format [13]. Night sky brightness data from the three photometers were filtered to remove values that were too bright, too dark, or otherwise unreliable, leaving values in the range 14.0 < m < 19.0 mag/arcsec2. The resulting files contain records with the date–time and sky brightness in their respective bands for each measure, typically one value per minute. AstMon proprietary software provides calibrated all-sky images along with sky brightness values for several locations in the sky. In this work, we used sky brightness at the zenith (3.8 arcminutes pixel scale) in order to compare with the SQM (Field of View (FoV) = 20◦) and TESS-W (FoV = 17◦) photometer data. The observational procedure of AstMon at UCM consists of a sequence of Johnson B, V, and R measurements (in that order) followed by a small interval, after which the sequence repeats. AstMon at UCM only observes during the astronomical night. AstMon data are stored in simple ASCII files whose records contain date and time of observation (Julian dates) and zenith sky brightness in the corresponding B, V, and R bands. Color indices (differences in magnitudes from two photometric bands) were derived later to determine changes in the color of the sky. In order to determine B-V and V-R color indices, the data have been grouped according to the observation time. The time of the V observation has been assigned to every B, V, and R sequence. An ASCII file containing date–time and color index data was created for the whole series.

2.4. Filtering Process The sky brightness at the zenith measured at the UCM astronomical observatory is a proxy for light pollution in the Madrid urban area. The relative contribution of different light sources depends on their intensity and distance from the photometers. Our purpose is to determine the changes in the number, intensity, and color of pollution sources using night sky brightness values. There are other factors that increase the brightness of the sky, particularly the presence of the moon over the horizon and variation in atmospheric conditions, particularly cloud cover and aerosol content. In this paper, we address the long-term effects of night sky brightness, produced by changes in street lighting systems. Effects produced by factors such as aerosols and atmospheric conditions are often short-term Remote Sens. 2021, 13, 1511 7 of 25

and can be interrelated (such as wind diluting aerosol content) and will be studied in a future paper. One approach to avoid the influence of clouds would be to select only the photometric nights in the astronomical sense: transparent nights with good conditions during the whole night. This procedure drastically reduces the dataset and throws away a large portion of otherwise useful data, including observations with clear skies during some intervals and non-photometric nights. Another approach is to use the complete dataset and perform an automatic filtering process to eliminate observations taken with cloud cover over the observatory. Since the monitoring station is located inside a highly polluted city, any cloud will increase the brightness. It would be straightforward to write code to filter out unexpected bright values by comparing them with previous and posterior observations. In our case, we elected to filter the data by visual inspection. This procedure is lengthy and tedious but yielded a filtered dataset that we could better trust. Our method offers the opportunity to visualize all the data, observe the variation in sky brightness throughout the night for different sky conditions, and learn about the effects of clouds and the presence of the moon on sky brightness. Each record of night sky brightness data collected by AstMon and the photometers was categorized into three conditions: clear sky without the moon (no moon, no clouds), clouds during the observation, and data obtained with the moon over the horizon. The procedure is as follows: (a) For each night, a graph is displayed showing the Johnson B, V, and R photometric data throughout the night (x-axis). The magnitude of the broadband provided by the photometers is displayed as a comparison. A second panel of the graph shows the variation of the B-V and V-R color indices. The Julian date when the moon is over the horizon is calculated and the corresponding records flagged. The data with the moon over the horizon are displayed with a different marker. (b) After visual inspection, the data suspected to include clouds were flagged in the dataset. (c) The data are displayed again to ensure that the process was done properly. After the process, all the records of the dataset (both magnitudes and colors) were flagged, and we performed statistics on the observations that were free of external influences (no moon, no clouds). Figure5 shows a typical graph of the variation throughout one night with observations during a cloudy sky and the presence of moonlight labeled. Figure6 shows the graph for a photometric night without the moon over the horizon. During the first part of the astronomical night (after twilight), the sky brightness decreases due to the progressive reduction in human activity, including traffic, occupation of office buildings and shops being open, among other anthropogenic factors. A step around 23 h is due to the switching-off of ornamental lights. After midnight, most of the light sources are those from public street illumination. According to Madrid city hall, the intensity of a fraction of the street lamps is reduced in the late evening, but the number has been inconsistent over time. The brightening from 04:45 h to 05:30 h is the result of street illumination returning to normal operational intensity (increase in electric power) before citizens leave for the daily commute. Finally, the end of the plot shows the twilight as recorded by SQM-LE/SQM-LU. The lower panel shows the B-V and V-R color indices. Lower values represent bluer skies. The sky is bluer at the beginning of the night due to the types of lights used in traffic, ornamentation, sports stadiums, and other light pollution sources, which are generally bluer. After midnight, the color reflects street illumination showing low night sky brightness variability. The observations during this part of the night are better suited to determine the nature of the lamps used for street illumination. After the filtering process, we used 119,296 measurements (no moon, no clouds), accounting for 28% of AstMon total observations. As regards the photometer data, we used 283,922 measurements, accounting for 41% of the SQM measurements, and 234,543 measurements, accounting for 39% of the TESS-W measurements. The percentage of useful data for this research (no moon, no clouds) depends on the experiment starting date, device operability, and the hourly instrument sampling of night sky brightness (Table1). Remote Sens. 2021,, 13,, x 1511x FORFOR PEERPEER REVIEWREVIEW 88 of of 26 25

FigureFigure 5. 5. BrightnessBrightness and color evolution ofof thethe skysky duringduring a a typical typical night night in in Madrid. Madrid. Blue, Blue, green, green, and andred solidred solid points points represent represent the Johnson the Johnson B, V, B, and V, R and data R withdata nowith moon no moon and no and clouds, no clouds, as measured as meas- by uredAstMon. by AstMon. Observations Observations with the with moon the over moon the over horizon the horizon are marked are marked with the with same the colorsame codecolor and codeopen and circles. open Observations circles. Observations under cloudy under skies cloudy (brighter skies and(brighter variable and values) variable are values) marked are with marked small withpoints. small Note points. how theNote color how B-V the andcolor V-R B--V indices and V (bottom--R indices graph) (bottom show graph) distinctive show distinctive values during values the duringcloud episodes. the cloud SQMepisodes. data SQM are displayed data are displayed as black points. as black Astronomical points. Astronomical magnitude magnitude is a logarithmic is a logarithmic inverse scale and mag/arcsec22 values are lower for brighter skies. logarithmicinverse scale inverse and mag/arcsec scale and 2mag/arcsecvalues are lowervalues for are brighter lower for skies. brighter skies.

FigureFigure 6. 6. BrightnessBrightness a andnd color color evolution evolution of of the the sky sky during during a photometric a photometric night night in inMadrid. Madrid. While While thethethe brightnessbrightness brightness decreasesdecreases decreases monotonicallymonotonically monotonically duringduring during thethe the firstfirst ﬁrst partpart part ofof ofthethe the night,night, night, thethe the curvecurve curve isis almostalmost is almost flatflat ﬂat afterafter midnight. midnight.

Remote Sens. 2021, 13, x FOR PEER REVIEW 9 of 26

After the filtering process, we used 119,296 measurements (no moon, no clouds), accounting for 28% of AstMon total observations. As regards the photometer data, we used 283,922 measurements, accounting for 41% of the SQM measurements, and 234,543 measurements, accounting for 39% of the TESS-W measurements. The percentage of useful data for this research (no moon, no clouds) depends on the experiment starting date, de- Remote Sens. 2021, 13, 1511 vice operability, and the hourly instrument sampling of night sky brightness (Table 1).9 of 25

3. Results 3.1. Madrid Night Sky Spectra 3. Results In Figure 4, we present a graph comparing the mean night sky spectra after midnight 3.1. Madrid Night Sky Spectra in January 2014 and January 2019. Examining this figure, we see that the 2019 night spectrum Inshows Figure the4, weblue present bump aof graph the LED comparing spectrum. the meanThe Johnson night sky B, spectraV, and afterR photometric midnight bandin January responses 2014 show and Januaryhow the 2019.changes Examining in spectra this are ﬁgure,related we to the see change that the in 2019 the nightsky’s spectrum shows the blue bump of the LED spectrum. The Johnson B, V, and R photometric color. band responses show how the changes in spectra are related to the change in the sky’s color. 3.2.3.2. AstMon AstMon InIn Figure Figure 7,7, we we show show histograms histograms of of the the Johnson Johnson B, B,V, V,and and R night R night sky skybrightness brightness and colorandcolor indices indices at the at zenith the zenith measured measured using using AstMon. AstMon. A progressive A progressive brightening brightening of the of Ma- the dridMadrid sky skyis detected is detected in the in the Johnson Johnson B photometric B photometric band. band. The The Johnson Johnson B, B, V, V, and and R R band band brightnessbrightness values values for for nights nights with with both both moon moon and and clouds clouds (lighter (lighter bars) bars) were were found found to be be to to thethe left left of of values values for for moo moonlessnless and and cloudless cloudless nights nights (darker (darker bars), bars), which which is is expected expected since since cloudsclouds brighten brighten the the night night sky. sky. The The B B-V-V color color index index value value increases increases up up to B B-V-V = 2.0 2.0 with cloudsclouds (more (more yellow, yellow, or or less less blue, blue, sky). sky). This This results results in in a a sky sky that that is is much much redder redder than than the the naturalnatural on one,e, which which may may have have a a relevant relevant impact impact on on the biological environment.

Figure 7. AstMon. From left to right, histograms for the Johnson B, V, R, B-V, and V-R color indices Figure 7. AstMon. From left to right, histograms for the Johnson B, V, R, B-V, and V-R color indi- cesfor for the the whole whole dataset, dataset, grouped grouped by year.by year. Darker Darker bars bars are are for for ﬁltered filtered data data (no (no moon moon over over the the horizon horizonand no clouds)and no clouds) and lighter and barslighter for bars unﬁltered for unfiltered data (with data moon (with and moon clouds). and clouds).

ComparingComparing Figures Figures 88 andand9 9 shows shows that, that, before before midnight, midnight, the the sky sky is is brighter brighter and and bluer bluer thanthan after after midnight whenwhen itit is is dimmer dimmer and and more more yellow. yellow. The The data data were, were therefore,, therefor splite, split into intobefore before and after and midnight after midnight to remove to remove the ornamental the ornamental light component, light component, headlights headlights from cars, and light from buildings that contribute to brighter and bluer skies before midnight. This process helped isolate the contribution of the street lighting component. In these ﬁgures, we present histograms that only include ﬁltered data. The range of values is reduced, and

the changes can be seen more clearly. We used a non-parametric Gaussian Kernel Density Estimation (KDE) to derive a representative value for each histogram (mode) and the underlying distribution. The KDE uses an optimization via the Scott rule of thumb, a data- based procedure for choosing an optimal bandwidth [27]. The bandwidth of the Gaussian kernel is chosen to minimize the magnitude mean integrated square error (goodness-of-ﬁt). There are several other options for selecting a Kernel, but we used the Gaussian reference Remote Sens. 2021, 13, x FOR PEER REVIEW 10 of 26

from cars, and light from buildings that contribute to brighter and bluer skies before midnight. This process helped isolate the contribution of the street lighting component. In these figures, we present histograms that only include filtered data. The range of values is reduced, and the changes can be seen more clearly. We used a non-parametric Gaussian Kernel Density Estimation (KDE) to derive a representative value for each histogram (mode) and the underlying distribution. The KDE uses an optimization via the Scott rule of thumb, a data-based procedure for choosing an optimal bandwidth [27]. The bandwidth of the Gaussian kernel is chosen to minimize the magnitude mean integrated square error (goodness-of-fit). There are several other options for selecting a Kernel, but we used Remote Sens. 2021, 13, 1511 the Gaussian reference because of the comparison with the optimal bin widths and differ-10 of 25 ent densities (within the kernel). The densities can then be rescaled into a standard form, which facilitates the comparison between the optimal bin and the different densities yearlybecause by of Johnson the comparison B, V, and withR photometric the optimal bands, bin widthsand associated and different color indices. densities (within the kernel).It should The be densitiesnoted that can the then Johnson be rescaled B, V, and into R photometric a standard form, bands which used facilitatesa logarithmic the 2 unitcomparison of mag/arcsec between, which the optimal is related bin inversely and the to different the radiance. densities The yearlyresultsby may Johnson differ after B, V, conversionand R photometric to linear bands, units. and associated color indices.

FigureFigure 8. 8. AstMon.AstMon. Night Night sky brightness and colorcolor indexindex histogramshistograms for for ﬁltered filtered data data before before midnight. midnight.From From left to left right, to right, histograms histograms for the for Johnsonthe Johnson B, V, B, R, V, B-V, R, B and-V, and V-R V color-R color indices. indices. Green Green points Remote Sens. 2021, 13, x FOR PEER REVIEW 11 of 26 pointsrepresent represent the modes. the modes. The black The curvesblack curves represent represent the Kernel the Kernel Density Density Estimation Estimation (KDE) (KDE that) can that be caninterpreted be interpreted as the goodness-of-ﬁt.as the goodness-of-fit.

FigureFigure 9. 9. AstMon.AstMon. Night Night sky sky brightness brightness and and color color index index histograms histograms for for filtered ﬁltered data after mid- midnight. night.From leftFrom to left right, to histogramsright, histograms for the for Johnson the Johns B, V,on R, B, B-V, V, R, and B-V, V-R and color V-R indices. color indices.

The evolution of night sky brightness and color after midnight with reduced ornamental lights, headlights, and light from buildings in Figure 9 shows a progressive evolution toward blue after 2015, when a significant streetlight retrofit replaced HPS lamps with LEDs. The light in the blue band brightened in comparison with less variable changes in the V band and a compensated change in the R band. It is interesting to note that the B-V color index highlights this bluing of the night sky (lower values of the B-V index). The change is also clear in the V-R color index. After midnight, the Johnson B photometric band brightened following the street light retrofit, seen in the movement of the peaks to the left (Figure 9), while the Johnson R photometric band dimmed, as seen in the movement of the peaks to the right. The Johnson V photometric band was less sensitive to the streetlight retrofit. The B-V and V-R color indices showed changes from reddish to bluish, allowing us to identify HPS, streetlight retrofit, and LED data clusters (Figure 10). For yearly night sky brightness and color statistics, see Tables S1–S5 of the Supplementary Materials.

Remote Sens. 2021, 13, 1511 11 of 25

It should be noted that the Johnson B, V, and R photometric bands used a logarithmic unit of mag/arcsec2, which is related inversely to the radiance. The results may differ after conversion to linear units. The evolution of night sky brightness and color after midnight with reduced ornamental lights, headlights, and light from buildings in Figure9 shows a progressive evolution toward blue after 2015, when a significant streetlight retrofit replaced HPS lamps with LEDs. The light in the blue band brightened in comparison with less variable changes in the V band and a compensated change in the R band. It is interesting to note that the B-V color index highlights this bluing of the night sky (lower values of the B-V index). The change is also clear in the V-R color index. After midnight, the Johnson B photometric band brightened following the street light retrofit, seen in the movement of the peaks to the left (Figure9), while the Johnson R photometric band dimmed, as seen in the movement of the peaks to the right. The Johnson V photometric band was less sensitive to the streetlight retrofit. The B-V and V-R color Remote Sens. 2021, 13, x FOR PEER REVIEWindices showed changes from reddish to bluish, allowing us to identify HPS, streetlight12 of 26

retroﬁt, and LED data clusters (Figure 10). For yearly night sky brightness and color statistics, see Tables S1–S5 of the Supplementary Materials.

FigureFigure 10.10.Color–color Color–color diagram diagram for for observations observations before before (left (left) and) and after after midnight midnig (rightht (right). The). whiteThe white dots represent dots represent the yearly the medianyearly median values, values, while the while change the change in time in is representedtime is represented by color. by color.

We visualizedvisualized thesethese changeschanges usingusing aa color–colorcolor–color diagram,diagram, wherewhere thethe observations areare color-codedcolor-coded andand binnedbinned byby observationobservation datedate (Figure(Figure 1010).). TheThe twotwo panelspanels inin FigureFigure 1010 showshow how the change of technology transformed the night sky color in the expected direction: bluer skies as thethe LEDsLEDs areare beingbeing introduced.introduced. TheseThese diagramsdiagrams showshow thethe HPS,HPS, streetstreet lightlight retroﬁt,retrofit, andand LEDLED datadata clustersclusters andand thethe shiftshift fromfrom reddishreddish inin thethe HPSHPS regionregion toto bluishbluish inin thethe LEDLED region.region. Splitting the datadata beforebefore andand afterafter midnightmidnight allowedallowed us,us, again,again, toto betterbetter account for changes inin thethe colorcolor indices.indices. 3.3. Sky Quality Meter SQM 3.3. Sky Quality Meter SQM The SQM photometer registered brighter values before the streetlight retroﬁt in 2015 The SQM photometer registered brighter values before the streetlight retrofit in 2015 (Figure 11); the values were similar to AstMon’s Johnson V band measurement in the (Figure 11); the values were similar to AstMon’s Johnson V band measurement in the same period (City Hall decreased bulb power of the HPS lamps (150W to 50W)). After the streetlight retrofit during 2015–2016, SQM registered dimmer values, while Johnson V registered brighter values of night sky brightness. These differences between the Johnson V and SQM measurements are later explained in terms of color indices correction applied to the SQM photometer (Section 3.5).

Remote Sens. 2021, 13, 1511 12 of 25

same period (City Hall decreased bulb power of the HPS lamps (150 W to 50 W)). After the streetlight retroﬁt during 2015–2016, SQM registered dimmer values, while Johnson V registered brighter values of night sky brightness. These differences between the Johnson Remote Sens. 2021, 13, x FOR PEER REVIEW 13 of 26 V and SQM measurements are later explained in terms of color indices correction applied to the SQM photometer (Section 3.5).

FigureFigure 11. 11. EvolutionEvolution of of SQM SQM magnitudes magnitudes by by grouping grouping the the data data into into years. years. Darker Darker bars bars represent represent filteredfiltered data data (no (no moon moon over over the the horizon horizon and and no no clouds), clouds), and and lighter lighter bars bars are are unfiltered unfiltered data data (with (with moonmoon and and clouds) clouds).. As As expected, expected, the the unfiltered unfiltered data data were were distributed distributed to to the the left left of of the the filtered filtered data, data, indicatingindicating brighter brighter skies skies on on nights nights with with moon moon and/or and/or clouds clouds (similar (similar to Figure to Figure 7). 7All). Allmeasures measures excluded twilight. excluded twilight.

TheThe SQM datadata recordedrecorded a a brighter brighter night night sky sky before before midnight midnight and and dimmer dimmer values values after aftermidnight midnight (Figure (Figure 12). As 12). with As AstMon,with AstMon, the procedure the procedure of splitting of splitting the data the in thisdata manner in this mannerguarantees guarantees a reduction a reduction in the ornamental in the ornamental lights and lights other and contributors other contributors to light pollution to light pollutionduring the during ﬁrst part the offirst the part night of the (See night Table (See S6 ofTable the SupplementaryS6 of the Supplementary Materials). Materials The year). The2015 year registered 2015 registered the highest the magnitudehighest magnitude (lower radiance)(lower radiance) of the series,of the whileseries, numericalwhile nu- mericaltrends of trends the median of the median were variable. were variable.

Remote Sens. 2021, 13, x FOR PEER REVIEW 14 of 26

Remote Sens. 2021, 13, 1511x FOR PEER REVIEW 1413 of 26 25

Figure 12. SQM histograms for filtered data showing night sky brightness values before (left) and after midnight (right). TheFigure green 12. pointsSQM histograms represent the for mode. filtered ﬁltered data showing night sky brightness values before (left) (left) and and after after midnight midnight (right). (right). The green points represent the mode. 3.4. TESS-W 3.4.3.4. TESS TESS-WFor Madrid-W night skies, the TESS-W registered brighter values compared to the SQM. ConsistentForFor Madrid Madrid with AstMonnight night skies, skies, and the the SQM, TESS TESS-W the-W data registered registered with clouds brighter brighter showed values values a compared compared bimodal distributionto to the the SQM. SQM. (FigureConsistentConsistent 13), with withwhile AstMon AstMon the data and and with SQM, no moonthe data and with no clouds clouds showed showed a a left bimodal-skewed distribution distribution.(Figure(Figure 1313),), whilewhile thethe datadata with with no no moon moon and and no no clouds clouds showed showed a left-skewed a left-skewed distribution. distribution.

FigureFigure 13. 13. TESSTESS-W-W magnitude magnitude by by year. year. Darker Darker bars bars represent represent the the result result for for filtered ﬁltered data data (no (no moon moon overFigureover t thehe 13. horizon horizon TESS- Wand and magnitude no no clouds) clouds) by and andyear. lighter lighter Darker bars bars bars for for representunfiltered unﬁltered the data data result (with (with for moon filtered moon and anddata clouds). clouds). (no moon All All measuresovermeasures the horizon exclude exclude and twilight. no clouds) and lighter bars for unfiltered data (with moon and clouds). All measures exclude twilight.

Remote Sens. 2021 13 Remote Sens. 2021,, 13,, 1511x FOR PEER REVIEW 1514 of of26 25

TESSTESS-W-W reported reported brighter brighter night night sky sky brightness brightness values values before before midnight midnight and and dimmer dimmer valuesvalues after after midnight midnight (Figure (Figure 13). 13). As As with with the the SQM, SQM, nothin nothingg can can be be said said about about the the evolu- evolu- tiontion of of color color indices. indices. Table Table S7 S7 shows shows a a decreasing decreasing trend trend for for the the median median magnitude magnitude values values (brighter(brighter skies) skies) for for data data after after midnight. midnight. This This result result is is expected expected because because of of the the removal removal of of HPSHPS streetlights (which projectproject inin thethered red spectrum), spectrum), changes changes in in streetlight streetlight bulb bulb power, power, and andchanges changes in the in diffuserthe diffuser technology. technology. TESS-W TESS is-W better is better suited suited for detecting for detecting night night sky bright- sky brightnessness variation variation due to due streetlight to streetlight retroﬁt retrofit because because it has it anhas extended an extended capacity capacity to measure to meas- in urered in spectra. red spectra. The progressive The progressi brighteningve brightening of the of sky the detected sky detected by TESS-W by TESS (Figure-W (Figure 14) is14) not isapparent not apparent in the in statistics the statistics of SQM of SQM (Figure (Figure 13). 13).

FigureFigure 14. 14. HistogramsHistograms of of night night sky sky brightness brightness values values measured measured by byTESS TESS-W-W before before (left) (left and) andafter after (right)(right) midnight. midnight. The The green green points points represent represent th thee mode. mode.

3.5.3.5. SQM SQM and and Johnson Johnson Photometry Photometry Comparison Comparison SánchezSánchez de Miguel etet al.al. [ 11[11]] examined examined the the light light response response properties properties of aof synthetic a synthetic SQM SQMusing using the Johnson the Johnson V magnitude V magnitude and the and B-V the color B-V color index. index. The synthetic The synthetic formula formula requires re- a quiresconversion a conversion from the from AB magnitudethe AB magnitude system tosystem radiance, to radiance, allowing allowing a comparison a compar withison the withJohnson the VJohnson band. TheyV band. showed They thatshowed irrespective that irrespective of the lamps of the used lamps for street used illumination, for street illu- the mination,formula SQM the formulas = V + 0.5SQM *(B-V),s = V + provides 0.5 *(B-V), a color provides index a correction,color index (B-V)correction, < 2 (see (B-FigureV) < 2 8 (seeof that Figure paper). 8 of that We paper). plot in We Figure plot in15 Figure a graph 15 a with graph the with Johnson the Johnson bands bands and SQM and SQM yearly yearlymeasurements measurements for Madrid. for Madrid. We add We the add results the results for the for synthetic the synthetic SQM (SQM SQMs ).(SQM Tables).2 Table shows 2the shows differences the differences in magnitude in magnitude for SQM-SQM for SQMs.-SQM Figures. Figure 15 shows 15 shows that the that SQM the SQM photometer pho- tometerdid not did follow not thefollow Johnson the Johnson B and VB trends.and V tre Afternds. performingAfter performing the color the color index index correction, correction,the SQM thes trends, SQMs trends, these agreed these agreed with the with Johnson the Johnson ones. ones. This discrepancyThis discrepancy in trends in trends likely likelyindicates indicates a tendency a tendency of SQM of opticalSQM optical elements elements to malfunction to malfunction as a consequence as a consequence of the ﬁlterof thebecoming filter becoming less transparent less transparent over time. ov Theer TESS-Wtime. The photometer TESS-W photometer did not need did a correction,not need a as correction,it showed theas it expected showed brighteningthe expected of brightening the night sky, of the supporting night sky, the supporting experimental the experi- outcome mentalthat the outcome SQM presented that the SQM opposite presented trends opposite from the trends ones expected from the [ 28ones]. expected [28]. Table 2. Comparison SQM and synthetic SQM (mag/arcsec2).

Year SQMs SQM SQM-SQMs 2013 18.53 18.24 −0.29 2014 18.53 18.32 −0.21 2015 18.69 18.59 −0.10 2016 18.49 18.38 −0.11 2017 18.54 18.62 +0.08 2018 18.34 18.53 +0.19 2019 18.46 18.57 +0.11

Remote Sens. 2021, 13, 1511 15 of 25 Remote Sens. 2021, 13, x FOR PEER REVIEW 16 of 26

FigureFigure 15. Comparison of of the the evolutio evolutionn of of the the yearly yearly median median values values after after midnight midnight of ofthe the Johnson Johnson BB and V V magnitude magnitude and and the the SQM. SQM. The The brown brown points points represent represent the synthetic the synthetic SQM SQMs obtaineds obtained by by combiningcombining V and (B (B-V)-V) data.

3.6.Table Variation 2. Comparison throughout SQM theand Night synthetic SQM (mag/arcsec2).

FigureYear6 shows a typical graphSQMs of the variation inSQM sky brightness andSQM color-SQM throughouts the night2013 for a photometric night.18.53 The position and shape18.24 of these curves has−0.29 been changing in recent years. In Figure 16 we present graphs with the composition of the curves for 2014 18.53 18.32 −0.21 filtered data collected using AstMon in Johnson B separated into seasons. The graphs have 2015 18.69 18.59 −0.10 been color-coded with the number of observations that represent the average curve for 2016 18.49 18.38 −0.11 each season. Since winter has the most prolonged nighttime duration in Madrid (more 2017 18.54 18.62 +0.08 than twelve hours), we use this season for comparison purposes. There2018 are four regions of18.34 interest related to anthropogenic18.53 factors (Figure+0.19 17). After twilight,2019 the sky progressively18.46 darkens as human activity18.57 decreases from+0.11 18:30 to 20:30, as it is the beginning of the night. At 21:30, some ornamental lights are turned off. The second3.6. Variation part ofthroughout the night the (after Night midnight) shows an almost flat curve. As the ornamental lightsFigure are off 6 andshows the a traffic typical is graph low, it of is the bestvariation time in to sky measure brightness the streetlight and color component. through- Finally,out the afternight 4:30, for a the photometric streetlight powernight. isThe increased position for and the shape morning of these commute. curves To has show been the evolutionchanging inof nightrecent curves years. in In winter Figure from 16 we 2010 present to 2019, graphs we binned with the the composition data fromAstMon of the incurves hourly for intervalsfiltered data and collected the color using index AstMon B-V (Figure in Johnson 18). We B separated highlight (ininto insets) seasons. on The each graphgraphs in have Figure been 18 color the time-coded of 03:30with the UT, number which wasof observations optimal for that streetlight represent signal the detection.average Thecurve winter for each data season. from SQMSince andwinter TESS-W has the are most also prolonged dividedinto nighttime hourly duration bins and in presented Madrid below(more t (Figureshan twelve 19 hours),and 20). we We use are this reporting season for the comparison night sky brightnesspurposes. median values and interquartile ranges (IQR)/2sigma for non-normally distributed data. The IQR represents a robust measure for scaling the variability in magnitudes. There are some interesting changes related to the Madrid streetlight retrofit that we will comment on in the discussion.

Remote Sens. 2021, 13, x FOR PEER REVIEW 17 of 26

Remote Sens. 2021, 13, 1511 16 of 25 Remote Sens. 2021, 13, x FOR PEER REVIEW 17 of 26

Figure 16. Seasonal variation of sky brightness throughout the night in the Johnson B photometric band. The winter here is defined from November to January. Panels (A–D) are for the Winter, Spring (February to April), Summer (May to July), and Fall (August to October) from 2010 to 2020.

There are four regions of interest related to anthropogenic factors (Figure 17). After twilight, the sky progressively darkens as human activity decreases from 18:30 to 20:30, as it is the beginning of the night. At 21:30, some ornamental lights are turned off. The second part of the night (after midnight) shows an almost flat curve. As the ornamental lights are off and the traffic is low, it is the best time to measure the streetlight component. Finally, after 4:30, the streetlight power is increased for the morning commute. To show the evolution of night curves in winter from 2010 to 2019, we binned the data from AstMon in hourly intervals and the color index B-V (Figure 18). We highlight (in insets) on each graph in Figure 18 the time of 03:30 UT, which was optimal for streetlight signal detection. The winter data from SQM and TESS-W are also divided into hourly bins and presented below (Figures 19 and 20). We are reporting the night sky brightness median values and interquartile ranges (IQR)/2sigma for non-normally distributed data. The IQR rep- Figureresents 16. a Seasonalrobust measure variation variation forof of sky skyscaling brightness brightness the variability throughout throughout inthe the magnitudes. night night in inthe the Johnson There Johnson Bare photometric B photometricsome inter- band.esting The changes winter here related is defined deﬁned to the from Madrid November streetlight to January. retrofit Panels that ( Awe–D )will are forcomment the Winter, on Springin the (FebruarySpringdiscussion. (February to April), to April), Summer Summer (May to (May July), to andJuly), Fall and (August Fall (August to October) to October) from 2010from to 2010 2020. to 2020.

There are four regions of interest related to anthropogenic factors (Figure 17). After twilight, the sky progressively darkens as human activity decreases from 18:30 to 20:30, as it is the beginning of the night. At 21:30, some ornamental lights are turned off. The second part of the night (after midnight) shows an almost flat curve. As the ornamental lights are off and the traffic is low, it is the best time to measure the streetlight component. Finally, after 4:30, the streetlight power is increased for the morning commute. To show the evolution of night curves in winter from 2010 to 2019, we binned the data from AstMon in hourly intervals and the color index B-V (Figure 18). We highlight (in insets) on each graph in Figure 18 the time of 03:30 UT, which was optimal for streetlight signal detection. The winter data from SQM and TESS-W are also divided into hourly bins and presented below (Figures 19 and 20). We are reporting the night sky brightness median values and interquartile ranges (IQR)/2sigma for non-normally distributed data. The IQR represents a robust measure for scaling the variability in magnitudes. There are some interesting changes related to the Madrid streetlight retrofit that we will comment on in the discussion.

FigureFigure 17.17. Detail of the winter curvecurve withwith somesome anthropogenicanthropogenic factorsfactors annotated.annotated.

Figure 17. Detail of the winter curve with some anthropogenic factors annotated.

Remote Sens. 2021, 13, 1511 17 of 25 Remote Sens. 2021, 13, x FOR PEER REVIEW 18 of 26

Figure 18. ChangeFigure in 18. positionChange and in positionshape of and the shapecurves of showing the curves the showingevolution the of evolutionJohnson B, of V, Johnson R and B B,-V V, color R and index B-V color index throughout thethroughout night in thewinter. night Data in winter. are grouped Data areby groupedyear and bybinned year andin hourly binned intervals. in hourly The intervals. data points The have data pointsbeen have been displaced ondisplaced the x-axis on to theavoidx-axis crowding. to avoid Inset crowding. shows Inset detail shows for the detail 3 h for30 them evolution. 3 h 30 m evolution. The night The sky night brightness sky brightness varia- variability bility is depictedis depicted with the with Inter the Quartile Inter Quartile Range Range(IQR). (IQR).

Remote Sens. 2021, 13, x FOR PEER REVIEW 19 of 26 Remote Sens. 2021, 13, 1511 18 of 25

Remote Sens. 2021, 13, x FOR PEER REVIEW 20 of 26

The color index for each winter curve before and after midnight was smoother throughout the night sky from 2010 to 2019 (Bottom of Figure 20). Before midnight, there was no marked change in the B-V color (~0.8). After the streetlight retrofit in 2015, the color index after midnight changed considerably from 1.2 to 1.0, indicating bluer skies. Thus, the color difference throughout the night is now significantly lower. Finally, higher

Figure 19. Change in positionvariability and shape in of the the IQR curves was showing due to anthropogenic the evolution of SQMactivities magnitude before throughout midnight, theand night the bright- in Figure 19. Change in position and shape of the curves showing the evolution of SQM magnitude throughout the night in winter. Data are grouped byness year change and binned at the in hourlyend of intervals.the night when lights return to the nominal power. winter. Data are grouped by year and binned in hourly intervals.

We report in Tables 3–5 the binned hourly night sky brightness values (50th quartile) for Johnson B, V, and R along with the associated interquartile range for years before (2011), during (2015), and after (2019) the streetlight retrofit in 2015. We selected three times throughout the night for Tables 3–5: (a) Switching off ornamental lights (21:30), (b) Optimal street light detection range (from 01:30 to 04:30), and (c) Streetlight power increase (05:30). The values indicate that most sky brightness variability happens in the first part of the night (before midnight). On average, the night sky brightness variability before midnight is 45%, 27%, and 20% in the Johnson B, V, and R photometric bands, respectively. Before midnight, Johnson B represents a combination of ornamental and commer- cial lighting, along with the LED streetlights. The Johnson R captures the dimming of HPS streetlights. These findings support previously reported values measured by the ISS [29].

Table 3. Binned Winter Evolution in Johnson B (mag/arcsec2).

Time Median (2010 to 2019) 2011 2015 2019 21:30 18.63 18.61 ± 0.13 18.83 ± 0.09 18.55 ± 0.06 01:30 19.13 19.13 ± 0.07 19.30 ± 0.07 18.97 ± 0.06 Figure 20. Change in position and shape of the curves showing the evolution of TESS-W magnitude throughout the night Figure 20. Change in position and05:30 shape of the curves showing19.08 the evolution of19.10 TESS -±W 0.05 magnitude 19.21 throughout ± 0.18 18.90the night ± 0.27 in winter. Data were grouped by year and binned in hourly intervals. in winter. Data were grouped by year and binned in hourly intervals. TableWe 4. Binned report Winter in Tables Evolution3–5 the in binned Johnson hourly V (mag/arcsec night sky2). brightness values (50th quartile) 4. Discussion for JohnsonTime B, V, andMedian R along (2010 with to the 2019) associated interquartile2011 range2015 for years before2019 (2011), 4.1.during Night (2015), Sky Brightness and after Changes (2019) theover streetlightTime retrofit in 2015. We selected three times 21:30 17.73 17.66 ± 0.09 17.81 ± 0.09 17.71 ± 0.05 throughoutWe have the presented night for Tablesa long3-–term5: (a) time Switching series offof observations ornamental lights resulting (21:30), from (b) Optimalcontinu- 01:30 18.00 17.93 ± 0.05 18.07 ± 0.11 17.96 ± 0.02 ous,street not light sporadic, detection monitoring range (from of the 01:30 Madrid to 04:30), night and sky. (c) StreetlightThe observations power increasefrom the (05:30).astro- 05:30 17.93 17.91 ± 0.07 18.03 ± 0.33 17.85 ± 0.09 nomicalThe values observatory indicate thatof Universidad most sky brightness Complutense variability de Madrid happens (inside in the the first city) part began of the in 2010night a (beforend was midnight). ongoing in On 2021. average, Both the broadband night sky panchromatic brightness variability photometers before (SQM midnight and Table 5. Binned Winter Evolution in Johnson R (mag/arcsec2). TESSis 45%,-W) 27%, and andthe All 20%-Sky in theTransmission Johnson B, Monitoring V, and R photometric Camera (AstMon) bands, respectively.provided measure- Before mentsmidnight,Time in the Johnson JohnsonMedian B B, represents V, and(2010 R to aphotometric combination2019) bands of2011 ornamental throughout and 2015the commercialnight, every2019 lighting,night. alongThe21:30 with ongoing the LED stree streetlights.tlight17.19 retrofits The Johnsonin Madrid R17.01 capturesprimarily ± 0.09 the replace dimming17.21 HPS ± 0.15 oflamps HPS17.22 with streetlights. ± 3000K 0.05 LEDs.These01:30 findingsChanges supportinclude previouslythe17.39 bulb, the reported lamp post values17.21 model, measured± 0.04 and 17.42a by modified the ± 0.07 ISS [ 29lighting].17.45 ±regime 0.03 (programmed05:30 change in the17.33 power of street illumination)17.21 ± 0.06 throughout17.39 ± 0.24 the night. 17.32 We± 0.07 obtained data before, during, and after a major streetlight retrofit that took place during the year 2015. Our methodology for analysis includes a manual visual filtering process to avoid noise in the data caused by measures taken during cloud coverage. The statistical analysis of the dataset allows us to determine the impact of the Madrid streetlight retrofits on the brightness and color of the night sky. Below, we discuss our findings on the evolution of night sky brightness and color over time in Madrid together with an analysis of the instrumentation used to collect our data.

4.1.1. SQM and TESS-W The SQM is a popular photometer in the light pollution community due to affordability and an easy-to-use user interface. The SQM bandpass was designed to mimic the human eye response and provide values similar to the Johnson V photometric band [30]. The TESS-W photometer bandpass is defined by a dichroic filter [12] that extends the spectral response to the red range, which allows measurement of the wide HPS line range as

Remote Sens. 2021, 13, 1511 19 of 25

Table 3. Binned Winter Evolution in Johnson B (mag/arcsec2).

Time Median (2010 to 2019) 2011 2015 2019 21:30 18.63 18.61 ± 0.13 18.83 ± 0.09 18.55 ± 0.06 01:30 19.13 19.13 ± 0.07 19.30 ± 0.07 18.97 ± 0.06 05:30 19.08 19.10 ± 0.05 19.21 ± 0.18 18.90 ± 0.27

Table 4. Binned Winter Evolution in Johnson V (mag/arcsec2).

Time Median (2010 to 2019) 2011 2015 2019 21:30 17.73 17.66 ± 0.09 17.81 ± 0.09 17.71 ± 0.05 01:30 18.00 17.93 ± 0.05 18.07 ± 0.11 17.96 ± 0.02 05:30 17.93 17.91 ± 0.07 18.03 ± 0.33 17.85 ± 0.09

Table 5. Binned Winter Evolution in Johnson R (mag/arcsec2).

Time Median (2010 to 2019) 2011 2015 2019 21:30 17.19 17.01 ± 0.09 17.21 ± 0.15 17.22 ± 0.05 01:30 17.39 17.21 ± 0.04 17.42 ± 0.07 17.45 ± 0.03 05:30 17.33 17.21 ± 0.06 17.39 ± 0.24 17.32 ± 0.07

The color index for each winter curve before and after midnight was smoother throughout the night sky from 2010 to 2019 (Bottom of Figure 20). Before midnight, there was no marked change in the B-V color (~0.8). After the streetlight retroﬁt in 2015, the color index after midnight changed considerably from 1.2 to 1.0, indicating bluer skies. Thus, the color difference throughout the night is now signiﬁcantly lower. Finally, higher variability in the IQR was due to anthropogenic activities before midnight, and the brightness change at the end of the night when lights return to the nominal power.

4. Discussion 4.1. Night Sky Brightness Changes over Time We have presented a long-term time series of observations resulting from continuous, not sporadic, monitoring of the Madrid night sky. The observations from the astronomical observatory of Universidad Complutense de Madrid (inside the city) began in 2010 and was ongoing in 2021. Both broadband panchromatic photometers (SQM and TESS-W) and the All-Sky Transmission Monitoring Camera (AstMon) provided measurements in the Johnson B, V, and R photometric bands throughout the night, every night. The ongoing streetlight retrofits in Madrid primarily replace HPS lamps with 3000 K LEDs. Changes include the bulb, the lamp post model, and a modified lighting regime (programmed change in the power of street illumination) throughout the night. We obtained data before, during, and after a major streetlight retrofit that took place during the year 2015. Our methodology for analysis includes a manual visual filtering process to avoid noise in the data caused by measures taken during cloud coverage. The statistical analysis of the dataset allows us to determine the impact of the Madrid streetlight retrofits on the brightness and color of the night sky. Below, we discuss our findings on the evolution of night sky brightness and color over time in Madrid together with an analysis of the instrumentation used to collect our data.

4.1.1. SQM and TESS-W The SQM is a popular photometer in the light pollution community due to affordability and an easy-to-use user interface. The SQM bandpass was designed to mimic the human eye response and provide values similar to the Johnson V photometric band [30]. The TESS-W photometer bandpass is deﬁned by a dichroic ﬁlter [12] that extends the spectral response to the red range, which allows measurement of the wide HPS line range as shown in Figure4 (upper panel). S ánchez de Miguel et al. [11] demonstrated how SQM Remote Sens. 2021, 13, x FOR PEER REVIEW 21 of 26

Remote Sens. 2021, 13, 1511 20 of 25

shown in Figure 4 (upper panel). Sánchez de Miguel et al. [11] demonstrated how SQM leads to misleading results by comparing sky brightness measured with SQM during a leads to misleading results by comparing sky brightness measured with SQM during a retrofit from HPS to LED. They proposed addressing this problem using a color index retroﬁt from HPS to LED. They proposed addressing this problem using a color index adjustment for night sky brightness data measured with SQM and encouraged the use of adjustment for night sky brightness data measured with SQM and encouraged the use of color observations to determine the real evolution of sky brightness. color observations to determine the real evolution of sky brightness. InIn this this work, work, we we h hadad the the opportunity opportunity to to compare compare the the data data collected by TESS TESS-W-W and SQM,SQM, as as well well as an astronomical camera with color information (AstMon).(AstMon). The average winterwinter evolution evolution curve throughout throughout the night for TESS TESS-W-W data data shows shows a a steady steady progressive progressive brighteningbrightening fromfrom 20162016 to to 2019 2019 (Figure (Figure 20 ).20). On On the the other other hand, hand, the SQMthe SQM data data (Figure (Figure 19) show 19) showa brighter a brighter sky in 2016sky in (following 2016 (following 2015, the 2015, retrofit’s the retrofit’s critical year), critica andl year), a progressive and a progressive darkening darkeningof the night of skythe night brightness sky brightness from 2016 from to 2019. 2016 to This 2019. is aThis prototypical is a prototypical example example of how of a howstreetlight a streetlight retrofit retrofit from HPS from to HPS LED to increases LED increases the nocturnal the nocturnal sky brightness, sky brightness, but a simple but a simpleanalysis analysis of the SQMof the data SQM yields data theyields opposite the opposi conclusion.te conclusion. Figure Figure21 shows 21 shows a comparison a com- parisonof the yearly of the median yearly brightnessmedian brightness measured measured by SQM andby SQM TESS-W. and InTESS addition-W. In to addition SQM data to SQMrequiring data a requiring color index a color correction index (Figure correction 15), Bar (Figureá et al. 15), [28 ]Bará and et Puschnig al. [28] etand al. Puschnig [31] recently et al.found [31] that recently SQM photometersfound that SQM do not photometers properly detect do not light properly over several detect years light (an over aging several effect). yearsThey found(an aging a correction effect). They drift found of around a correction +0.05(6) magSQM/arcsecdrift of around +0.05(6)2 per year. magSQM/arcsec This correction2 perlikely year. explains This correction the discrepancy likely inexplains SQM brightness the discrepancy trends with in SQM AstMon brightness and TESS trends data. with The AstMonSQM was and four TESS years data. older The than SQM the TESS-W. was four Therefore, years older a time-dependent than the TESS sensitivity-W. Therefore, change a timemight-dependent also play asensitivity role in these change trends. might also play a role in these trends.

FigureFigure 21. 21. MedianMedian trends trends of of SQM SQM and TESS TESS-W-W data before and after midnight.

4.1.2.4.1.2. ASTMON ASTMON Brightness Brightness and and Color Color Evolution Evolution ToTo show show the the evolution evolution in in brightness brightness and and color color of of the the Madrid Madrid sky sky during during the the night, night, in Figurein Figure 22 we22 wepresent present changes changes in the in median the median values values of brightness of brightness and color and color indices indices for each for yeareach in year the infirst the and ﬁrst second and second parts of parts the night of the using night AstMon using AstMondata. The data. dispersion The dispersion of values isof higher values during is higher the duringfirst part the of ﬁrst the partnight of when the night there when is variation there isin variation human activity, in human as expected.activity, as The expected. second The part second of the partnight of is the consistently night is consistently darker. The darker. evolution The evolution of median of nightmedian sky night brightness sky brightness shows a showscompensated a compensated difference difference of magnitude of magnitude in the Johnson in the Johnson B and B and Johnson R photometric bands. While the Johnson R band increases in magnitude (towards lower brightness), the Johnson B presents the opposite trend, showing increasing

Remote Sens. 2021, 13, x FOR PEER REVIEW 22 of 26

Johnson R photometric bands. While the Johnson R band increases in magnitude (towards lower brightness), the Johnson B presents the opposite trend, showing increasing brightness. The B-V and V-R color indices decrease in value over the years, showing a bluer night sky. The Johnson V photometric band was less sensitive to the streetlight retrofit and did not show a marked trend. To check the significance of the trends in Figure 22, we performed a model selection using leave-one-out cross-validation [32]. The two competing models are: a simple linear Remote Sens. 2021, 13, 1511 trend and a constant value, both including the variance of the data points. We fit the21 mod- of 25 els before and after midnight, and for each magnitude and color. Before midnight, the best model for the Johnson R band and V-R is a linear trend (increase in magnitude and decrease in V-R color), whereas for the rest of magnitudes and colors the best model is flat. Thebrightness. best model The B-Vin all and cases V-R after color midnight indices decrease is a linear in value trend, over except the years, for Johnson showing V awhich bluer wasnight less sky. sensitive The Johnson and did V photometric not show a bandmarked was trend. less sensitive to the streetlight retroﬁt and did not show a marked trend.

. Figure 22. Median trends of the Johnson B, V, and R photometric bands and the associated B-V and Figure 22. Median trends of the Johnson B, V, and R photometric bands and the associated B-V andV-R colorV-R color indices indices before before and afterand after midnight. midnight. To check the significance of the trends in Figure 22, we performed a model selection For the first part of the night, while the increase in Johnson R is significant, the dis- using leave-one-out cross-validation [32]. The two competing models are: a simple linear persion in the median values of Johnson B is enough to explain the changes in this mag- trend and a constant value, both including the variance of the data points. We fit the nitude. In the second part of the night, the evolution of median night sky brightness shows models before and after midnight, and for each magnitude and color. Before midnight, the a compensated difference of magnitude in the Johnson B and Johnson R photometric best model for the Johnson R band and V-R is a linear trend (increase in magnitude and bands. While the Johnson B band presents the opposite trend, showing increasing bright- decrease in V-R color), whereas for the rest of magnitudes and colors the best model is flat. ness. The B-V and V-R color indices decrease in value over the years in the second part of The best model in all cases after midnight is a linear trend, except for Johnson V which was theless night, sensitive showing and did bluer not night show sky. a marked trend. For the first part of the night, while the increase in Johnson R is significant, the dispersion in the median values of Johnson B is enough to explain the changes in this magnitude. In the second part of the night, the evolution of median night sky brightness shows a compensated difference of magnitude in the Johnson B and Johnson R photometric bands. While the Johnson B band presents the opposite trend, showing increasing brightness. The B-V and V-R color indices decrease in value over the years in the second part of the night, showing bluer night sky. We observed that the night sky brightness increased primarily in the blue part of the spectrum (Table6), and most of the detected change took place during 2015 when the major streetlight retrofit from HPS to LEDs occurred. Remote Sens. 2021, 13, 1511 22 of 25

Table 6. Yearly median night sky brightness after midnight.

Johnson Photometric Band (mag/arcsec2) Color Indices Year B V R B–V V–R 2011 19.15 ± 0.17 17.82 ± 0.24 17.11 ± 0.25 1.32 ± 0.20 0.78 ± 0.08 2015 19.27 ± 0.18 18.07 ± 0.22 17.42 ± 0.23 1.20 ± 0.11 0.66 ± 0.04 2019 18.98 ± 0.14 17.95 ± 0.16 17.44 ± 0.18 1.02 ± 0.08 0.51 ± 0.03

4.1.3. Winter Curves The winter curve is uniquely suited to demonstrate the effect of anthropogenic factors on sky brightness data throughout the night because the night is longer than in other seasons, providing more sampling points. Examining the winter curve, we could identify the part of the night when anthropogenic factors were reduced to an extent that the remaining light was primarily from streetlights. These winter curves (Figures 18–20) allow us to understand the effect of the 2015 streetlight retroﬁts on the Madrid night sky brightness and color. We emphasize the importance of identifying the streetlight signal with reduced noise. The large sample, available due to continuous nightly measurement, allowed us to identify this signal and is a strength of our design.

4.1.4. Comparison with Other Studies Using data gathered by satellites, Kyba et al. [33] detected a 2% increase in global night sky brightness in lit outdoor areas. However, they found that measurements by local devices reported greater increases in some regions, suggesting that satellite data sometimes underestimate radiance on a local scale. Our study supports these findings, where TESS-W, a device that measures in the yellowish-red spectra, reported a 10% increase in brightness following a major streetlight retrofit in 2015. We found that the transition towards solid-state LED technology impacts night spectra, supporting findings by Posch et al. [34]. According to Bará et al. [6], night sky brightness responses behave differently for narrower bands. Therefore, an assessment of increases in sky radiance due to transitions from HPS to LED has to include measures of the brightness in the channels of major changes. We measured in the Johnson B, V, and R photometric bands and found that the bands did indeed behave differently. Madrid City Hall did not make significant changes to the number or manufacturing type of streetlights in Madrid after 2015. Hence, the night sky brightening reported in the B and V Johnson photometric band and the TESS-W photometer following the 2015 streetlight retrofit have to be produced by other sources. Potential sources for brightening in the night sky at the UCM observatory after 2015 include: (a) inaccurate Madrid City Hall data (we consider this unlikely, but not impossible), (b) changes to street light diffusers in Madrid that increased the upper light ratio, and (c) street lighting changes in Madrid’s surrounding areas have impacted the night sky brightness measured at the UCM observatory. 5. Conclusions Examining filtered data (no moon, no clouds) from the winter season in Madrid at 03:30 (when other anthropogenic light pollutants are minimized, allowing optimal streetlight signal detection), we found: As regards SQM: 1. A 38% darkening of the night sky from 2012 to 2019, as measured by SQM. After the streetlight retrofit in 2015, the SQM photometer did not follow the trends detected by AstMon in the Johnson V band and the broadband TESS-W photometer. Although the change in shifting lighting types could play a small role, the main effect is the loss of the SQM photometer sensitivity. This experimental outcome has far-reaching implications for long-term global monitoring of night sky brightness, if confirmed by similar studies. As regards TESS-W: Remote Sens. 2021, 13, 1511 23 of 25

2. A 10% brightening of the night sky from 2016 to 2019, as measured by TESS-W. This percentage value followed the trends of Johnson B and V photometric bands, as detected by AstMon. We conclude that due to the inclusion of the dichroic filter, TESS-W is better suited to tracking changes in night sky brightness compared to SQM, allowing for the detection of night sky brightness in the yellowish-reddish range of spectra (HPS range). Regarding the AstMon data: 3. We compared brightness trends in the Johnson photometric bands before and after the streetlight retrofit in 2015. The sky brightness detected in the Johnson B band darkened 14% from 2011 to 2015 and brightened by 32% from 2015 to 2019. The sky brightness detected in the Johnson V band darkened by 13% from 2011 to 2015 and brightened by 11% from 2015 to 2019. The sky brightness detected in the Johnson R band darkened by 19% from 2011 to 2015; it continued to darken from 2015 to 2019, darkening by 1% in this period. The Johnson B band reached its darkest values in 2015, while the Johnson V was darkest in 2014. 4. Radiance in the Johnson R band decreased due to less abundant HPS technology, while radiance in the Johnson B band increased due to more abundant LED technology (at least until 2015). 5. Before midnight, we did not find a marked change in the winter color index. After midnight, following the streetlight retrofit in 2015, the color index shifted considerably toward bluer skies. Prior to the retrofit, there was a shift in the color of the Madrid night sky from bluish (B-V = 0.8) before midnight, to yellowish (B-V = 1.2) after midnight. After the retrofit, the difference in color before midnight (B-V = 0.8) and after midnight (B-V = 1.0) decreased. 6. The B-V and V-R yearly color indices showed a steady decrease in value from 2010 to 2019 (B-V = 1.24 ± 0.09 to 1.03 ± 0.03) and (V-R = 0.78 ± 0.10 to 0.52 ± 0.03), indicating an increased blue component of the light scattered in the night sky of Madrid. The streetlight retrofit from HPS to LED in Madrid: 7. Madrid’s HPS streetlamps remain a strong contributor to the night sky brightness V and R components. The new LED streetlights changed the proportions of the night sky brightness B and V components. 8. Without careful design, a white LED streetlight retrofit is not optimal to reduce light pollution, even after reducing the electric power because of the high content of blue light. The relevant energy savings are not due to the lamps’ energy efficiency but the dimming capabilities and the lower ULOR of the LEDs compared to traditional lighting. An alternative way to reduce light pollution consists of reducing the power of in-place sodium lamps. 9. There was a small decrease in sky brightness following the streetlight retrofit in 2015, but these changes were later reversed.

Supplementary Materials: The following are available online at https://www.mdpi.com/article/10 .3390/rs13081511/s1, Table S1: Johnson B statistics, Table S2: Johnson V Statistics, Table S3: Johnson R Statistics, Table S4: B-V Color Index Statistics, Table S5: V-R Color Index Statistics, Table S6: SQM Statistics for Data After Midnight, and Table S7: TESS Statistics for Data After Midnight. Author Contributions: Conceptualization, J.Z., S.P., J.R., and A.S.d.M.; formal analysis, J.R., J.Z., and S.P.; investigation, J.R., J.Z., S.P., A.S.d.M.; resources, J.Z. and S.P.; data curation, J.R., S.P., and J.Z.; writing original draft, J.R., J.Z., and S.P.; writing review and editing, J.R., J.Z., and K.J.G.; visualization, J.R., J.Z., and S.P.; supervision, J.Z., S.P., and J.R.; project administration, J.G. and J.Z.; funding acquisition, J.G., and J.Z. All authors have read and agreed to the published version of the manuscript. Remote Sens. 2021, 13, 1511 24 of 25

Funding: This research was funded by Secretaría Nacional de Ciencias y Tecnología de Panamá and Instituto para la Formación y Aprovechamiento de Recursos Humanos (270-2018-642); ASTRID (P-ESP-000361-0505); TEC2SPACE (S2018/NMT-4291), Programa Estatal Español de I+D+i Orientada a los Retos de la Sociedad (RTI2018-096188-B-I00); STARS4ALL (H2020-ICT-2015-688135); ACTION (H2020-SwafS-2018-1-824603); and EMISSI@N (NE/P01156x/1). Data Availability Statement: The data presented in this research are openly available in https: //zenodo.org/record/4633001 (accessed on 9 April 2021). Acknowledgments: The original manuscript was improved with the comments and suggestions received from four referees We thank The Cities at Night project for its supports. Conﬂicts of Interest: The Authors declare no conﬂict of interest.

Abbreviations The following abbreviations are used in this manuscript. UCM Universidad Complutense de Madrid NSB Night Sky Brightness LED Light-Emitting Diode HPS High-Pressure Sodium MV Mercury Vapor SQM Sky Quality Meter SQM-LE Sky Quality Meter Lens Ethernet SQM-LU Sky Quality Meter Lens USB SQMs Synthetic Sky Quality Meter (via color index correction) TESS-W Telescope Encoder and Sky Sensor via WiFi AstMon All Sky Transmission Monitor Camera SAND Spectrometer for Aerosol Night Detection AERONET AErosol RObotic NETwork AOD Aerosol Optical Depth 3D Three Dimensional ULOR Upward Light Output Ratio UTC Coordinated Universal Time PySQM Python Sky Quality Meter ASCII American Standard Code for Information Interchange KDE Kernel Density Estimation ISS International Space Station DMSP-OLS Defense Meteorological Program—Line-Scan System SNPP Suomi National Polar-Orbiting Partnership VIIRS-DNB Visible Infrared Imaging Radiometer Suite-Day/Night Band

References 1. Falchi, F.; Cinzano, P.; Duriscoe, D.; Kyba, C.C.M.; Elvidge, C.D.; Baugh, K.; Portnov, B.A.; Rybnikova, N.A.; Furgoni, R. The new world atlas of artificial night sky brightness. Sci. Adv. 2016, 2, e1600377. [CrossRef] 2. Gaston, K.J.; Bennie, J.; Davies, T.W.; Hopkins, J. The ecological impacts of nighttime light pollution: A mechanistic appraisal. Biol. Rev. 2013, 88, 912–927. [CrossRef][PubMed] 3. Bennie, J.; Davies, T.W.; Cruse, D.; Gaston, K.J. Ecological effects of artificial light at night on wild plants. J. Ecol. 2016, 104, 611–620. [CrossRef] 4. Sanders, D.; Frago, E.; Kehoe, R.; Patterson, C.; Gaston, K.J. A meta-analysis of biological impacts of artificial light at night. Nat. Ecol. Evol. 2021, 5, 74–81. [CrossRef] 5. Gaston, K.J.; Holt, L.A. Nature, extent and ecological implications of night-time light from road vehicles. J. Appl. Ecol. 2018, 55, 2296–2307. [CrossRef] 6. Bará, S.; Rigueiro, I.; Lima, R.C. Monitoring transition: Expected night sky brightness trends in different photometric bands. J. Quant. Spectrosc. Radiat. Transf. 2019, 239, 106644. [CrossRef] 7. Barentine, J.C.; Walker, C.E.; Kocifaj, M.; Kundracik, F.; Juan, A.; Kanemoto, J.; Monrad, C.K. Skyglow changes over Tucson, Arizona, resulting from a municipal LED street lighting conversion. J. Quant. Spectrosc. Radiat. Transf. 2018, 212, 10–23. [CrossRef] 8. Tong, K.P.; Kyba, C.C.; Heygster, G.; Kuechly, H.U.; Notholt, J.; Kolláth, Z. Angular distribution of upwelling artificial light in Europe as observed by Suomi–NPP satellite. J. Quant. Spectrosc. Radiat. Transf. 2020, 249, 107009. [CrossRef] Remote Sens. 2021, 13, 1511 25 of 25

9. Estrada-García, R.; García-Gil, M.; Acosta, L.; Bará, S.; Sanchez-de-Miguel, A.; Zamorano, J. Statistical modelling and satellite monitoring of upward light from public lighting. Light. Res. Technol. 2016, 48, 810–822. [CrossRef] 10. De Miguel, A.S.; Kyba, C.C.M.; Zamorano, J.; Gallego, J.; Gaston, K.J. The nature of the diffuse light near cities detected in nighttime satellite imagery. Sci. Rep. 2020, 10, 7829. [CrossRef] 11. De Miguel, A.S.; Aubé, M.; Zamorano, J.; Kocifaj, M.; Roby, J.; Tapia, C. Sky Quality Meter measurements in a colour-changing world. Mon. Not. R. Astron. Soc. 2017, 467, 2966–2979. [CrossRef] 12. Zamorano, J.; García, C.; Tapia, C.; de Miguel, A.S.; Pascual, S.; Gallego, J. STARS4ALL Night Sky Brightness Photometer. Int. J. Sustain. Light. 2017, 18, 49–54. [CrossRef] 13. Kyba, C.C.M.; Lolkema, D.E. A community standard for recording skyglow data. Astron. Geophys. 2012, 53, 6.17–6.18. [CrossRef] 14. Kyba, C.C.M.; Ruhtz, T.; Fischer, J.; Hölker, F. Red is the new black: How the colour of urban skyglow varies with cloud cover. Mon. Not. R. Astron. Soc. 2012, 425, 701–708. [CrossRef] 15. Ramírez, L.A. Medida de La Luminosidad de Fondo de Cielo Del Observatorio de U.C.M. E-Prints Complutense. 2000. Available online: http://guaix.fis.ucm.es/~{}jaz/Documentos/Ramirez_MeCO_2001.pdf (accessed on 2 August 2020). 16. Ruiz, Á.C. Constantes Fotométricas del Observatorio de la UCM. E-Prints Complutense. 2003, p. 40. Available online: https: //webs.ucm.es/info/Astrof/users/jaz/TRABAJOS/FOTOM/ (accessed on 2 August 2020). 17. Kozlovska, B. La Cámara Polar de Gran Campo del Observatorio UCM; UCM: Madrid, Spain, 2007; p. 33. 18. Ramírez Moreta, P. Brillo de Fondo de Cielo con AstMon-UCM. Available online: https://eprints.ucm.es/15000/ (accessed on 2 August 2020). 19. Nievas Rosillo, M.; Nievas Rosillo, M. Absolute Photometry and Night Sky Brightness with All-Sky Cameras. Available online: https://eprints.ucm.es/24626/ (accessed on 2 August 2020). 20. Zamorano Calvo, J.; Sánchez de Miguel, A.; Nievas Rosillo, M.; Tapia Ayuga, C.; Ocaña González, F.; Izquierdo Gómez, J.; Gallego Maestro, J.; Pascual Ramírez, S.; Zamorano Calvo, J.; Sánchez de Miguel, A.; et al. Light Pollution Spanish REECL SQM Network; E-Prints Complutense; Honolulu, HI, USA, 2015. Available online: https://eprints.ucm.es/id/eprint/33274/ (accessed on 2 August 2020). 21. Aubé, M.; Kocifaj, M.; Zamorano, J.; Solano Lamphar, H.A.; Sanchez de Miguel, A. The Spectral Amplification Effect of Clouds to the Night Sky Radiance in Madrid. J. Quant. Spectrosc. Radiat. Transf. 2016, 181, 11–23. [CrossRef] 22. Tapia Ayuga, C.E.; Calvo, J. LED Luminaire Report at LICA-UCM. E-Prints Complutense. 2016. Available online: https: //www.researchgate.net/publication/301432539_LED_luminaire_report_at_LICA-UCM (accessed on 2 August 2020). 23. Zamorano, J.; Sánchez de Miguel, A.; Ocaña, F.; Pila-Díez, B.; Gómez Castaño, J.; Pascual, S.; Tapia, C.; Gallego, J.; Fernández, A.; Nievas, M. Testing Sky Brightness Models against Radial Dependency: A Dense Two Dimensional Survey around the City of Madrid, Spain. J. Quant. Spectrosc. Radiat. Transf. 2016, 181, 52–66. [CrossRef] 24. Sanchez de Miguel, A.; Castaño, J.G.; Jaime, Z.; Pascual, S.; Ángeles, M.; Cayuela, L.; Kyba, C.C.M. Atlas of Astronaut Photos of Earth at Night. Astron. Geophys. 2014, 55, 4–36. 25. Aceituno, J.; Sanchez, S.F.; Aceituno, F.J.; Galadi-Enriquez, D.; Negro, J.J.; Soriguer, R.C.; Gomez, G.S. An All Sky Transmission Monitor: ASTMON. Publ. Astron. Soc. Pac. 2011, 123, 1076–1086. [CrossRef] 26. Nievas Rosillo, M.; Zamorano Calvo, J.; Nievas Rosillo, M.; Zamorano Calvo, J. PySQM the UCM Open Source Software to Read, Plot and Store Data from SQM Photometers. Available online: https://eprints.ucm.es/25900/ (accessed on 2 August 2020). 27. Scott, D.W. Scott’s Rule. WIREs Comput. Stat. 2010, 2, 497–502. [CrossRef] 28. Bará, S.; Marco, E.; Ribas, S.J.; Gil, M.G.; de Miguel, A.S.; Zamorano, J. Direct Assessment of the Sensitivity Drift of SQM Sensors Installed Outdoors. arXiv 2020, arXiv:2011.15044. 29. Sánchez de Miguel, A. Variación Espacial, Temporal y Espectral de la Contaminación Lumínica y sus Fuentes: Metodología y Resultados. Ph.D. Thesis, Universidad Complutense de Madrid, Madrid, Spain, 2016. 30. Cinzano, P. Night Sky Photometry with Sky Quality Meter. ISTIL Int. Rep. 2005, 9, 1–4. 31. Puschnig, J.; Näslund, M.; Schwope, A.; Wallner, S. Correcting Sky Quality Meter Measurements for Aging Effects Using Twilight as Calibrator. Mon. Not. R. Astron. Soc. 2021, 502, 1095–1103. [CrossRef] 32. Vehtari, A.; Gelman, A.; Gabry, J. Practical Bayesian Model Evaluation Using Leave-One-out Cross-Validation and WAIC. Stat. Comput. 2017, 27, 1413–1432. [CrossRef] 33. Kyba, C.C.M.; Kuester, T.; de Miguel, A.S.; Baugh, K.; Jechow, A.; Hölker, F.; Bennie, J.; Elvidge, C.D.; Gaston, K.J.; Guanter, L. Artificially Lit Surface of Earth at Night Increasing in Radiance and Extent. Sci. Adv. 2017, 3, e1701528. [CrossRef] 34. Posch, T.; Binder, F.; Puschnig, J. Systematic Measurements of the Night Sky Brightness at 26 Locations in Eastern Austria. J. Quant. Spectrosc. Radiat. Transf. 2018, 211, 144–165. [CrossRef]