Light pollution is an emerging issue in the evaluationof environmental modifications that are caused byhuman settlement and activity. The level and theextent of artificial illumination at nighttime in agiven region are practical as well effective measuresof the overall activity of the local community and takeinto account both light from industrial and residen-tial sources. The total power emitted by artificiallight sources is, on average, directly related to theoverall energy consumption of the community.1–3
The level of light pollution is therefore a generalmeasure of how human activity modifies its environ-ment, and hence is, although indirectly, an importantparameter for monitoring and modeling all kinds ofartificial pollution. This parameter is also relevantfor economic investigations, because the amount ofpower �nocturnal light emission� that any countrydemands is strictly related to its gross domestic prod-uct.
Light pollution that affects the biosphere is impor-
The authors are with the Institute of Applied Physics “NelloCarrara,” Consiglio Nazionale delle Ricerche, Via Panciatichi 64,50127 Florence, Italy. A. Barducci’s e-mail address [email protected].
Received 10 September 2002; revised manuscript received 14March 2003.
tant for many scientific purposes, such as improve-ment of ground-based astronomical observations.4–9
Because of the remarkable importance of the issuesmentioned above, great effort is being devoted to as-sessing the intensity of scattered radiation that illu-minates the night sky. Two kinds of light-pollutionmeasurements are already being carried out:
Ground-based measurements of the scattered radi-ation level during nighttime is a reliable option forcharacterization of a single geographic site of a mod-erate spatial size. This technology permits accuratemeasurements to be made and is widely employed forchoosing optimal sites for the installation of astro-physical instruments.
Satellite observations instead are often hindered bythe low level of the available signal. The lightsources’ emission, in fact, is spatially localized, andonly a fraction of the radiated power is reflected fromthe ground to the sensor. Therefore specialized in-struments specifically designed for this type of mea-surement have to be used. Nowadays, only onehighly sensitive satellite instrument �the DefenseMeteorological Satellite Program’s Operational Linescan System # �DMSP-OLS�� meets this requirement,and it is often used for remote assessment of lightpollution.10–12
Ground-based observations have high accuracy butcannot encompass sufficiently large terrain to pro-vide light-pollution maps over large areas; satellite-based measurements suffer from coarse spatialresolution and lower accuracy. Moreover, available
satellite measurements do not provide the user withinformation about the spectrum of the source �theDMSP-OLS has only one broadband channel to coverthe entire visible spectral range�, thus making foradditional difficulty in subsequent data processing�e.g., modeling of atmospheric effects and computa-tion of atmospheric light scattering�. To overcomethis difficulty several authors have performed nu-merical simulations of the expected stray-light levelonly, relying on average population density data forthe estimate of surface light emission.
Surface-light emission e�x, y� is the basic quantityused as input information for any theoretical modelintended to compute the stray-light level in a givenviewing direction. The essential structure of thistype of calculation is13
L� x, y� � � e��,�� f � x, y; �,��d�d� , (1)
where f�x, y; �,�� is an empirical propagation functionthat describes the way in which artificial radiationemitted at location �, � is conveyed by scattering atlocation x, y in a given propagation direction. Undersome selective hypotheses �e.g., when only small geo-graphical areas with no height variation are investi-gated�, this integral equation may be reduced to asimple convolution product, and integral kernel f� �will become shift invariant. As has been pointed outby other authors,1 the use of population density datafor estimating surface-light emission e�x, y� suffersfrom some errors and uncertainties:
1. The average population density data of severalcountries are collected by use of different methodsand timetables, which may be out of date,
2. The relation between population density andsurface-light emission is only partially known andcan vary noticeably with geographical location andwith time as a result of economic development andtype of industrial production; and
3. Function f�x, y; �, �� is not known with the re-quired accuracy.
Conversely, maps of urban light gathered at nightfrom remote-sensing instruments represent, in a firstinstance, a good approximation of quantity e�x, y� aslong as the atmosphere is optically thin and diffuseradiance is, on average, far below the radiance of therelated sources. The discussion above has shownhow data on light pollution are important for manypurposes and in scientific domains and that assess-ment of light pollution is still far from reaching sat-isfactory accuracy. In this connection thedevelopment of new sensors specifically designed fornocturnal Earth observations in the visible spectralrange or the application of existing sensors to thisproblem will be important. Anyway, the selectedsensors are required to provide higher spatial reso-lution than the DMSP-OLS does and acceptable spec-tral discriminability. The availability of surfaceemission measurements with high spatial and spec-
tral accuracy will lead to the calculation of more-detailed light-pollution maps for various visiblesubbands.
In this paper we report the results of research de-voted to investigating the monitoring of light pollu-tion by standard hyperspectral imagers that areroutinely employed for daytime remote sensing of theEarth. In particular, we describe the results of anocturnal remote-sensing campaign performed inSeptember 2001 to assess levels of light pollution ofurban and industrial origin. For the first time to ourknowledge, nocturnal light pollution was measuredat high spatial and spectral resolution by use of twoairborne hyperspectral sensors, namely, the Multi-spectral Infrared and Visible Imaging Spectrometer�MIVIS� and the Visible Infrared Scanner �VIRS-200�. These sensors, which provide spectral cover-age from the visible to the thermal infrared, wereflown in Italy, over the Tuscany coast, on board aCasa 212 airplane. The acquired images showed amaximum scene brightness that was comparable tothat observed during similar daytime measurements,whereas their average luminosity was 3 orders ofmagnitude lower. The measurements, performedunder clear-sky conditions before moonrise and in-volving only small-sized towns, confirm that artificialillumination hinders astronomical observations andproduces noticeable effects even at great distancesfrom its source.
We found that only a small number of the instru-ments usually employed for remote-sensing cam-paigns can be successfully converted for monitoring ofnocturnal light pollution, which requires fine spectralresolution and good radiometric sensitivity. Weshow that if the Earth is remotely imaged with asensor that has a small-bandwidth �e.g., 2 nm� spec-tral channel that is resonant with an emission spec-tral line of common city lights, the related lightsources are easily recognized and that the reliabilityof source detection is higher than that allowed bypanchromatic or low-resolution sensors.
In Section 2 we give details about the technicalcharacteristics and imaging capabilities of the twodigital imagers described here, together with infor-mation regarding the experiment. Section 3 is de-voted to showing the raw data that were obtained andthe processing method adopted to validate the obser-vations and to compute higher-level remote sensingproducts. Finally, in Section 4 we draw some con-clusions and point out some unresolved questions.
2. Remote-Sensing Experiment
To investigate the use of standard hyperspectral sen-sors to monitor levels of nocturnal stray light pro-duced by artificial sources, we performed an aerialsurvey of the Tuscany coast in September 2001.Two hyperspectral sensors, the MIVIS and the VIRS-200, were selected to provide fine spatial and spectralresolution and extended spectral coverage. Theremote-sensing campaign was preceded by some pre-liminary activity that is documented here.
The remote-sensing campaign was preceded by in-field activity for observation of the signals emittedfrom light sources, information that allowed us tooptimize the sensor configuration.
Several urban lights were observed at night fromthe ground by a portable spectrophotometer operat-ing in the range 300–950 nm. The instrument isbased on an optical-fiber-fed Rowland reflection grat-ing whose output is sampled by a silicon photodiodearray manufactured by Hamamatsu. Spectra com-prise 256 10-bit independent samples acquired with aspectral resolution of 7 nm and pixel dispersion of2.96 nm�pixel. A thorough discussion of the maintechnical characteristics and performance of thisspectrometer can be found in Ref. 14.
Ground-based observations of the spectral radi-ance emitted by urban lights was aimed at recogniz-ing the main emission lines from more-diffuse lamps.This knowledge was in its turn utilized for wave-length selection of the 20 spectral channels acquiredby the VIRS-200 imager. The basic assumption wasthat because artificial sources emit most of their en-ergy in narrow spectral features they can easily bedetected by those standard remote sensing sensors�planned for daytime activity� that have fine spectralresolution. In this context the VIRS-200 appearedto be an ideal sensor because it collects radiance dataat a spectral resolution of 2.5 nm and therefore has abandwidth that is comparable with the widths of thelines emitted by the observed sources.
Figures 1–3 show the more interesting spectral fea-tures obtained from ground-based observations afterten independent measurements were averaged foreach plot. Because of the uncalibrated radiometricresponse of the instrument, the relative intensities ofthe various spectral lines shown in these figures maybe unrealistic. Let us note that often-utilized illu-mination sources that also include many residentiallights are based on mercury or sodium lamps, asshown in Figs. 1–3.
We note that the spectrum plotted in Fig. 3 con-tains spectral lines from various elements, amongwhich we have identified mercury and probably neonand krypton–helium. Another kind of source that iseasily found comprises filament lamps, which emitcontinuous spectra in the visible and the infrared.These sources do not need specific preflight calibra-tion.
B. Flat-Field Measurement
We examined the application of new and simpleequipment that is able to produce a spatially andangularly flat radiation source. We used this sys-tem to assess the response of the VIRS-200 push-broom sensor.
In its simplest form this system is made from twoplanar diffusers placed on parallel planes at a givendistance. Ideally, the first diffuser produces homo-geneous illumination of the second diffuser, uponwhich, however, a nonisotropic E.M. field is placed.The second diffuser then homogenizes the impingingradiation between the different propagation direc-
Fig. 1. Spectrum of a city lamp emitting yellow–orange light.The measurement was made at nighttime with a portable spectro-photometer operating in the range 300–950 nm. The source isprobably a high-pressure sodium lamp, whose lines are easily rec-ognized.
Fig. 2. Spectrum of a city lamp emitting white–yellow light. Themeasurement was made at nighttime with a portable spectropho-tometer operating in the range 300–950 nm. Typical mercuryspectral lines can be seen.
Fig. 3. Spectrum of a city lamp emitting quasi-white light. Themeasurement was made at nighttime with a portable spectropho-tometer operating in the range 300–950 nm. Some spectral linesof mercury can be seen; however, the main spectral features aredue to other elements, probably neon, krypton, or helium.
tions, hence producing an accurate approximation ofan extended, uniform, and isotropic source.
Perfect diffusers produce exactly uniform and iso-tropic output. In real circumstances the diffuserswill be characterized by a variable phase functionwhose value will decrease significantly for scatteringangles that exceed a specific threshold , for in-stance, the FWHM of the phase function. In thiscase we achieve uniformity of illumination over thesecond diffuser by constraining distance H betweenthe diffusers to obey the following relationship:
H tan � ��2� � D�2 , (2)
where D is the size of each diffuser. This systembehaves as a uniform source of radiation that emitsan almost constant radiation field up to an angle .
The flat-field measurements were executed withtwo high-color-temperature lamps illuminating aSpectralon plate that we used as standard white.The flat-field calibration was compared with that ob-tained with an independent method that retrievesthe changes in sensitivity of the detector by usingonly image information, without any laboratory da-ta.15 The agreement of the two measurements wasexcellent.
C. Airborne Sensors
The VIRS-200 is a push-broom imaging spectrometerequipped with a two-dimensional array detector�CCD� and operating in the visible and the near-infrared spectral ranges. The sensor acquires 20 of240 available spectral channels that are uniformlyspaced from 400 to 1000 nm in 2.5-nm increments.The wavelengths of the 20 channels, which are digi-tized with 10-bit accuracy, are chosen by the operatorof the spectrometer before the overflight. The maincharacteristics of the sensor are detailed in Table 1,and additional information concerning this sensor isgiven in other publications.16,17
Let us note that the images gathered by this kindof sensor are affected by systematic patterns of dis-turbance that originate from fluctuations in the sen-
sitivity of the two-dimensional array detector. Toprevent occurrence of this artifact we performed someflat-field calibration measurements in the laboratorybefore and after the overflight.
Following the preliminary ground-based measure-ments described in Subsection 2.B, we selected aspectral configuration that was able to map the prin-cipal emission lines observed so far. In particular,four spectral channels were chosen for observation ofthe wide spectral emissions from the high-pressuresodium sources shown in Fig. 1 �broad features lo-cated roughly at 570 and 600 nm�. Six channelswere devoted to the measurement of three mercuryemission lines �at 435.85 at 546.07 nm and an unre-solved doublet 576.96–579.07 nm; Fig. 2�. Six otherspectral channels were selected for monitoring ofthree emission lines from mixed vapor sources, thathad wavelengths near 450, 533, and 586 nm �Fig. 3�.The remaining four spectral channels were used asreferences and tuned to the following wavelength po-sitions: 506.25, 701.25, 801.25, and 936.25 nm.These channels could also observe continuum emis-sion from filament, fluorescent, and high-pressurelamps. The channel at 936.25 nm falls within astrong atmospheric absorption band caused by H2Ovapor. The final spectral configuration of the VIRS200 is shown in Table 2.
The MIVIS is a scanning �whisk-broom� imagingspectrometer, which collects image data from 102 in-dependent spectral channels that cover the visible,the near and medium infrared to 2500 nm, and thethermal infrared spectral ranges with ten channels.Extended spectral coverage together with fine radio-metric accuracy �12 bits� makes the MIVIS a usefulsensor for many remote-sensing applications. Inour case the MIVIS also represents an importantreference for VIRS 200 data because of the availabil-ity of ten thermal infrared channels that can be use-ful for fire recognition �rejection� and geolocation.
Because of its operating mode the MIVIS does notneed flat-field calibration, whereas standard radio-metric and spectral calibrations are routinely per-
Table 1. Technical Characteristics of the MIVIS and VIRS-200 Instruments
Sensor
MIVIS VIRS-200
Type Whisk-broom Push-broomNumber of channels 102 20 on 240 selectableFree spectral range 20 channels from 0.43 to 0.83 m �20 nm FWHM� 0.4–1.0 m �2.5 nm
FWHM�8 channels from 1.15 to 1.55 m �50 nm FWHM�64 channels from 1.985 to 2.479 m �9 nm FWHM�10 channels from 8.21 to 12.7 m �0.36 m FWHM�
Instantaneous FOV �mrad� 2.0 1.0Spatial sampling interval �mrad� 1.64 1.33Number of Cross-track samples 755 512Scan rate �scans�s� 6.25, 8.3, 12.5, 16.7, 25 12, 20, 30Quantization accuracy �bits� 12 10Signal-to-noise ratio �at albedo 0.5� 200–800 up to 1.3 m 20–400
formed. Table 1 summarizes the main technicalcharacteristics of the MIVIS.
D. Remote-Sensing Campaign
On 13 September 2001, a CASA 212 airplaneequipped with the two imaging spectrometers de-scribed above was flown at heights that ranged from1500 to 3000 m. The flight was begun at approxi-
mately 1:00 A.M. �GMT�, roughly 2 h before moon-rise, from north– northwest toward south–southeast,over the Tuscany coast. Two small towns �Livornoand Viareggio� and some isolated buildings were im-aged under clear-sky conditions. Figure 4 sketchesthe path of the overflight.
The duration of the overflight was �1 h, duringwhich some in-field measurements were performed tovalidate the remotely sensed spectra. These mea-surements were made with the portable spectropho-tometer described in Subsection 2.A.
The VIRS-200 dark signal was collected twice dur-ing the flight, immediately after noticeable environ-mental temperature changes originated fromvariation in altitude of the plane. These ancillarydata were subsequently used to improve the radio-metric calibration of the sensor.
3. Results and Data Processing
Remotely sensed data preprocessing to produceradiance-calibrated images comprised the followingsteps:
1. Dark-signal subtraction,2. Flat-field correction �only for the VIRS-200�,3. Conversion to radiance units �nW cm�2 sr�1
nm�1�,4. Geometric correction and image geolocation.
This preprocessing produces a first level of cali-brated remote-sensing data that is useful for our re-
search: the at-sensor radiance. Let us note thatthe radiance observed by the sensor is not that emit-ted by the corresponding source. Generally, thepresence of a shield prevents the emitted light fromreaching the sensor; hence the urban lamp radiationis initially reflected from the ground and then reachesthe sensor after having been dimmed by the atmo-sphere. The main effects of the atmosphere on vis-ible and near-infrared radiation are absorption andscattering; in the thermal infrared range appreciableatmospheric emission is observed, and scattering be-comes negligible.
Therefore the radiance obtained from the foursteps outlined above is generally underestimated,and one should compute the at-source radiance byremoving the main atmospheric effects. This goalrequires reliable theoretical modeling of the radiativetransfer inside the atmosphere.
The basic relationship that drives the first twosteps is
g� x, y, �� � �i� x, y, �� * H����S� x, �� � g0� x, y, �� ,(3)
where x and y are the horizontal and vertical imagecoordinates, respectively, g is the observed image, i isthe imaged source, g0 is the average dark signal, S isthe detector sensitivity, H is the spectral point-spread function for the monochromator that is used,and i * H is the ideal image to be observed.
Dark-signal removal is important because it re-moves the average effects of noise and bias �g0� thatare superimposed upon faint as well as upon rarelight sources. To eliminate false signal �noise� thatmay have survived this Processing, we also filteredthe raw image at a 3 g threshold, where g is thedark signal’s standard deviation, which is assumed tobe equal to the noise. The point sources validated sofar represent with more than 98% confidence, realsources of radiation. The two parameters g0 and gneeded for processing were deduced by dark-signalmeasurements performed during the flight. Notethat for the VIRS-200 both g and average dark sig-nal g0 depend on the image column index �i.e., theacross-track x coordinate�.
We performed a conversion to radiance units byusing calibration data measured during a remote-sensing campaign carried out in 2000. We correctedthe MIVIS image geometry taking by into accountonly the perspective effect introduced by the scanningunit, because the survey was conducted over a flatcoastal terrain where orography was negligible. Wenote that the VIRS-200, because of its push-broomoperating mode, directly produces instrumental-distortion-free images.
A. Data Analysis
Figure 5 shows a MIVIS composite image of Livornoobtained by use of thermal infrared �100th channel asblue� and visible �22th channel as green and 24thchannel as red� radiometrically calibrated data.Compared this image with the corresponding VIRS-
200 image, which is shown in Fig. 6 and maps the15th channel �586.25 nm� as red, the 13th channel�578.75 nm� as green, and the 9th channel �546.25nm� as blue. The VIRS-200 image portrayed in Fig.6 corresponds to partially processed data that re-mained after subtraction of the dark signal, calibra-tion of the flat field, and conversion to radiance units.Additional filtering aimed at removing faintersources and residual noise was also applied. Thewavelength positions of the spectral channels usedfor generating these false colorimages were selectedsuch that we could map the more important emissionlines recognized during the preflight activity andshown in Figs. 1–3. It is impressive to note how wellthe VIRS-200 imager detects even faint city lights.
Comparison of Figs. 5 and 6 clearly shows that, asexpected, the VIRS-200 provides higher sensitivity tourban lights than does the MIVIS. This importantbehavior is ascribed to the circumstance that a narrowspectral channel tuned at a wavelength resonant witha source’s emission line senses a higher signal than abroadband channel does. In other words, it can besaid that urban lights emit, on average, significantlylower power than solar radiation reflected by soil.However, the difference in emitted radiance betweenthe two sources becomes small in the narrow spectralbands in which city light emission is concentrated.
Figure 7 shows a VIRS-200 image of a peripheralzone of Livorno together with the spectrum of ex-tended and powerful urban light. This source prob-ably is a low-pressure lamp that still has not beenidentified. This type of source has rarely been ob-served during in-field measurements performed tovalidate the remote-sensing products. Figure 8shows the next image, which portrays the downtownarea and the spectrum of a typical light source. Thisurban light has been identified as that of a low-
Fig. 5. MIVIS image acquired over Livorno on 19 September 2001at nighttime. We obtained this false color composite by mappingthe 8th channel �580.0 nm� as red, the 6th channel �540.0 nm�, asgreen, and the 100th channel �11450.0 nm� as blue. The imagedata �radiometrically calibrated� were flat-field calibrated and cor-rected for the dark signal.
pressure mercury lamp. Note the high power out-put by the two sources in Figs. 7 and 8, which isroughly 1 order of magnitude less than diurnal soilbrightness in the same spectral range.
Spectra of various types of city light retrieved fromthe VIRS-200 data at several locations are shown inFigs. 9 and 10. Any curve represents the spectralradiance of a single source, which often appears to beextended in the image. Thus the various spectracontained in a given figure are related to variousimage pixels that seem to belong to the same urbanlight. These spectra from adjacent pixels may orig-inate from the ground halo of a single powerful cen-tral source as well as from unresolved secondarysources. We point out that imaged spectra show alarge variety of city lights, a circumstance that isusually found in urban areas. The relatively highpower emitted by most of these sources suggests thepresence of a certain number of unshielded �or par-tially shielded� lights, the radiation of which perhapsis directly seen by the sensor. These city lightsshould, however, be recognized even in the MIVIS
images, which instead reveal only a small number ofsources �Fig. 5�.
In particular, Fig. 9�a� shows the spectrum of apowerful source recognized as a high-pressure so-dium lamp, and Fig. 9�b� illustrates the spectralemission from a high-pressure sodium source that isprobably mixed with a smaller mercury component.We were not able to assess whether this spectrumresults from mixing of two independent sources lo-cated close to each other or whether the source is asingle lamp with a nonstandard gas fill. Figures 9�c�and 9�d� detail the spectral emissions of two low-intensity city lights from sodium �Fig. 9�c�� and mer-cury �Fig. 9�d�� lamps.
Figure 10�a� shows strong spectral emission from acontinuous source, probably a halogen lamp. Figures10�b� and 10�c� indeed show the spectra of two citylights that have not been identified but both of whichcontain mercury spectral lines. In particular, thebright source of Fig. 10�c� is of the same type as thatshown in Fig. 3, whose spectrum was observed duringthe in situ campaign. We point out that the radiance
Fig. 6. VIRS 200 image acquired over Livorno on 19 September 2001 at nighttime. We obtained this false color composite by mappingthe 15th channel �586.25 nm� as red, the 13th channel �578.75 nm� as green, and the 9th channel �546.25 nm� as blue. The image datawere flat-field calibrated and corrected for the dark signal.
signal revealed by the VIRS 200 sensor after the 3 gfiltering described in Section 2 is still affected by thelow-level diffuse signal observed almost everywhere inthe calibrated images. This signal does not show anymeaningful spatial structure; nonetheless its ampli-tude and average power are too high to be interpretedas noise that was not removed by the 3 g filtering.Figure 11, which portrays a rural scene essentially freefrom lights and inhabited areas, is an example of thisphenomenon. Figure 12 shows the average stray-light spectrum retrieved from a 30 � 30 window in thecenter of the image in Fig. 11. It can be seen that theamplitude of this diffuse signal is far below the city-light intensity shown in Figs. 9 and 10 and that itgrows with decreasing wavelengths. Let us note thatthe first spectral channel of the VIRS-200 �first point ofthe plot� is less accurate because of degraded sensorperformance at the beginning of its free spectral range�worst signal-to-noise ratio�.
B. Simulations
We also simulated stray light affecting the night skyas perceived by an observer looking at zenith from the
ground. To this purpose we utilized the mathemat-ical framework depicted by Eq. �1� with the followingpoint-spread function:
f � x, y, �, �� � a���� U� x � ��2 � � y � ��2 � h2
�V
�� x � ��2 � � y � ��2 � h2�1�2
� exp��k����� x � ��2 � � y � ��2 � h2�1�2� , (4)
Here x, y indicate the observer’s location, �, � are thesource coordinates, and surface light emission e�x, y�has been estimated from the radiometrically cali-brated VIRS-200 images. The average atmosphericextinction coefficient k ��� was estimated from a MODT-RAN run in the spectral range covered by the VIRS 200sensor. We used a MODTRAN estimation of atmo-spheric spectral transparency to compute the opticalthickness of the whole atmosphere, a result that wecompared with theoretical predictions of the verticalprofile of the extinction coefficient modeled as an ex-ponential decay law. Height h of the scattering layerwas fixed at 2.4 km; the values of U and V together
Fig. 7. �a� VIRS 200 radiometrically calibrated image of a periph-eral area of northern Livorno acquired on 19 September 2001 atnighttime. We obtained this image by summing the radiometri-cally calibrated radiances collected at the 8th �543.75 nm�, the 12th�576.25 nm�, and the 14th �583.75 nm� spectral channels. �b�Spectrum of a powerful source �not reliably identified yet� outlinedin the image by a white square.
Fig. 8. �a� VIRS 200 radiometrically calibrated image of a periph-eral area of northern Livorno acquired on 19 September 2001 atnighttime. We obtained this image by summing the radiometri-cally calibrated radiances collected at the 8th �543.75 nm�, the 12th�576.25 nm�, and the 14th �583.75 nm� spectral channels. �b�Spectrum of a powerful source �probably a low-pressure mercurylamp� outlined in the image by a white square.
Fig. 9. Spectra of various urban lights extracted from the VIRS 200 image after processing for dark-signal subtraction, flat-fieldcalibration, and transformation to radiance units. �a� Emission from a powerful source, recognized as a high-pressure sodium lamp. �b�spectral emission from a high-pressure sodium source probably mixed with a smaller mercury component. It is not clear whether thisspectrum results from mixing of two independent sources located close to each other or whether it is from a single lamp with a nonstandardgas fill. �c� Low-intensity city light whose source is probably a sodium lamp. �d� Emission from low-intensity city light whose source wasprobably a mercury lamp.
Fig. 10. Spectra of various city lights extracted from the VIRS 200 image after processing for dark-signal subtraction, flat-fieldcalibration, and transformation to radiance units. �a� Strong spectral emission from a continuous source, probably a halogen lamp. �b�Spectrum of a faint city light that has not been identified, which contains contributions mercury spectral lines. �c� Spectrum of anunidentified bright source, which contains mercury spectral lines. This spectrum seems to be of the same type as that shown in Fig. 3that was observed during the in situ campaign.
with the relative spectrum of amplitude factor a ���were taken from the literature.18,19 We note that pre-viously reported studies were generally aimed to esti-mate the nocturnal stray-light level by use ofpopulation data only. Hence the empirical valuefound for amplitude factor a ��� converts populationdensity into radiance. However, we possess an accu-rate map describing the radiance coming from grounddistribution of urban light; as a consequence werescaled amplitude a ��� by a factor of 105 to obtain arough estimate of the radiance produced by a givenground location. Nonetheless, we remark that a goodestimation of the true radiance observed at a givenpoint would require more-accurate modeling of groundradiance e��,��, which could be affected by local groundreflectance20–25 and other unpredictable factors.
A final point regarding the integration in Eq. �1�that was made with the VIRS 200 radiance-calibrated signal used as the ground distribution oflight e��,�� is related to the narrow swath imaged bythis sensor, i.e., �1.2 km. The result obtained fromour simulations is always underestimated becauseour calculation excludes city lights outside the VIRS-200 image �1.2 km wide�, whereas the point-spreadfunction that we used �Eq. �4�� has a full width of notless than h �2.4 km�.
Results from numerical simulations are shown inFigs. 14 and 15. Figure 13 shows the input imageused as a sources map: it is a scene that comprisesthe center of the town of Livorno and therefore the
more polluted region observed during our overflight.The nocturnal radiance distribution that would beobserved in the same geographic area by seeing atzenith is displayed in Fig. 14, which also shows anunrealistic side effect �darkening� caused by thesmall spatial extent of the available image data.The spectrum of this scattered radiance as averagedon a 30 � 30 window located in the center of theimage �which corresponds to the center of the town� isplotted versus wavelength in Fig. 15. As can beseen, this spectrum is very similar to the spectrum ofdiffuse light perceived by the VIRS 200 far from anyground source �Fig. 12�.
4. Conclusions
The problem of light pollution that affects night skynear urban areas has been reexamined. The optionof measuring diffuse light at nighttime by use of stan-dard remote-sensing instruments has been investi-gated. A remote-sensing campaign was performedin September 2001 to assess the stray-light level nearthe Tuscany coast �Italy� and to investigate the opti-mal experimental methodology for this detection.Two hyperspectral sensors, the MIVIS and the VIRS-200, were flown on board a Casa 220 airplane at1.5–3.0-km heights. The acquired images showed amaximum scene brightness �at-sensor radiance� thatwas three to five times less than that that observedduring similar daytime measurements. The mea-surements have confirmed that observation of noc-turnal stray light is possible by use of narrowbandhyperspectral imagers that have bandpass channelscentered about the emission features of common ar-tificial light sources.
A spectrum of nocturnal diffuse light was retrievedfrom the processed �calibrated� data and has shownsome interesting properties: As a general rule, in-creasing diffuse light toward the blue was found.This property was to a degree expected, because Ray-leigh scattering efficiency increases strongly atshorter wavelengths. The spectrum of stray lightalso showed the spectral signatures of commonly
Fig. 11. Gray-scale VIRS-200 image acquired at the north of thetown of Viareggio over a ground area free from light sources. Thedata were processed to remove the dark signal, flat-field calibrated,and transformed to radiance units. This image was filtered up toa 3 g threshold to mitigate the effects of false signals caused bynoise. The weak diffuse signal shown in this image cannot beexplained as a residual effect of noise; probably it represents straylight. This image was stretched to enhance the low-level part ofthe signal, which is evident in this picture.
Fig. 12. Upwelling radiance �stray light� measured with the VIRS200 over a ground area that was free from light sources. The datawere processed to remove the dark signal �filtered up to 3 g�,flat-field calibrated, and transformed to radiance units. Thestray-light spectrum averaged over a 30 � 30 pixel windows in thecenter of the scene in Fig. 11 is shown.
used line emission lamps. The amplitude of this at-sensor radiance, which originated from scattering oflight emitted by ground sources outside the imagedscene, is 3–4 orders of magnitude less than the cor-responding amplitude at sensor radiance measuredduring daytime observations. This experimental re-
sult confirms that light pollution is a relevant hand-icap to astronomical observation and even to theintegrity of environment.
The authors thank the Istituto Geografico MilitareItaliano for allowing the use of the VIRS-200 sensor
Fig. 13. Colorcomposite of the VIRS-200 image �input data� obtained by mapping of the 8th channel �543.75 nm� as blue, the 12th channel�576.25 nm� as green, and the 14th channel �583.75 nm� as red. We used this VIRS 200 calibrated image as a map of ground-emittedradiance and integrated it with a response function to estimate the radiance that a ground-based observer would measure on looking atnighttime in the zenith direction.
that has been widely utilized in this research. Thisresearch has been partially supported by the ItalianSpace Agency.
References1. C. D. Elvidge, K. E. Baugh, E. A. Kihn, H. W. Kroehl, E. R.
Davis, and C. W. Davis, “Relation between satellite observedvisible-near infrared emissions, population, economic activityand electric power consumption,” Int. J. Remote Sens. 18,1373–1379 �1997�.
2. M. L. Imhoff, W. T. Lawrence, C. D. Elvidge, T. Paul, E. Levine,M. V. Privalsky, and V. Brown, “Using nighttime DMSP�OLSimages of city lights to estimate the impact of urban land use
on soil resources in the United States,” Remote Sens. Environ.59, 105–117 �1997�.
3. S. Isobe and S. Hamamura, “Light energy loss and its appli-cation to estimating a global energy usage,” in HyperspectralRemote Sensing of the Land and Atmosphere, W. L. Smith andY. Yasuoka, eds., Proc. SPIE 4151, 237–245 �2001�.
4. R. Berry, “Light pollution in southern Ontario,” J. R. Astron.Soc. Can. 70, 97–115 �1976�.
5. R. H. Garstang, “Model for artificial night-sky illumination,”Publ. Astron. Soc. Pac. 98, 364–375 �1986�.
6. R. Pike, “A simple computer model for the growth of lightpollution,” J. R. Astron. Soc. Can. 70, 116–126 �1976�.
7. P. Cinzano, F. Falchi, and C. D. Elvidge, “The first world atlasof the artificial night sky brightness,” Mon. Not. R. Astron. Soc.328, 689–707 �2001�.
8. P. Massey and C. B. Foltz, “The spectrum of the night sky overMount Hopkins and Kit Peak: changes after a decade,” Pub.Astron. Soc. Pacific �to be published�.
9. S. Isobe and H. Kosai, “Star watching observations to measurenight sky brightness,” Astron. Soc. Pac. Conf. Ser. 139, 175–184 �1998�.
10. W. T. Sullivan, “A 10 km resolution image of the entire night-time Earth based on cloud-free satellite photographs in the400–1100 nm band,” Int. J. Remote Sens. 10, 1–5 �1989�.
11. C. D. Elvidge, K. E. Baugh, J. B. Dietz, T. Bland, P. C. Sutton,and H. W. Kroehl, “Radiance calibration of DMSP-OLS low-light imaging data of human settlements,” Remote Sens. En-viron. 68, 77–88 �1999�.
12. M. L. Imhoff, W. T. Lawrence, D. C. Stutzer, and C. D. Elvidge,“A technique for using composite DMSP�OLS ‘City Lights’satellite data to map urban area,” Remote Sens. Environ. 61,361–370 �1997�.
13. M. F. Walker, “Light pollution in California and Arizona,”Publ. Astron. Soc. Pac. 85, 508–519 �1973�.
14. A. Barducci, P. Marcoionni, I. Pippi, and M. Poggesi, “Devel-opment of a solar spectro-irradiometer for the validation ofremotely sensed hyperspectral images,” presented at the 21stSymposium of European Remote Sensing Laboratories,Marne-la-Vallee �Paris�, 14–16 May 2001.
15. A. Barducci and I. Pippi, “Analysis and rejection of systematicdisturbances in hyperspectral remotely sensed images of theEarth,” Appl. Opt. 40, 1464–1477 �2001�.
16. A. Barducci and I. Pippi, “The airborne VIRS for monitoring ofthe environment,” in Sensors, Systems, and Next-GenerationSatellites �EUROPTO ’97�, H. Fujisada, ed., Proc. SPIE 3221,437–446 �1998�.
18. Ch. Leinert, S. Bowyer, L. K. Haikala, M. S. Hanner, M. G.Hauser, A.-Ch. Levasseur-Regourd, I. Mann, K. Mattila, W. T.Reach, W. Sclosser, H. J. Staude, G. N. Toller, J. L. Weiland,J. L. Weinberg, and A. N. Witt, “The 1997 reference of diffusenight sky brightness,” Astron. Astrophys. Suppl. Ser. 127,1–99, �1998�.
19. J. H. Joseph, Y. J. Kaufman, and Y. Mekler, “Urban lightpollution: the effect of atmospheric aerosols on astronomicalobservations at night,” Appl. Opt. 30, 3047–3058 �1991�.
20. C. D. Elvidge, K. E. Baugh, V. H. Hobson, E. A. Kihn, H. W.Kroehl, E. R. Davis, and D. Cocero, “Satellite inventory ofhuman settlements using nocturnal radiation emission: acontribution for the global toolchest,” Global Change Biol. 3,387–395 �1997�.
21. D. R. Steinwand, “Mapping raster imagery into the inter-rupted Goode Homolosine Projection,” Int. J. Remote Sens. 15,3463–3472 �1993�.
22. P. Sutton, D. Roberts, C. Elvidge, and H. Meij, “A comparisonof nighttime satellite imagery and population density for the
Fig. 14. Gray-scale image of simulated downwelling radiance�stray light� computed with the VIRS-200 calibrated image of Fig.13 used as a map of ground-emitted radiance. We integrated theVIRS-200 signal with a response function to estimate the radiancethat a ground-based observer would measure on looking at night-time in the zenith direction.
Fig. 15. Stray-light spectrum extracted from the center of thesimulated image of Fig. 14. This spectrum should be comparedwith that of Fig. 12.
continental United States,” Photogram. Eng. Remote Sens. 63,1303–1313 �1997�.
23. C. D. Elvidge, K. E. Baugh, E. A. Kihn, H. W. Kroehl, and E. R.Davis, “Mapping lights with nighttime data from the DMSPOperational Linescan System,” Photogram. Eng. Remote Sens.63, 727–734 �1997�.
24. F. E. Volz, “Measurements of the skylight scattering function,”Appl. Opt. 26, 19, 4098–4105 �1987�.
25. P. Cinzano, F. Falchi, C. D. Elvidge, and K. E. Baugh, “Theartificial night sky brightness mapped from DMSP satelliteOperational Linescan System measurement,” Mon. Not. R.Astron. Soc. 318, 641–657 �2000�.
