1 Introduction

The night sky brightness, together with number of clear nights, seeing, transparency, photometric stability and humidity, are some of the most important parameters that qualify a site for front-line ground-based astronomy. While there is almost no way to control the other characteristics of an astronomical site, the sky brightness can be kept at its natural level by preventing light pollution in the observatory areas. This can be achieved by means of extensive monitoring programmes aimed at detecting any possible effects of human activity on the measured sky brightness.

For this purpose, we have started an automatic survey of the UBVRI night sky brightness at Paranal with the aim of both getting for the first time values for this site and building a large database. The latter is a fundamental step for the long term trend which, given the possible growth of human activities around the observatory, will allow us to check the health of Paranal's sky in the years to come.

The ESO-Paranal Observatory is located on the top of Cerro Paranal in the Atacama Desert in the northern part of Chile, one of the driest areas on Earth. Cerro Paranal (2635 m, 24$^\circ$40$^\prime$ S, 70$^\circ$25$^\prime$ W) is at about 108 km S of Antofagasta (225,000 inhabitants; azimuth 0°.2), 280 km SW from Calama (121 000 inhabitants; azimuth 32°.3), 152 km WSW from La Escondida (azimuth 32°.9), 23 km NNW from a small mining plant (Yumbes, azimuth 157°.7) and 12 km inland from the Pacific Coast. This ensures that the astronomical observations to be carried out there are not disturbed by adverse human activities like dust and light from cities and roads. Nevertheless, a systematic monitoring of the sky conditions is mandatory in order to preserve the high site quality and to take appropriate action, if the conditions are proven to deteriorate. Besides this, it will also set the stage for the study of natural sky brightness oscillations, both on short and long time scales, such as micro-auroral activity, seasonal and sunspot cycle effects.

The night sky radiation has been studied by several authors, starting with the pioneering work by Lord Rayleigh in the 1920s. For thorough reviews on this subject the reader is referred to the classical textbook by Roach & Gordon (1973) and the recent extensive work by Leinert et al. (1998), which explore a large number of aspects connected with the study of the night sky emission. In the following, we will give a short introduction to the subject, concentrating on the optical wavelengths only.

Figure 1: Night sky spectrum obtained at Paranal on February 25, 2001 02:38UT in the spectral region covered by B, V, R and I passbands (from top to bottom). The original FORS1 1800 s frame was taken at 1.42 airmasses with a long slit of 1 $^{\prime \prime }$ and grism 150I, which provide a resolution of about 22 Å (FWHM). The dashed lines indicate the passband response curves. Flux calibration was achieved using the spectrophotometric standard star Feige 56 (Hamuy et al. 1992) observed during the same night. The absence of an order sorting filter probably causes some second order overlap at wavelengths redder than 6600 Å.

The night sky light as seen from ground is generated by several sources, some of which are of extra-terrestrial nature (e.g. unresolved stars/galaxies, diffuse galactic background, zodiacal light) and others are due to atmospheric phenomena (airglow and auroral activity in the upper Earth's atmosphere). In addition to these natural components, human activity has added an extra source, namely the artificial light scattered by the troposphere, mostly in the form of Hg-Na emission lines in the blue-visible part of the optical spectrum (vapour lamps) and a weak continuum (incandescent lamps). While the extra-terrestrial components vary only with the position on the sky and are therefore predictable, the terrestrial ones are known to depend on a large number of parameters (season, geographical position, solar cycle and so on) which interact in a largely unpredictable way. In fact, airglow contributes with a significant fraction to the optical global night sky emission and hence its variations have a strong effect on the overall brightness.

To illustrate the various processes which contribute to the airglow at different wavelengths, in Fig. 1 we have plotted a high signal-to-noise, flux calibrated night sky spectrum obtained at Paranal on a moonless night (2001, Feb. 25) at a zenith distance of 45$^\circ$, about two hours after the end of evening astronomical twilight. In the B band the spectrum is rather featureless and it is characterised by the so called airglow pseudo-continuum, which arises in layers at a height of about 90-100 km (mesopause). This actually extends from 4000 Å to 7000 Å and its intensity is of the order of 3$\times$10^-7 erg s^-1 cm^-2 Å^-1 sr^-1 at 4500 Å. All visible emission features, which become particularly marked below 4000 Å and largely dominate the U passband (not included in the plots), are due to Herzberg and Chamberlain O₂ bands (Broadfoot & Kendall 1968). In light polluted sites, this spectral region is characterised by the presence of Hg I (3650, 3663, 4047, 4078, 4358 and 5461 Å) and NaI (4978, 4983, 5149 and 5153 Å) lines (see for example Osterbrock & Martel 1992) which are, if any, very weak in the spectrum of Fig. 1 (see Sect. 9 for a discussion on light pollution at Paranal). Some of these lines are clearly visible in spectra taken, for example, at La Palma (Benn & Ellison 1998, Fig. 1) and Calar Alto (Leinert et al. 1995, Figs. 7 and 8).

The V passband is chiefly dominated by [OI]5577 Å and to a lesser extent by NaI D and [OI]6300, 6364 Å doublet. In the spectrum of Fig. 1 the relative contribution to the total flux of these three lines is 0.17, 0.03 and 0.02, respectively. Besides the aforementioned pseudo-continuum, several OH Meinel vibration-rotation bands are also present in this spectral window (Meinel 1950); in particular, OH(8-2) is clearly visible on the red wing of NaI D lines and OH(5-0), OH(9-3) on the blue wing of [OI]6300 Å. All these features are known to be strongly variable and show independent behaviour (see for example the discussion in Benn & Ellison 1998), probably due to the fact that they are generated in different atmospheric layers (Leinert et al. 1998 and references therein). In fact, [OI]5577 Å, which is generally the brightest emission line in the optical sky spectrum, arises in layers at an altitude of 90 km, while [OI]6300, 6364 Å is produced at 250-300 km. The OH bands are emitted by a layer at about 85 km, while the Na ID is generated at about 92 km, in the so called Sodium-layer which is used by laser guide star adaptive optic systems. In particular, [OI]6300, 6364 Å shows a marked and complex dependency on geomagnetic latitude which turns into different typical line intensities at different observatories (Roach & Gordon 1973). Moreover, this doublet undergoes abrupt intensity changes (Barbier 1957); an example of such an event is reported and discussed in Sect. 9.

In the R passband, besides the contribution of NaI D and [OI]6300, 6364 Å, which account for 0.03 and 0.10 of the total flux in the spectrum of Fig. 1, strong OH Meinel bands like OH(7-2), OH(8-3), OH(4-0), OH(9-4) and OH(5-1) begin to appear, while the pseudo-continuum remains constant at about 3$\times$10^-7 erg s^-1 cm^-2 Å^-1 sr^-1. Finally, the I passband is dominated by the Meinel bands OH(8-3), OH(4-0), OH(9-4), OH(5-1) and OH(6-2); the broad feature visible at 8600-8700 Å, and marginally contributing to the I flux, is the blend of the R and P branches of O₂(0-1) (Broadfoot & Kendall 1968).

Several sky brightness surveys have been performed at a number of observatories in the world, most of the time in B and V passbands using small telescopes coupled to photo-multipliers. A comprehensive list of published data is given by Benn & Ellison (1998). All authors agree on the fact that the dark time sky brightness shows strong variations within the same night on the time scales of tens of minutes to hours. This variation is commonly attributed to airglow fluctuations. Moreover, as first pointed out by Rayleigh (1928), the intensity of the [OI]5577 Å line depends on the solar activity. Similar results were found for other emission lines (NaI D and OH) by Rosenberg & Zimmerman (1967). Walker (1988b) found that B and V sky brightness is well correlated with the 10.7 cm solar radio flux and reported a range of $\sim$0.5 mag in B and V during a full sunspot cycle. Similar values were found by Krisciunas (1990), Leinert et al. (1995) and Mattila et al. (1996), so that the effect of solar activity is commonly accepted (Leinert et al. 1998). For this reason, when comparing sky brightness measurements, one should also keep in mind the time when they were obtained with respect to the solar cycle, since the difference can be substantial. A matter of long debate has been the so-called Walker effect, named after Walker (1988b), who reported a steady exponential decrease of $\sim$0.4 mag in the night sky brightness during the first six hours following the end of twilight. This finding has been questioned by several authors. We address this issue in detail later (Sect. 6 and Appendix D).

Here we present for the first time UBVRI sky brightness measurements for Paranal, obtained on 174 nights from 2000 April 20 to 2001 September 23 which, to our knowledge, makes it the largest homogeneous data set available. Being produced by an automatic procedure, this data base is continuously growing and it will provide an unprecedented chance to investigate both the long term evolution of the night sky quality and to study in detail the short time scale fluctuations which are still under debate.

The paper is organized as follows. After giving some information on the basic data reduction procedure in Sect. 2, in Sect. 3 we discuss the photometric calibration and error estimates, while the general properties of our night sky brightness survey are described in Sect. 4. The results obtained during dark time are then presented in Sect. 5 and the short time-scale variations are analysed in Sect. 6. In Sect. 7 we compare our data obtained in bright time with the model by Krisciunas & Schaefer (1991) for the effects of moonlight, while the dependency on solar activity is investigated in Sect. 8. In Sect. 9 we discuss the results and summarize our conclusions. Finally, detailed discussions about some of the topics are given in Appendices A-D.