Site testing at astronomical sites: evaluation of seeing using satellite-based data

Transcription

1 Mon. Not. R. Astron. Soc. 419, (2012) doi: /j x Site testing at astronomical sites: evaluation of seeing using satellite-based data S. Cavazzani, 1 S. Ortolani 1 and V. Zitelli 2 1 Department of Astronomy, University of Padova, Vicolo dell Osservatorio 3, I-35122, Padova, Italy 2 INAF Osservatorio Astronomico di Bologna, via Ranzani 1, I-40127, Bologna, Italy Accepted 2011 October 4. Received 2011 October 4; in original form 2011 July 15 ABSTRACT We present, for the first time, a new method to estimate seeing using remote sounding from the infrared night-time data of the Geostationary Operational Environmental Satellite 12.We discuss the correlation found between the ground-based data and the satellite-derived values from the analysis of the sites located at Cerro Paranal (Chile) and Roque de los Muchachos (Canary Islands, Spain). We obtain a ground satellite correlation percentage of about 90 per cent. Finally, by studying the correlation between the data from the afternoon and the following night, we are able to provide a forecast for the photometric night quality. Key words: turbulence atmospheric effects methods: statistical site testing. 1 INTRODUCTION The ability to optimize scientific requirements to observing conditions is a challenge that must be met in order to improve performances and to increase the final efficiency of the telescope instrumentation system. This is mainly important for very large telescopes. The first parameter needed for this goal is knowledge about usable nights. In the last decades, this evaluation has suffered from biases as a result of personal judgement, because the evaluation has been based on visual inspection. The use of satellite data has made a vast improvement. The second important parameter for site selection and site characterization is image quality. It is well known that image quality affects the scientific quality of the results in many fields of astronomical research. Since the first campaigns for site selection, the criteria have been based simply on a direct analysis of the size of the stellar images. Now, with the increase of knowledge in this area, we know that the seeing is characterized by multiple parameters and it is affected by, or simply linked to, several local and wide-scale conditions, such as the external air temperature and gradients (Lombardi et al. 2006, hereafter Paper I), pressure, wind velocity (Lombardi et al. 2007, hereafter Paper II) and the link between these parameters and optical turbulence (Cavazzani et al. 2011). It is also crucial to know the evolution of the seeing with time in both short and long time-scales for the optimization of the flexible scheduling. This is particularly important for the future giant telescopes. In general, the testing campaigns of the past were expensive and time-consuming, and they were limited to a few pre-selected sites. The use of archive satellite data, instead, is very important because it allows us to simultaneously investigate several sites over many years. Erasmus & van Rooyen (2006) and stefano.cavazzani@unipd.it Erasmus & Sarazin (2002) have recently carried out a quantitative survey of cloud coverage and water vapour content above several astronomical sites, using both satellite and ground-based data. They are among the first to demonstrate that it is possible to use satellite data to obtain the number of useful nights. Della Valle et al. (2010, hereafter Paper III) used a similar analysis and, using independent data, they found that the numbers of clear nights found by satellite and ground-based data at La Palma had an agreement of about 80 per cent. We presented an evolution of this analysis in Cavazzani et al. (2011), where we used a more sophisticated method and introduced the concept of a satellite stable night, which is the best approximation of the concept of photometric nights. In this paper, we present for the first time an estimation of the seeing obtained using satellite remote sounding. We analyse the correlation between ground-based seeing and satellite-based seeing. This analysis is applied to two very important astronomical international sites Cerro Paranal (Chile) and Roque de Los Muchachos (La Palma, Canary Islands, Spain) in order to validate the code in two very different climatic and topographic conditions. The locations of the two sites are shown in Fig. 1. La Palma and Paranal are two sites where the astronomical community has built several facilities, because of the good sky conditions. Moreover, the community is strongly interested in maintaining the high performance of this instrumentation. For this reason, several authors have focused their attention on the characterization of these two sites (Murdin 1985; Sarazin 2004; Varela et al. 2008). The staff at the European Southern Observatory (ESO) pioneered this topic. The long record of data collected at Paranal is a useful tool to analyse the connection between the astrophysical and physical environmental conditions. The differences in the microclimate at La Palma have been discussed in Papers I, II and III. In Paper I, a complete analysis of the vertical temperature gradients and their correlation with the astronomical seeing is given. In Paper II, an C 2011 The Authors

2 3082 S. Cavazzani, S. Ortolani and V. Zitelli Figure 1. The locations of the two sites involved in the analysis are shown. As seen in the insets, the selected sites present very different topographical conditions: La Palma is a sharp island and Paranal is an isolated peak over the Atacama desert. The position of GOES 12 is projected on to the map. The figure also shows a comparison of one matrix at Paranal and La Palma. The deformation is a result of the satellite s observation angle. analysis is given of the correlation between wind and astronomical parameters, as well as the overall long-term weather conditions at La Palma. In Paper III, a statistical fraction of clear nights has been derived using satellite data and a basic approach to test the ability of the satellite to select clear nights. The main conceptual difference between the analyses of Erasmus & van Rooyen (2006) and Paper III is that Erasmus & van Rooyen (2006) used the radio sounding vertical profile temperature as an absolute reference to compare with the brightness infrared temperature measured by the satellite, while we have used relative deviations from the bulk of the data to detect the presence of clouds. In particular, we have selected two bands sensitive to the clouds, and we have plotted one band versus the other. The calibration of the plot gives the statistical fraction of usable nights. The use of the two bands separately is efficient to sense thick clouds, but it has some limitations in the case of partial coverage or thin clouds. For this reason, we have refined the analysis by introducing a new band that is sensitive to the local phenomena and by introducing a mathematical code to correlate the three bands. This analysis discriminates successfully between the changes in airmasses, showing also a first connection with seeing variations, as presented in Cavazzani et al. Table 1. Geographic characteristics of the sites analysed and of GOES 12. The angle of view is obtained using θ = ( LAT) 2 + ( LONG) 2. Site LAT LONG Altitude Angle of view (km) Paranal La Palma GOES (2011). In this paper, in order to better analyse the correlation between satellite reflectivity and ground-based image quality at La Palma and Paranal, we have used ground- and satellite-based data sampling for We have used the Geostationary Operational Environmental Satellite (GOES), in order to have homogeneous results with the previous papers and to make it easier to compare and discuss the results. Table 1 shows the geographical positions and view angles of the satellite for each site. The paper is organized as follows. In Sections 2 and 3, we describe the ground- and satellite-based data, respectively. In Section 4

3 GOES 12 remote sounding 3083 we describe the satellite acquisition procedure, and in Section 5 we describe the mathematical model. In Section 6 we describe the atmospheric correlation function, in Section 7 we describe our approach to detecting small clouds and local perturbations and in Section 8 we describe the night classification obtained from the satellite. In Section 9 we describe the satellite seeing, and in Section 10 we describe the temporal forecasting seeing. Finally, in Section 11, we discuss the results. Table 2. Characteristics of the GOES 12 bands used and the resolution at nadir. Window Passband Resolution (µm) (km) BAND3 H 2 O BAND4 Infrared BAND6 CO GROUND-BASED DATA In this analysis, we compare the satellite data with image quality in terms of FWHM obtained using differential image motion monitoring (DIMM) at the two sites. Data at the Observatorio del Roque de los Muchachos (ORM) are derived from the Robotic Differential Image Motion Monitor (known as RoboDIMM 1 )ofthe Isaac Newton Telescope (INT). The INT RoboDIMM, like all classical DIMMs, relies on the method of differential image motion of telescope subapertures to calculate the seeing Fried parameter r 0. RoboDIMM forms four separate images of the same star, and measures the image motion in two orthogonal directions. From this, it derives four simultaneous and independent estimates of the seeing. The data interpretation makes use of the DIMM algorithm described in Sarazin & Roddier (1990), whichis based on the Kolmogorov theory of atmospheric turbulence in the free atmosphere. At present, we do not have other seeing data to check possible local differences. However, La Palma is the only site that has several DIMMs distributed along the top of the mountain. We are planning to follow this analysis by using all the available DIMM data to check possible local differences. In this way, we hope to obtain a better characterization of the site and to correlate direct measurements, such as Cn 2 (h), with satellite seeing. The seeing data at Paranal are obtained by measuring the seeing of the DIMM at the Very Large Telescope (VLT) observatory. The file also contains measurements of the flux of a reference star, and thus the flux of the star can trace the presence of clouds. The ground-based classification of night quality has been obtained using the nightly observing log of each telescope. 3 SATELLITE-BASED DATA In this analysis, we have used GOES because it can detect the infrared night-time emitted radiation, which allows us to compare the ground- and satellite-based data simultaneously. A detailed discussion of the performance of this satellite is presented in Cavazzani et al. (2011). The main advantage of GOES with respect to other satellites is that GOES is able to observe the full Earth disc, and it has on board an imager with five channels, allowing the collection of five simultaneous images of almost half of the Earth s hemisphere. We choose to use the infrared channels because they allow the detection of the thermal radiation emitted during the night from different atmospheric layers and/or from the soil. An appropriate choice of the wavelength allows us to choose the optimal layer emission height above the site. If this occurs well above the soil surface, the signal becomes independent of the specific soil properties and of low-level conditions. Phenomena occurring below the selected site (such as fog, low clouds, etc.) are also avoided. At some sites (e.g. La Palma), this aspect is of crucial importance. 4 SATELLITE DATA ACQUISITION For the purposes of this work, we used GOES 12 equipped with the imager. We have analysed the year We have selected the water vapour channel (B3 band) centred at 6.7 µm, which can detect high-altitude cirrus clouds, the infrared channel (B4 band) centred at 10.7 µm, which can detect middle-level clouds, and the CO 2 band (B6 band) centred at 13.3 µm, which can sense small particles, such as fog, ash and semitransparent high clouds. We selected the infrared channels in order to detect clouds at different heights, because water vapour absorbs electromagnetic radiation and then re-emits it at various wavelength bands, in particular in the infrared region at 6 7 µm. All the data obtained are measurements of the thermal radiation emitted during the night by the Earth and received by the satellite detector. If clouds are not present, the emissions at 10.7 µm that reach the satellite are largely not absorbed by the atmosphere, so the measured radiance values are a result of emission from the surface of the ground. However, when clouds are present, the emissivity drops because the ground radiation is blocked. Data are prepared by the Comprehensive Large Array-data Stewardship System (CLASS) 2 and are processed using the freeware software MCIDAS-V-1.0-BETA4. For each site, we have identified and extracted a subimage of 1 1, with the central pixel close to the coordinates given in Table 1. The use of the matrix is justified by the high correlation with the single pixel. We have previously described this correlation (Cavazzani et al. 2011). Here, we report the correlation coefficient values for 2009: the matrix correlation coefficients at Paranal and La Palma are 0.97 and 0.93, respectively. The use of the matrix reduces the satellite noise and also allows us to observe the entire sky above the site. Table 2 shows the main characteristics of the selected bands. For each night, we have extracted the observations at different hours in local time: at 17:45, 20:45, 23:45, 02:45, 05:45, 7:45, 8:45 and 9:45. The evaluation of the amount of useful hours is carried out by using observations for the whole night except for those at 17:45 and 9:45. We have used the brightness temperature at both 17:45 and 9:45 to check for a possible day night correlation. The last column of Table 1 shows the satellite view angle. The insets in Fig. 1 show the two different projections obtained from each acquisition at La Palma and Paranal. 5 THE CODE In Cavazzani et al. (2011), there is an exhaustive description of the mathematics approach. Here, we summarize the main definition for a complete exposition. The emitted monochromatic radiation intensity at a given λ and along a vertical path at the top of the atmosphere, incident at a 1 See 2

4 3084 S. Cavazzani, S. Ortolani and V. Zitelli satellite instrument, is given by R λ = (I 0 ) λ τ λ (z 0 ) + B λ [T (z)] K λ (z)dz. (1) z 0 Here, K λ (z) = dτ λ (z)/dz is the weighting function (WF), B λ [T(z)] is the Planck function profile as a function of the vertical temperature profile T,(I 0 ) λ is the emission from the Earth s surface at height z 0 and τ λ (z) is the vertical transmittance from height z to space. For a viewing path through the atmosphere at angle θ to the vertical, we have [ ] τ λ (z, θ) = exp secθ κ λ (z)c(z)ρ(z)dz. (2) z Here, ρ(z) is the vertical profile of the atmospheric density, κ λ (z)is the absorption coefficient and c(z) is the absorbing gas mixing ratio. The emitted radiation intensities in each considered band λ are then R λ3 = (I 0 ) λ3 τ λ3 (z 0 ) + B λ3 [T (z)] K λ3 (z)dz, R λ4 = (I 0 ) λ4 τ λ4 (z 0 ) + z 0 z 0 B λ4 [T (z)] K λ4 (z)dz R λ6 = (I 0 ) λ6 τ λ6 (z 0 ) + z 0 B λ6 [T (z)] K λ6 (z)dz. Each band considered is characterized by a WF that gives the variation of the efficiency of the system as a function of the height. The peak of the efficiency specifies the layer from which the radiation is emitted, and then the region of the atmosphere, which can be sensed from space at fixed λ. Assuming a standard atmosphere, the GOES 12 WFs assign the following median height values to each band: 3 BAND3 BAND4 K λ3 (z) = dτ λ 3 (z) dz K λ4 (z) = dτ λ 4 (z) dz 8000 m; 4000 m; BAND6 K λ6 (z) = dτ λ 6 (z) 3000 m. dz The elevation assigned by the WF depends on the location of the selected sites, but we can assume that these values can be assigned to both sites that we are interested in, which have heights ranging between 2 and 3 km. 6 ATMOSPHERIC CORRELATION FUNCTION Instead of using each band separately in this analysis, we have introduced a code to correlate the three bands. The correlation function F C.A. (t) isgivenby F C.A. = I λ3 ( ) I λ6 I λ4. (3) In mathematical terms, this model provides the brightness temperature of the B3, B4 and B6 combination, given by F C.A. = R λ 3 + R λ4 R λ6 τ(z 0 ) z 0 B λ3 [T (z)]k λ3 + B λ4 [T (z)]k λ4 B λ6 [T (z)]k λ6 dz. τ(z 0 ) (4) 3 See Figure 2. Atmospheric correlation function at Paranal for 2009 August. The top panel shows the monthly plots of the three bands.the central panel shows F C.A. for August, where the solid gray line is the F C.A. (t) linear regression. The brightness temperature is expressed as the number of satellite counts, extracted using MCIDAS-V. The bottom panel shows the distribution of clear and stable nights as a function of the sensed height. The physical meaning of this model is that the brightness temperature of the atmosphere reaches the satellite sensor, as shown by the combination of the B3, B4 and B6 bands. This is shown by equation (4). Therefore, F C.A. provides information about the atmospheric evolution of the surveyed site. Moreover, F C.A. provides information about both the height and quality of the perturbations over the surveyed site, which are both a function of the T brightness. Fig.2showstheF C.A. (t) obtained at Paranal for 2009 August. The top plot in Fig. 2 shows the monthly trend of the three bands used. The central plot shows the F C.A. of August obtained from the three bands, where the solid grey line is the F C.A. (t) linear regression. The bottom plot shows the distribution of clear and stable nights, as discussed in Section 8. 7 DETECTION OF SUBTLE PHENOMENA Here, we describe two different algorithms that are introduced to detect perturbations in two cases: low-level perturbations located spatially very close to the telescope or very far from the telescope (located at the wedge of the matrix area) this means that we have an incoming perturbation. 7.1 Detection of small clouds in the matrix area Della Valle et al. (2010) obtained the reflectivity flux from the pixel of the matrix centred close to the coordinates of the specific site.

5 GOES 12 remote sounding 3085 Figure 3. Example of a low standard deviation of the pixel array. A high value of I RS shows the presence of a perturbation, even at a low level of signal-to noise ratio, which is not detectable using the simple standard deviation and the matrix average because of the high number of averaged pixels. In particular, we use the following mathematical classifications: clear, I λ4 I λ4 (1 pixel) 2σ ; subtle phenomena, I λ4 I λ4 (1 pixel) > 2σ. Fig. 5 present an example of the average corresponding to clear nights. In this case, the matrix/1 pixel remote sounding (RS) shows the presence of local stationary phenomena that are not detected by the mean value of the matrix. By using both the matrix/1 pixel and the standard deviation for all the data, we can better detect local phenomena and thin clouds, as shown in Fig. 6 for La Palma. In fact, it can be seen that the plot of the B4 band, shown by circles in the upper panel, is flat, typical of a clear sky. In the bottom panel of Fig. 6, the plot of the matrix/1 pixel difference shows variations that indicate the presence of local phenomena. Checking with the logbooks, we find the presence of high humidity and ice on these Figure 4. Example of a high standard deviation of the pixel array. With the aim of reducing the instrumental noise and looking for a wider field of view, we decided to replace the one-pixel flux reflectivity with the mean value of the 1-deg matrix even centred at the coordinates of the specific site. Moreover, to better discriminate small clouds distributed in the matrix area, which are missed when considering the limitation of the model, as described in the previous section, we computed the standard deviation of each matrix. In fact, a high standard deviation signifies the presence of perturbations in the wall area. We can also see incoming clouds approaching the edge of the matrix area. Figs 3 and 4 show two examples where the average value of the matrix in both figures corresponds to clear nights. However, the standard deviation of Fig. 4 is high, showing a non-real flat distribution of the satellite counts. This is the case of incoming perturbations to the telescope site. Considering the standard deviation, we obtain the following classifications: clear, standard deviation (I λ4 ) 2σ ; subtle phenomena, standard deviation (I λ4 ) > 2σ. Figure 5. Remote sounding between the average and single pixels. An example is shown of the large difference between the average and single pixels. 7.2 Detection of local phenomena Finally, in order to better detect the presence of local phenomena close to the telescope, we introduce the difference between the mean matrix reflectivity and the single pixel reflectivity using I RS (matrix/1 pixel) = I λ4 I λ4 (1 pixel). (5) Figure 6. The trend of the B4 band (upper panel) shows no indications of local phenomena, while the plot of RS matrix/1pixel (bottom panel) does indicate the presence of local phenomena. In fact, the logbooks describe the presence of high humidity and ice. The brightness temperature is expressed as the number of satellite counts, extracted using MCIDAS-V.

6 3086 S. Cavazzani, S. Ortolani and V. Zitelli Table 3. Satellite mean monthly percentage at Paranal and La Palma in Month Paranal La Palma Clear time Subtle phenomena Clear time Subtle phenomena January February March April May June July August September October November December Mean Figure 8. Temporal distribution of the GOES 12 B4 and B6 band emissivity at Paranal in The classification of sky quality has been carried out using the Paranal log. Figure 9. Temporal distribution of the GOES 12 B4 and B6 band emissivity at La Palma in The classification of sky quality has been carried out using the La Palma log. Figure 7. Subtle phenomena at La Palma in nights. We stress that these cases are rare. In fact, Table 3 shows the statistical results of this analysis for 2009 at Paranal and La Palma, giving the mean monthly percentage of clear nights and the fraction of clear nights with low-level phenomena. We see that only 1 per cent of the 91 per cent of clear nights at Paranal are affected by low-level phenomena, compared with 3 per cent of the 67 per cent at La Palma. In both cases, this is a very low number. At Paranal, in May, there are a high number of subtle phenomena. Checking with the logbooks, we find that there was a high wind coming from the sea, which means high humidity, thus justifying the high value given by the satellite. Fig. 7 shows the how many clear nights there were at La Palma in 2009; the grey shading shows the percentage of subtle phenomena. 8 SATELLITE TEMPORAL CLASSIFICATION To obtain a reliable prediction of the night quality, we have used a high temporal resolution for each night using the series of data for the following times: 20:45, 23:45, 2:45, 5:45, 7:45 and 8:45. Using the brightness temperature obtained for the hours considered, we have obtained the monthly atmospheric correlation function. Figs 8 and 9 show the obtained temporal emissivity of the B4 band versus B6 for 2009 at Paranal and La Palma, respectively. Nights are classified according to the comments found in the observing logs. Clear time presents high values of emissivity at both sites. As in Paper III, the classification of satellite time quality is done by assuming that the maximum monthly brightness temperature in the B4 band (Tb max ) occurs in clear conditions. The other hourly brightness temperatures are correlated with Tb max using the following classifications: clear, Tb max T b 2σ ;mixed,2σ < Tb max T b 3σ ; covered, Tb max T b > 3σ. Here, T b is the brightness temperature of the 1 1 matrix. Table 4 shows the obtained percentages of clear, mixed and covered nights at Paranal and La Palma for 2009 using all the algorithms previously described. The ground-based classification is derived from the comments found in the night logbook. We found a very good agreement for both sites between groundand satellite-based data. The last row of Table 4 shows the percentage of accuracy associated with each obtained fraction of nights. The uncertainty is computed as follows: clear/mixed is clear/mixed uncertainty; clear/covered is clear/covered uncertainty; mixed/covered is mixed/covered uncertainty. Usually, the quality of clear nights obtained using ground-based data is divided between photometric and spectroscopic nights. Also, for the satellite classification, we have introduced a similar definition introducing the concepts of stable night (photometric) and clear night (spectroscopic). Considering the value of the F C.A. (t) linear regression Tb trendline, we define stable, T b Tb trendline 1σ ; clear, 1σ < T b Tb trendline 2σ ; covered, T b Tb trendline > 2σ. Here, Tb trendline is the brightness temperature of the F C.A. (t) linear regression computed in one month and T b is the brightness temperature of the 1 1 matrix in 1 h. Table 5 shows the mean monthly percentage of clear and stable time at Paranal and La Palma, obtained using satellite-based data. Figs 10 and 11 show the distribution of clear, stable and covered time at Paranal and La Palma for 2009, obtained from F C.A. (t). The maximum of the distribution shows the sensed height, and it gives the height at which the atmospheric phenomena occur. Fig. 12 shows the monthly distribution of clear and stable nights at Paranal

7 Table 4. Temporal data analysis of clear/mixed/covered time at Paranal and La Palma in GOES 12 remote sounding 3087 Site Ground Satellite Clear Mixed Covered Clear Mixed Covered Paranal 90.1 per cent 2.2 per cent 7.8 per cent 90.8 per cent 2.6 per cent 6.6 per cent La Palma 65.8 per cent 5.0 per cent 29.3 per cent 67.0 per cent 4.5 per cent 28.5 per cent Paranal La Palma Uncertainty clear/mixed clear/covered mixed/covered clear/mixed clear/covered mixed/covered Percentage 1.2 per cent 0.4 per cent 0.8 per cent 1.3 per cent 0.5 per cent 0.8 per cent Table 5. Satellite mean monthly percentage for Month Paranal La Palma Clear time Stable Clear time Stable January February March April May June July August September October November December Mean Figure 11. Histogram of the annual atmospheric stability at La Palma. The light-grey bars represent the stable nights, the grey bars represent clear but unstable nights and the black bars represent covered nights. Figure 10. Histogram of the annual atmospheric stability at Paranal. The light-grey bars represent the stable nights, the grey bars represent clear but unstable nights and the black bars represent covered nights. for We see that during the winter months the percentage of stable time is low and the nights are mostly clear. 9 SATELLITE CALCULATION OF SEEING In Cavazzani et al. (2011), we have shown that the adopted code is able to successfully discriminate variations of the atmospheric stability function (F C.A. (t) ) from optical turbulence, showing the first connection between F C.A. (t) and seeing. In this paper, we analyse this in more depth. In order to better analyse the correlation between satellite reflectivity and ground-based image quality at La Palma and Paranal, we have used ground- and satellite-based data sampling the year In particular, we introduce for the first time the concept of satellite seeing. The F C.A. (t) measures the temperature in different atmospheric layers. Also, because the ground-based CT 2 is linked to r 0 and to the FWHM, it is possible to derive a satellite-based CT 2, Figure 12. Fractions of clear and stable nights at Paranal in 2009 obtained from GOES 12. and consequently Cn 2. The zero-point is given empirically in this analysis. Using the basic equations of seeing theory, such as Fried s radius r 0,wehave r 0 = [ π 2 ] (3/5) 1 C λ 2 n 2 cos(θ zen ) dz. (6) Here, Cn 2 is the refractive index structure parameter: [ Cn 2 = P ] CT 2 T. The FWHM is given by FWHM = 0.98 λ r 0. (7) The FWHM of the satellite is obtained through our empirical model. If we assume Tb Tb trendline C 2 T Cn 2, then we can replace the Cn 2 value in equation (6) and obtain a satellite

8 3088 S. Cavazzani, S. Ortolani and V. Zitelli Table 6. Mathematical and statistical uncertainties of the model at Paranal and La Palma in Site total N(G; S) statistical Paranal 1.4 per cent per cent La Palma 1.5 per cent per cent calculation of r 0 : [ r 0,sat = π ] 2 λ (θ) T b Tb trendline (3/5). (8) 2 z Finally, using this value we obtain the following equation for the satellite FWHM: ] T FWHM sat = 0.58λ [4π (1/5) 2 b Tb trendline 3/5 (θ). (9) z Here, (θ) is an empirical constant defined by (θ) = cos θ, (10) where θ is the satellite s angle of view. Fig. 13 shows the comparison between the ground- and satellitebased FWHM computed for the same hours. We note the very good agreement between the two set of data. Fig. 14 shows the dispersion of this correlation and its linear regression. A tentative physical interpretation of our correlation can be related to the Richardson number R i, which is dependent on the vertical temperature gradient. Tables 7 and 8 show comparisons between the seeing given by ground-based data and computed by satellite using equation (9). We make the following observations of the values obtained. The ground values at Paranal are the DIMM data and not the VLT values. At La Palma, we have calculated the correlation coefficients only for the months in which the RoboDIMM gives us values for more than ten nights. Moreover, the mean seeing only refers to the clear time, because the RoboDIMM does not work on cloudy nights. 10 TEMPORAL FORECASTING SEEING ANALYSIS In this section, we analyse for the first time the possibility of giving a forecasting value of the seeing a few hours before starting the observations. We have proceeded in two different ways to check the capability and the best procedure. In the first test, we have correlated the brightness temperature obtained from the value at 9:45 with the brightness temperature obtained using the values of the previous night. In the second test we have correlated the brightness temperature obtained from the afternoon value at 17:45 with the values from Table 7. Satellite FWHM at Paranal for Month sat ground Corr. coeff. Figure 13. Correlation between ground- and satellite-based data, comparing the FWHM calculated from the ground and the satellite FWHM, for Paranal in 2009 January (correlation coefficient is 0.91). The satellite FWHM is calculated using equation (9). January February March April May June July August September October November December Table 8. Satellite FWHM at La Palma for We have calculated the correlation coefficients only for the months in which the RoboDIMM gives us values for more than ten nights. Moreover, the mean seeing only refers to the clear time because the RoboDIMM does not work on covered nights. Month sat ground Corr. coeff. Figure 14. Correlation between ground- and satellite-based data for Paranal in 2009 January. The figure shows the dispersion of this correlation and its linear regression (correlation coefficient is 0.91). January February March April May June July August September October November December

9 GOES 12 remote sounding 3089 Table 10. Meteorological changes at Paranal for Month 5 p.m. 6 a.m. 9 p.m. 10 a.m. 10 a.m. 5 p.m. January February March April May June July August September October November December Table 11. Meteorological changes at La Palma for Month 5 p.m. 6 a.m. 9 p.m. 10 a.m. 10 a.m. 5 p.m. Figure 15. GOES 12 emissivity in the B3, B4 and B6 bands (upper panel) at Paranal in 2009 January. The bottom panel shows the B4 and B6 vertical scale zoom. The brightness temperature is expressed as the number of satellite counts, extracted using MCIDAS-V. the same night. Fig. 15 shows the monthly distribution of the three bands at Paranal for 2009 January (upper panel) and the zoom (bottom panel), in which is possible to see the new day point used for the forecasting seeing. We see that, for high day values, the following night is stable (photometric night). It is interesting to note that in this analysis we are able to give a percentage of useful nights instead of useful time. Table 9 shows the monthly values of the derived correlation at the two sites. Column 1 shows shows the month, column 2 shows the number of days used, columns 3 and 5 show the afternoon to night correlation (A N is the correlation between the afternoon and the following night) and columns 4 and 6 show the night to morning correlation (N M is the correlation between the morning and the previous night). For our analysis, we are interested in columns 3 and 5, which give the correlations of all the available day night data. We see that at both Paranal and La Palma, the correlation decreases during the winter months. Tables 10 and 11 show the period of the day in which the meteorological variation occurred. In both tables, column 1 shows January February March April May June July August September October November December the month, column 2 shows the percentage of the variation that occurred between 5 p.m. and 6 a.m., column 3 shows the variation that occurred between 9 p.m. and 10 a.m., column 4 shows the percentage obtained for the part of the day that was not analysed (10 a.m. 5 p.m.) obtained from the difference. These numbers are obtained through the percentage of clear time (Table 5) and the correlation percentages. For example, if we have clear time of 70 per cent, then covered time is 30 per cent. Thus, we have an A N correlation of 95 per cent and an N M correlation of 90 per cent. This means that 5 per cent of the meteorological changes occurred between 5 p.m. and 6 a.m., 10 per cent between Table 9. Forecast at Paranal and La Palma for A N is the correlation between the afternoon and the following night and N M is the correlation between the morning and the previous night. Paranal La Palma Month Days A N correlation N M correlation A N correlation N M correlation January February March April May June July August September October November December

10 3090 S. Cavazzani, S. Ortolani and V. Zitelli Figure 16. The trend of F C.A. (t). We have highlighted the new points used for the forecast of night quality (forecast points). Using the positions of these points, we can predict whether the night will be stable (stable time) or clear (clear time). In fact, the figure also shows the DIMM FWHM values at Paranal (shown in black). We note that at these stable time points we have low seeing values; in contrast, we have high seeing values at clear time points (the y-axis on the right shows the DIMM FWHM values). The brightness temperature is expressed as the number of satellite counts, extracted using MCIDAS-V. Figure 17. Fried s radius values calculated by satellite (r 0,sat ) at Paranal in 2009 January. The satellite FWHM calculated with these values has a correlation coefficient of 0.91 with the ground FWHM (see Fig. 13). 9 p.m. and 10 a.m. and the remaining 15 per cent between 10 a.m. and 5 p.m. We can see that most of the changes occur during the day (from 10 a.m. to 5 p.m.), so it is possible to correlate the afternoon satellite data with the satellite data from the following night. Fig. 16 shows the trend of F C.A. (t) obtained using all the brightness values and the DIMM seeing. The grey line is the best fit of the monthly plot. We have highlighted the new points used for the night quality forecast (forecast points). Using the positions of these points, we can predict whether the night will be stable (stable time) or clear (clear time). Fig. 17 shows the values of r 0 obtained by satellite. These values are obtained using the model described in Section 9 (equation 8). Moreover, the value of r 0 is computed by taking into account the height of Paranal, and it is given in the visible range. We obtained values close to those obtained using ground-based data. In fact, the FWHM calculated using the value of r 0,sat has a high correlation coefficient with the ground-based FWHM (see Tables 7 and 8 and Fig. 14). 11 CONCLUSION In this paper, we have introduced the concept of satellite seeing using remote sounding from the infrared night-time data from GOES 12. We have discussed the correlation derived between the groundbased data and the values derived from the satellite from an analysis of the sites located at Cerro Paranal (Chile) and Roque de Los Muchachos (Canary Islands, Spain) for In this analysis, we have used F C.A. (t), which is obtained by correlating the monthly mean values from a 1-deg matrix of each of the three selected bands. This function, which is a measure of the temperature gradient of the three layers sampled by the three bands, is linked to r 0, as well as the ground-based CT 2. The values of r 0 derived using F C.A. (t) (see Section 9) at Paranal and La Palma are close to the groundbased values. In particular, the FWHM calculated using r 0,sat has a high correlation coefficient with the ground-based FWHM (see Tables 7 and 8 and Fig. 14). In this first analysis, we have obtained empirically the zero-point using the DIMM seeing from each site. We have demonstrated that the plot of the seeing from the satellite is in good agreement with the DIMM seeing of the same month (see Fig. 13), showing at Paranal a monthly correlation ranging between 80 and 97 per cent. Fig. 14 shows an example of the dispersion of this correlation and its linear regression for 2009 January at Paranal. We found a better correlation at La Palma (89 95 per cent), because the correlation only refers to the clear time (the RoboDIMM does not work on cloudy nights). We cannot give a fixed zero point of the seeing value, but we intend to refine our procedure. As a further step, for the first time, we give the forecasting seeing from the satellite (see Fig. 16). We have used two ways to select the best procedure. In the first test, we correlated the brightness temperature of the morning value at 9:45 with the values of the previous night. In the second test, we correlated the brightness temperature of the afternoon value at 17:45 with the following night. The two procedures seem to show similar results, with a marginally higher percentage in the night morning values for both sites. However, to predict the image quality for the upcoming observing night, we can use the afternoon night correlation. At Paranal, we can see that the correlation decreases during the winter months, whereas we found a more homogeneous distribution at La Palma. Using this afternoon night relationship, we can estimate the photometric night quality. In fact, in Section 9, ground we have demonstrated a high correlation between sat and (see Tables 7 and 8 and Fig. 14). In addition, in Section 10 we have shown how the afternoon data are correlated with the night data (see Table 9, column 3).

11 GOES 12 remote sounding 3091 With these results, we have a model that can provide a satellite calculation of seeing and a forecast. The results shown in Tables 10 and 11 are interesting. These tables show the monthly percentages of the changes in the observing conditions at the two sites during We have found that at Paranal variations in the meteorological conditions occur during the day, but for February, March and April they occur between 5 p.m. and 9 p.m. During the night, the weather is almost stable. At La Palma, we have shown that the variations occur during the day. ACKNOWLEDGMENTS This activity is supported by the European Community (Framework Programme 7, Preparing for the Construction of the European Extremely Large Telescope, Grant Agreement Number ) and by the Strategic University of Padova funding under the title Quantum Future. Most of the data used in this paper are based on CLASS, which is an electronic library of NOAA environmental data. The CLASS website 4 provides opportunities for finding and obtaining data, particularly data from the NOAA GOES. Finally, we acknowledge the staff responsible for the website of the Liverpool Telescope. REFERENCES Cavazzani S., Ortolani S., Zitelli V., Maruccia Y., 2011, MNRAS, 411, 1271 Della Valle A., Maruccia Y., Ortolani S., Zitelli V., 2010, MNRAS, 401, 1904 (Paper III) Erasmus D., Sarazin M., 2002, in Vernin J., Benkhaldoun Z., Munoa- Tunon C., eds. ASP Conf. Ser. Vol. 266, Astronomical Site Evaluation in the Visible and Radio Range. Astron. Soc. Pac., San Francisco, p. 310 Erasmus D., van Rooyen R., 2006, in Stepp L. M., ed., Proc. SPIE, 6267, Ground-based and Airborne Telescopes. SPIE, Bellingham, p Lombardi G., Zitelli V., Ortolani S., Pedani M J., 2006, PASP, 118, 1198 (Paper I) Lombardi G., Zitelli V., Ortolani S., Pedani M., 2007, PASP, 119, 292 (Paper II) Murdin P., 1985, Vista Astron., 28, 449 Sarazin M., 2004, The VLT Astronomical Site Monitor ( gen-fac/pubs/astclim/paranal/asm/verif/20years-climatologyofparanal- Oct2004.pdf) Sarazin M., Roddier F., 1990, A&A, 227, 294 Varela A. M., Bertolin C., Munoz-Tunon C., Ortolani S., Fuensalida J. J., 2008, MNRAS, 391, This paper has been typeset from a TEX/LATEX file prepared by the author.