Identification of Potential Sites for Astronomical Observations in Northern South America

The use of images of the earth in different wavelengths from infrared cameras on board geostationary satellites is a powerful tool for the identification of new potential sites for astronomical observations in optical, infrared, and submillimeter wavelengths (Erasmus & Sarazin 2002; Cavazzani et al. 2012; Cavazzani & Zitelli 2013). Sites in remote areas with complex relief and scarce or null meteorological records can be identified by previous computation of the clear-sky fraction (CSF). As the cloud covering is a function of the height of the atmosphere, images from satellites are useful due to the detectors on board that measure the emissivity of the atmosphere at different layers.

Several studies for the determination of the CSF are found in the literature. Soden & Bretherton (1993) conducted an analysis of upper tropospheric humidity and clouds using GOES observations, finding a correlation between humidity and cloud cover at high altitudes. Erasmus & Sarazin (2002) were among the first to successfully use long-term satellite data for CSF computations for a wide range of altitudes. Erasmus & van Rooyen (2006) and Della Valle et al. (2010) reproduced with high precision the CSF records of ground data at La Palma with a model based on satellite data. Cavazzani et al. (2011, 2012) found percentages of clear nights at five astronomical observatories around the world using long-term GOES data. Hidayat et al. (2012) used 15 years of GOES infrared images in 6.7 and 10.7 μm bands to compute the percentage of clear sky over a wide region in Indonesia. Recently, Marín et al. (2015) computed the CSF at high-altitude sites by using a new clearance algorithm based on global meteorological models. CSF values were then used for precipitable water vapor estimations at two astronomical sites in the north of Chile.

Even though most of the methods are based on the comparison of radiances, fluxes, or temperatures at each pixel of the satellite image with threshold values previously defined there is no standard methodology for CSF computations. Cavazzani & Zitelli (2013) used monthly thresholds defined as the maximum radiance at clear nights along the month. Thresholds may also be defined by using an atmospheric model (9). In this study, we introduce a novel methodology involving the use of monthly thresholds of temperature at different altitudes from the surface up to the upper troposphere. Our thresholds definitions were based on the altitude of the terrain, long-term records of weather stations, and radiosonde data archives.

The thresholds definitions are based on monthly mean records of both air temperature at the site and vertical temperature profile between 8 and 10 km of altitude, which makes the predictions for clear covering more precise than those involving atmospheric models for thresholds computations. Our methodology was applied in northern South America over the tropical region covering Colombia, western Venezuela, and northern Ecuador. Even though the zone is far from being interesting for infrared and submillimeter astronomy due to the requirement of excellent weather conditions, many universities and institutes from those countries are planning to develop optical observatories. The countries are located within the intertropical convergence zone (ITCZ) showing complex topography and with a climate governed by a bimodal annual cycle which results of the double passage of the ITCZ for the region (Poveda et al. 2011). The Andes affect the climate by blocking zonal flows, influencing regional wind patterns and precipitation, and hosting diverse microclimates. The main goal of this study is to seek prospective locations with middle quality CSF values over the Andes of those countries.

The paper is organized as follows. In § 2, we describe the satellite data used in this study. Threshold temperatures calculations are presented in §§ 3 and 4. The methodology and validation are discussed in §§ 5 and 6, while in §§ 7 and 8, we present the results and the selected sites.

The GOES 12 and GOES 13 satellites are equipped with instrumentation capable of capturing images in several infrared bands. For our purposes, we concentrate our attention on the B3 and B4 bands centered at 6.7 and 10.7 μm, respectively. Images in B4 are strongly affected by emission and absorption of water vapor found at middle altitudes (∼4 km), whereas layers of the upper troposphere between 8 and 10 km are sensed by images in the B3 band. Other bands as those centered in 13.3 μm provide information about fog and very low atmosphere characteristics.

Since the data extend for decades, these satellites are commonly used for long-term studies. The temporal domain for this study begins on 2008 January 1 and ends on 2012 December 31 with missing data lower than 7% in the worst cases. GOES missing data have different causes: technical troubles in GOES-12 (2008 December 15–2009 January 6); when GOES-13 replaced GOES-12 (2010 April 14–28); and due to anomalies in the operational system (2012 September 23–October 18). We assumed the clear-sky fraction as null for the days with GOES missing data.

GOES-12 and GOES-13 images are freely available through an electronic library of the National Oceanic and Atmospheric Administration (NOAA) in format GVAR (GOES variable format). We selected a region of the image to download covering a large area of Colombia, the western part of Venezuela and the north of Ecuador. For each night, we used available observed time at the three different hours 00:15:15, 06:45:13, and 09:45:14 UTC in order to cover the entire night. GVAR is an instrumental radiance given in counts. We used the library of idl routines provided by the National Climatic Data Center (NCDC) for converting the counts to radiance at each pixel. Usual calibration of the images was done using the scaling bias and gain values available from the NCDC website ¹ in order to get the radiance in mW/[m² sr cm^-1].

Brightness temperatures at the instants i = 1, 2, 3 of the night are computed from calibrated radiance values by assuming black body emission. Conversion to actual temperatures is done using the relation , where α and β are conversion coefficients dependent on each band (Weinreb et al. 1997; Rossow et al. 1985). Reported values available from NCDC website, for the GOES-12 satellite are β = 1.01 and α = -4.76 K for the B3 band and β = 1.00 and α = -0.36 K for the B4 band. The differences between the actual and brightness values increase with decreasing temperature but usually are lower than 4%. For instance, a brightness temperature of 10,000 K corresponds to temperatures of 1095 K and 9999.6 K for the B3 and B4 bands, respectively. For most of cases, those differences are negligible for the CSF computations. We indicate actual temperatures in B3 and B4 bands as T_B3,i and T_B4,i , respectively.

The spatial resolution of GOES images is 4 × 4 km in both bands, and thus a single pixel subtends only a small field-of-view from the ground compared with the total field-of-view available for an observer at ground. As the validation of our method involves comparison with log-books of the Observatorio Nacional de Llano del Hato in Venezuela (hereafter, CID), we degraded the original resolution in order to make more real comparisons. Other authors have used similar or even lower resolutions. Cavazzani et al. (2012) computed GOES temperatures for each site over a 1° × 1° submatrix which corresponds to a linear projection of about 100 × 100 km centered on the coordinates observatories at ground. We used a Delaunay interpolation of actual temperatures T_B3,i and T_B4,i on a grid of 51 × 51 pixels obtaining a resolution of 28 × 18 km which we consider wide enough for validation with ground data.

Detection of midaltitude clouds at a height of 4000 m above a pixel location of the image in B4 band and at a fixed instant during the night requires the previous evaluation of a threshold or reference temperature. In this study, we compute monthly thresholds of temperature based on long-term records of weather stations located inside the region covered by the GOES B4 and B3 images.

We used long-term records of meteorological data obtained from the Instituto de Hidrología, Meteorología y Estudios Ambientales, hereafter IDEAM. The IDEAM data archive comprise periods from 7 to 40 years of monthly mean values of temperature and precipitation for a set of weather stations in Colombia. The complete archive contains 400 records at altitudes between 400 and 4700 m above sea level covering 25% of the country. The archive presents a high number of gaps, especially those of temperature. We considered only long-term records of temperature with a maximum of one monthly absence obtaining the reduced number of 63 weather stations irregularly distributed in a region comprising latitudes between 3° and 7° North, and longitudes from -76° up to -72° West. We will refer to this region as the control region, defined by the spatial distribution of the IDEAM long-term historical records. The longest records of these selected stations have more than 40 yr as is shown in the right panel of Figure 1. Locations are widely distributed in different altitudes with a high number of them close around Bogotá (BOG) (h = 2600 m) as shown in the left panel of the same figure.

Zoom In Zoom Out Reset image size

The average of monthly mean records at each station for periods spanning at least 7 years is expected to be normally distributed around a central value 〈T_s 〉 being very likely that the monthly mean values are within 3σ_s of the distribution, where σ_s is the standard deviation. The coldest temperature at each station may be then defined as T_min = 〈T_s 〉-3σ_s . However, an additional correction term must be applied since the measurements do not represent values during the night. We recall that the IDEAM archive consists only of mean temperatures, thus we must correct them for night-time observations. The monthly threshold temperature at each station may therefore computed as:

where T_min ,tr is the trend of the minima records of temperature given by short-term records of weather stations indicated with a solid line in Figure 2 and represents the correction for nighttime observations. We note that these thresholds correlate very well with the altitude of the station as illustrated in the same figure, suggesting that monthly threshold temperatures may be completely determined from the altitude of the stations. From linear fits for each month we established monthly relationships between threshold and altitude of the form: T₀(h) = A × h + B where A and B are constants for each month with values indicated in Table 1. Slight differences along the year are observed for these two parameters with slope or lapse rates variations of the order of 10^-1°C and negligible variations in the intercepts.

Zoom In Zoom Out Reset image size

3.1. Kriging Interpolation of Altitudes

Thresholds in the control region at locations where no weather stations are available require interpolation of the altitudes. We used the geostatistical method called Kriging, which is a spatial interpolation method that in its simplest form allows to compute altitudes at locations with no data, based on the spatial autocorrelation. This means that two close stations have similar values of altitude whilst two stations farther apart do not. Briefly, the height at the location s₀ referred as is given by a weighted combination of the IDEAM all records of altitude:

where λ₀ is the vector of Kriging weights w_i and h is the vector of available samples of altitudes. Our sample is defined by the altitudes of all available records of weather stations. The system of equations (2) is equivalent to the matrix form λ₀ = C^-1·c₀, where C represents the matrix of covariances between all sampled values and c₀ the vector of covariances between sampled points and the point in which the value is assessed. The vector of covariance is obtained from the semivariogram γ(d) defined as:

where d represents the unit distance or lag from which semivariance is determined. The two semivariograms indicated in Figure 3 were computed for lags up to 200 km along two mutual perpendicular directions. Both semivariograms present similar spatial variances and are well described by a spherical model (dashed line) with a nugget of 0.048, sill of 1.01, and range of 120. The spherical model was found as the most suitable for modeling the covariances (3) and for solving the weights in the system of equations (2) by using GSL-GNU numerical libraries for matrix inversion. A comparison of the interpolated heights with real altitudes from all-term IDEAM weather stations is shown in Figure 4. Cross-validation of the Kriging interpolator was done by hiding each IDEAM data point during the computation of the weights and then comparing the prediction with the real value. Differences with an ideal Kriging interpolation (solid line) are within 3σ which we consider enough good for our purposes.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Based on Kriging results for altitude interpolation in the control region, we computed the monthly thresholds using the values appearing in Table 1. Results are very similar for all the months with small differences due to the bimodal annual cycle. Threshold temperatures for January with resolution of 28 × 18 km are shown in Figure 4b . The coldest region (0–2)° C corresponds to the Nevado del Huila, the highest snow peak in the central branch of the Andes. Thresholds between (2–7)° C are observed in the eastern Andes comprising Villa de Leyva (VDL), Pisba (PIS), and Macaravita (MAC). Warm regions with thresholds above 20° C characterize the valley of Magdalena river between the central and eastern branches of the Andes and the southeastern regions on zero altitude zones.

For comparison purposes, we show in Figure 4a the thresholds obtained using Delaunay triangulation of the 63 IDEAM long-term monthly records. We note that the spatial domain is constrained in this case and defined by the triangulation of the 63 irregularly distributed long-term IDEAM weather stations. Differences between both interpolations are significant mostly in regions surrounded by strong topographic gradients in which a better estimation of the threshold temperatures using Kriging is expected. This is particularly noticed in the poor IDEAM sampled region centered on PIS where a maximum difference of ± 8°C is observed. Other regions with low number of weather stations are the south of Bucaramanga (BUC) and the south-east of Medellín (MED). On these regions, the combination of the lack of weather stations and strong gradients in altitude makes the Delaunay interpolation unrealistic. Mean differences of ± 3°C over the eastern Andes are attributed mainly to strong gradients as the region is well sampled by the IDEAM data archive. Thresholds obtained from altitude are more precise than those obtained by interpolation of weather station records and are used as a reference for cloud detection in the B4 band.

Uncertainties in the threshold temperature determination are computed using the sigma error of the Kriging interpolation. The largest errors up to 6° C are observed over the sea-level locations and into the valleys, while errors over the Andes are always below 1.5° C as shown in Figure 4c . For the control region, the thresholds are computed with an uncertainty of ± 2°C.

With the aim of computing monthly threshold temperatures over a wider region including the Observatorio Nacional de Llano del Hato where log-books of cloud cover fraction are available, we apply monthly threshold-altitude relations over what we denoted as extended region covering latitudes between 0 and 13 N and longitudes between 70 and 78 W. A set of 3000 altitudes over this region were obtained from Google Maps and interpolated using the methodology discussed above. Monthly threshold temperatures between 21 and 24° C are observed across Caribbean coastal regions and in the valleys of the rivers Magdalena and Cauca in Figure 5. In the proximity to the Andes, thresholds get colder, especially in the highest mountains indicated with purple in the figure, such as the four highest locations over the Colombian and Venezuelan Andes highlands with threshold temperatures below 0° C. Uncertainties in threshold temperatures over the extended region are assumed to be equal to those obtained in the control region.

Zoom In Zoom Out Reset image size

Interpolated threshold temperatures covering the surface of the extended region are then used in our algorithm for cloud detection in the B4 band. For a fixed instant, if the temperature obtained from the GOES image at 10.7 μm is colder than T₀ then the presence of a cloud above the location is inferred.

Clouds along the upper troposphere are detected following a similar schema, defining first a threshold temperature for high altitudes where the weighting functions of B3 show the maximum value. We denote the altitude of the atmosphere measured from the ground with H, different to the altitude of a weather station h. We recall that the weighting functions depending on both, the wavelength and H, determine the layer from which the emitted radiation originates. For the B3 band, the median height values of the weighting functions are in the interval 8–10 km of altitude above the ground with a maximum at 9 km (Cavazzani et al. 2011).

Threshold temperatures at upper troposphere may be obtained using different approaches. Hydayat et al. (2012) considered the reference atmospheric model of Kneizys et al. (1996), which provides a vertical tropospheric profile of temperature for latitudes within ± 15° and adopted a threshold equal to -30.15°C at the observatory location based on a χ² analysis with data of sky visual inspection. Our approach is based on the use of three thresholds each one at heights of 8, 9, and 10 km obtained from radiosonde upper-air measurements.

We make a strong emphasis on the reliability of the temperature at upper troposphere. Particularly, we are interested in finding trustworthy threshold values between 8 and 10 km. Data were obtained twice a day from the freely available database of the University of Wyoming ² which provides information of temperature, humidity, pressure and mixing ratio as a function of the altitude for two stations located inside the extended region, 80222 SKBO (74°9'0''5 W, 4°42'0''N, 2546 m) in central Andes and 78988 TNCC (68°57'36''W, 12°11'60'', 54 m) in the Caribbean coast. Upper-air temperatures of both stations exhibit similar patterns as a function of the altitude. Around 60 radiosonde records were used per month adding data from both stations. Altitude resolution is variable with typical values ranging from 200 up to 600 m.

We compute monthly mean temperatures by considering measurements made twice daily (00 and 12 UTC) during the years 2008–2012 at each H using linear interpolation along a uniform vertical grid of 21 points between 8 and 10 km. In order to follow the same approach than the one used for ground thresholds, we define the threshold temperature at the altitude H as T_H  ≡ 〈T_H 〉-3σ_H , where 〈T_H 〉 is the monthly mean value and σ_H is the standard deviation. Monthly thresholds as a function of the altitude are displayed in Figure 6. Monthly standard deviations change with altitude and time showing yearly variations up to ± 2°C in the worst of cases. In the tropics, the temperature anomalies at high altitudes is enhanced by the presence of moderate size clouds usually associated with shallow convection responsible for anomalies up to ± 2°C at altitudes from 5 up to 14 km (Folkins 2008). This tropospheric warming has been largely observed and we suggest that is the main source of the sigma variations.

Zoom In Zoom Out Reset image size

The further assumption that these thresholds describe the upper troposphere of the whole extended region in many aspects is a strong hypothesis; however, the area of study is affected by atmospheric phenomena including the westerly low level Chocó jet coming from the Pacific coasts of Colombia (Poveda & Mesa 2000) and the Caribbean trade winds which leads to similar daily patterns of high-altitude clouds. The middle and right panels in Figure 7 show a typical evolution during the night of 2009 April 8. We see two images in B3 taken at the beginning and at the end of the night. High-altitude clouds shown in green color in the left panel coming from the Pacific ocean are blocked by the Andes system leaving clear zones in the Andes of Venezuela and over the Amazon zones.

Zoom In Zoom Out Reset image size

We note that the fact of extrapolating thresholds over all the extended region, the possible changes in the instrumentation and the differences between night and daytime observations may affect the observed temperatures and therefore introduce considerable noise in the computed thresholds. However, the monthly thresholds obtained with these two stations are in agreement with the predictions of the tropical atmospheric model of Kneizys (1996) indicated with the solid line in Figure 6.

We implemented an algorithm for percentage of clear-sky calculation during the night through a comparison pixel to pixel of GOES 12/13 interpolated temperatures with monthly threshold values obtained from long-term records of weather stations and radiosonde databases. We constructed four monthly threshold images, one for band B4 labeled as T₀ and three for B3 labeled as T_H each one at heights H = 8, 9, and 10 km. Daily comparisons were done in both bands at 00:15:15, 06:45:13 and 09:45:14 UTC with similar approach to those described in Hidayat et al. (2012) and Erasmus & Sarazin (2002). First, we used daily radiances in the B4 band to compute brightness and actual temperatures. If the satellite derived temperature is colder than the monthly threshold at B4 the presence of middle high clouds above the pixel image at the instant i is inferred:

where T₀ is a monthly constant depending only on the altitude of the terrain. If a pixel is classified as covered in B4, it will be also covered in the B3 band. On the contrary, if the pixel is clear in B4 we still have to verify the presence of high-altitude cirrus clouds by doing a further and similar analysis using the B3 band. The pixel is classified as clear or covered in B3 at the altitude H in the following way:

where 〈T_H 〉 denotes the monthly mean temperature of the upper troposphere at height H and σ_H the standard deviation at that height. We compute three values of clear-sky fraction at the instant i and at each altitude H using inequalities (4) and (5) by counting the number of clears or covered pixels. The total clear-sky fraction at altitude H is then obtained from the average of the three instants i along the night. One value of CSF is computed in this way at each altitude.

5.1. Classification of the Nights

Based on total CSF values obtained with SNS and PNS outputs, monthly averages at altitude H were computed. Cloud cover along the upper troposphere is sensitive to the altitude, being larger at low altitudes, mostly at H = 8 km, indicating that high cirrus does not affect considerably the net cloud fraction. We used the monthly CSF averages at H = 8, 9, and 10 km to compute a weighted mean for the three altitudes at each instant during the night. Weights were computed through a comparison of the amount of clear-sky fraction for photometric nights coming from our model with those obtained from daily number of observing hours reported by log books of CID or Site 2 in Venezuela. Site 2 is at 3600 m altitude and is enclosed by high mountains such as Pico Bolívar (4978 m) at less than 40 km away. As the resolution is 28 km in longitude and 18 km in latitude, the threshold temperature at Site 2 is highly affected by the altitude of the closest higher mountains, giving colder thresholds than expected. Using PNS instead of SNS produces more precise validation because PNS output is closer to the condition of total clear night.

Figure 8 corresponds to the clear hours above the reference Site 2 or CID as reported in the log book. Despite the bias present in the log books and the missing data for the years 2009–2010, due mainly to bad weather, we did not find considerable differences between observed and expected monthly clear-sky fraction values when different weights were used. Therefore, for the purposes of this study, we consider it sufficient to use equal weights for the three altitudes.

Zoom In Zoom Out Reset image size

The validation of our code is limited due to factors such as the lack of completeness of daily log books, technical issues, high humidity, strong winds, and uncertainties of our model. We used all the information available in the log books as the general description of the night done by the astronomer to compute the amount of daily clear-sky fraction for each month of the years 2008–2012 at Site 2. Comparison with our monthly results using PNS is displayed in Figure 9. The solid line represents no differences between predicted and measured monthly clear-sky fraction values. There is a large spread during the considered years, especially for clear-sky fraction values between 20% and 60%, but in general terms they are in reasonable agreement.

Zoom In Zoom Out Reset image size

Monthly percentages of clear sky for the years 2008–2012 over the extended region were obtained using the method described above. The main common feature is that monthly averages display maxima at the same locations along the five years. Maximum monthly percentages of clear sky are observed during two periods in the years 2008–2012, one from December to February and another from June to July according with the bimodal annual cycle of precipitation (Poveda & Álvarez 2011). The year 2012 stands out as the best year displaying the highest percentages during both periods.

We take the year 2012 as representative of the monthly distributions of clear sky. As illustrated in Figure 10, our monthly averages of CSF for SNS nights exhibit differences along the year with the highest CSF values always located over a wide region in the central Venezuelan Andes with a maximum CSF≥80% at Site 1 during both periods. For Site 2 or CID, corresponding to Observatorio Nacional de Llano del Hato, we get a CSF mean value of 58.5%, lower in comparison with Site 1 during both periods.

Zoom In Zoom Out Reset image size

We have found two regions in Colombia with similar night qualities to those observed at Site 2. The northern part of the Sierra Nevada de Santamarta close to Nabusimake (NAB) and the central Serranía del Perijá toward the east of Codazzi (COD). Percentages at these two regions are in the range of 80–90% during the first period, 50–70% during the second, and below 30% from March to May and between August and October. Sites over this region contain the best places in Colombia found in this study.

Somewhat low-quality places are found over the northeastern Andes of Colombia, over mountain systems along BUC, MAC, and PIS, with clear-sky fractions between 30–50% during both periods of 2012. Southern Andes regions in the Nudo de los Pastos, especially those in northern Ecuador, display CSF values below 30% from November to May and ∼70% from June to October. Very low percentages of clear-sky fraction below 20% characterize the central Andes of Colombia over the regions comprising Popayán (POP) and BOG, with the exception of the region centered at the high site Nevado del Huila (75°55'12''W, 2°51'36''N) at 3700 meters with a mean of 35% between December–February and June–July and 33% for the other months.

Regarding monthly averages of CSF for PNS nights, the same regions were found but with low percentages in comparison, as shown in Figure 11. The absence of spectroscopic nights in central Colombia from May to October confirms that November–February and June–July are marked by the largest CSF percentages, while March–May and August–November are characterized by lower CSF values. We also observe such behavior for other years. The existence of this bimodal behavior with maxima during the dry season and minima during the rainy season is attributed, as we mentioned already, to the double passage of the intertropical convergence zone for the extended region. Therefore, we conducted yearly SNS and PNS simulations, bringing together the months December, January, February, June, July, and August into one group that we call the dry season, and March, April, May, September, October, and November into another group denoted the rainy season.

Zoom In Zoom Out Reset image size

SNS percentages for both seasons show maxima at the same locations than those observed for the year 2012 with seasonal differences of ∼20% with the exception of low-latitude locations close to the equator, where slightly smaller differences are observed. Percentages for PNS simulations result as 10% lower in comparison to SNS nights, but with identical CSF distribution over the extended region.

Annual clear-sky fraction averages obtained from our SNS simulations during rainy seasons for the years 2009 and 2012 are larger than for the other years. It seems to correlate with the changes in the Oceanic Niño index (ONI) reported by the climate prediction center NOAA. ³ Changes from positive to negative anomalies in April 2010 associated to a La Niña event increase the number and intensity of tropical storms in central Colombian Andes, thus decreasing the clear-sky percentages. The Nudo de los Pastos and the Sierra Nevada de Santamarta are not considerably affected by this El Niño Southern Oscillation (ENSO).

For selection purposes, we give preference to the simulations for spectroscopic nights during the dry seasons. We have selected 12 sites based on five-year averages during the dry seasons that can be suitable for astronomical observations by selecting locations with CSF≥30%. Factors such as accessibility, security, service, local humidity, and atmospheric seeing were not considered, and we need additional studies to confirm these locations as potential astronomical sites. These 12 selected sites are distributed mostly along the following four zones: the central Andes of Venezuela, comprising the states of Mérida and Trujillo, the northeast of the Sierra Nevada de Santamarta around the sites NAB and COD, the northeastern Andes of Colombia in the proximity of El Espino (ESP), and a wide region on the southern Andes in Nudo de los Pastos containing two locations in Ecuador. We show in Figure 12 the contours of clear-sky fraction for these selected zones during the dry season with a spatial pixel resolution of 28 × 18 km.

Zoom In Zoom Out Reset image size

Even though the selection criteria involves only spectroscopic nights during the dry seasons, we give a total number of clear nights per year at selected sites by adding the CSF five-year averages for each season and then multiplying by a factor of 1.827. Tables 2 and 3 contain properties of the selected sites indicating the identifier, the nearest populated location, geographic coordinates of the site, altitude, yearly five-year averages for the dry and rainy seasons, five-year average of clear-sky fraction, standard deviation, and the total number of clear nights per year.

8.1. Central Venezuelan Andes

8.2. Sierra Nevada de Santamarta

8.3. Eastern Andes of Colombia

8.4. Nudo de los Pastos

This region consists of tropical humid forests where the Andes splits into the western and central branches. Sites 9 and 11 in north of Ecuador stand out as the best places in the region with (105 ± 36) and (101 ± 42) clear nights per year, respectively. With (24.3 ± 3.9)% below our selection limit we include the Site 12 Laguna de la Cocha (LCO) located at the Colombian side of the Nudo de los Pastos.

Minimum seasonal differences are found at the sites in this region. Compared with the other sites, the percentage of clear sky during dry and rainy seasons is very similar in sites within the Nudo de los Pastos, confirming that the sky cover does not change significantly from one season to another at locations near the equator. We note from Tables 2 and 3 that the minimum differences between CSF values for dry and rainy seasons for all 12 sites are observed especially for the sites 9, 11, and 12 with differences of 4, 2, and 1%.

Five-year average for the selected sites exhibits an extended minima in 2010–2011 during rainy seasons, as shown in Figure 13. We found a possible correlation to La Niña episodes, since positive values for the Southern Oscillation Index (SOI) are observed for these two years. Interior plots in Figure 13 show SOI values obtained from NOAA website for 2008–2012. The symbols for the months go along with those for the sites. Positive values mostly for 2008, 2010, and 2011 are associated to intensification of storms over the Andes and then to low CSF values, as observed.

Zoom In Zoom Out Reset image size

In this work, we describe a method to compute the clear-sky fraction from GOES data based on comparisons with long-term records of temperature from weather stations and radiosondes, over a wide region of northern South America, including sites in Colombia, Venezuela and Ecuador.

We conducted a temporal analysis based on three daily values along five years doing a careful comparison of GOES12/13 temperatures in bands B3 and B4 with thresholds at different altitudes. This approach enabled us to carefully determine the percentage of clear sky simultaneously over the whole region. Similar percentages were obtained for all locations along the five years confirming the reliability of our methodology.

Threshold monthly temperatures as a function of the altitude of the terrain were obtained for a small region of control through Kriging interpolation of long-term records of temperature at weather stations. Monthly linear correlations between thresholds and altitude were found and extrapolated to an extended region including Observatorio Nacional de Llano del Hato. Threshold monthly temperatures at 8, 9, and 10 km along the upper atmosphere were also computed based on daily radiosonde data from two stations located at sea level and in the central Andes.

The method differs from other cloud fraction algorithms that use long-term records of temperature to obtain the CSF. Our approach gives more precise CSF values as long as we have used reliable reference temperatures at the surface and at the high troposphere. Furthermore, our model can be easily extended to other tropical regions of astronomical interest.

Nights are classified as spectroscopic or photometric based on the three CSF values during the night. The total number of spectroscopic and photometric nights per year at a fixed site are computed separately from 5-year averages for dry and rainy seasons.

The best location found in this study is Site 1 (70°28'48''W, 9°5'60''N) in the central Venezuelan Andes with (220 ± 42) spectroscopic clear nights per year and (157 ± 24) photometric clear nights per year. Site 1 is placed at an altitude of 3480 m and 35 km north-east from Timotes within a protected area of Venezuela. Two lower quality locations were found in Sierra Nevada de Santamarta, Site 3 (73°31'12''W, 10°55'12''N) and Site 4 (73°2'24''W, 10°8'24''N) with (126 ± 34) and (111 ± 27) spectroscopic clear nights per year, respectively. Site 3 stands out as the best site identified in Colombia.

The eastern Andes of Colombia host two regions with more than 90 spectroscopic clear nights per year. The region centered at Site 5 (72°43'12''W, 7°1'12''N) in Norte de Santander with (107 ± 23) clear nights per year and the wide zone centered at El Espino comprising Macaravita and El Cocuy in Boyacá with (94 ± 13) clear nights per year. The nearest locations to the equator in the Nudo de los Pastos region are Site 9 (77°50'24''W, 0°46'48''N) and Site 11 (77°50'24''W, 0°15'36''N) with (105 ± 36) and (101 ± 42) clear nights per year, respectively, presenting minimum seasonal differences of number of clear nights as expected. Finally, CSF values below 25% observed during the rainy seasons of the years 2010 and 2011 seems to be correlated to positive values of the Southern Oscillation Index.

The selected sites obtained in this study may be considered as a first step to decide the location of an astronomical observatory. However, a site survey is absolutely required before starting any astronomical facility. Results of an observational campaign with the aim to characterize the sites identified in this study will be reported in a forthcoming paper.

We would like to further point out that the value of our approach goes beyond the possibility of being useful for identification of potential sites in the region. Our approach allows us to quantify the contribution of the different layers of the atmosphere to the CSF above the site, in particular, the high troposphere and the surface. Although we have had to make a number of simplifying assumptions, the results are self-consistently and provide a coherent picture of the CSF over the extended region.