Climatological Study for the Cherenkov Telescope Array North Site at the Canary Islands I: Temperature, Precipitation, and Relative Humidity

We obtained climate normals over a period of 10 years (2003–2012) at Izaña (2395 m), Tenerife (Spain) with the aim of characterizing the mesoscale climate conditions governing the two astronomical observatories in the Canary Islands: Teide Observatory, located on the summit of Izaña and Roque de los Muchachos Observatory (ORM), approximately 140 km away, on the island of La Palma. Both sites are at the same height (≈2300 m) and share very similar regional climatological conditions. The study was carried out as part of the site selection campaign for the Cherenkov Telescope Array (CTA) in the northern hemisphere, which was finally installed at the ORM. We used long-term and high-quality climate data series recorded by the Centro de Investigación Atmosférica de Izaña (CIAI) at the Izaña Atmospheric Observatory (IZO; WMO#60010). In this paper, we analyze the temperature (central tendency, maxima, minima, and ranges), precipitation (intensities per hour and daily, monthly, and yearly cumulative), and relative humidity. The precipitation data are divided into liquid, snow, and hail. All the results are presented in triplicate, covering the whole series, nighttime, and daytime. The results show a smooth temperature profile with sporadic but moderate extremes, a low daily (24 h) temperature range (median 7.1 °C), and extremely low nighttime temperature range (median 2.7 °C). Rainfall is very rare and sporadic (287 mm a−1), with low intensities. The relative humidity is also low (median 29% and 25th percentile 15%) and follows the precipitation cycle. These results confirm the stability and optimal conditions required to fully operate CTA and other astronomical facilities. The requirements specified in the CTA project documentation are fulfilled.


Introduction
Ground meteorological conditions are crucial for astronomical site testing. The requirements for a good astronomical site include small day-night air temperature gradients; high thermal stability during the night; low relative humidity; and moderate wind speed and wind gusts (McInnes & Walker 1974;Muñoz-Tuñón 1998, 2002. A proper characterization of these parameters is important not only when estimating the potential useful time, but also for operational and design specifications. The most accurate description of the meteorological conditions at a particular location is obtained from a statistical treatment of the longest available climate series. Recording these series has been one of the major efforts in the recent site-testing campaigns for the forthcoming class of optical telescopes >10 m, for example the European Extremely Large Telescope (E-ELT; Varela et al. 2014) and the Thirty Meter Telescope (TMT; Schöck et al. 2009). In the case of unshielded large open facilities such as the Cherenkov telescopes, the meteorological requirements are particularly important, as they may compromise the safety of the installation. Where this is the case, it is essential to characterize additional parameters, such as the amount, type, and intensity of precipitation, as well as carrying out a detailed study of extreme events (Medina et al. 2009;Actis et al. 2011).
Assuming that climate is continuously evolving over time for either natural or anthropogenic reasons, the most suitable length of timeseries for a proper description of present climate conditions has been widely discussed; see, for example, Kunkel & Court (1990) or Lamb & Changnon (1981). The World Meteorological Organization (WMO) recommend a period of 30 years (Arguez & Vose 2011), where possible, to update the published climatological normals for their widespread set of stations (WMO 2017). Shorter periods have also been proposed for practical applications (Lamb & Changnon 1981;Dixon & Shulman 1984;Huang et al. 1996). All these studies have concluded that 10 years suffice, in most cases, to provide an accurate picture of climate conditions to minimize distortions in the series resulting from global warming (GW) and other trends. This criterion is also commonly used by the industrial demand for accurate forecasts (Arguez & Vose 2011).
The Cherenkov Telescope Array 3 (CTA) is the next generation of ground-based observatories for gamma-ray astronomy. It is currently the world's largest and most sensitive high-energy gamma-ray observatory, with more than 100 Cherenkov telescopes to be installed at two different locations: one in the northern hemisphere, and another in the south. Southern hemisphere candidate sites included Namibia (Aar), Argentina (Leoncito, Leoncito++, and San Antonio de los Cobres), and Chile (an area close to Cerro Paranal); the northern sites were the USA (Meteor Crater, AZ), Mexico (San Pedro Mártir), and the Canary Islands (Spain), with two different locations at the same altitude: Roque de los Muchachos Observatory (ORM) on the island of La Palma, and Teide Observatory (OT) in Izaña, on the island of Tenerife (see Section 1.1 for details). On completion of the site selection process, Paranal and the ORM were finally chosen as the locations for the CTA on 2015 July 15. 4 To support the CTA site-testing campaign, we obtained climate normals over a period of 10 years at Izaña, taking advantage of the long-term and high-quality climate series recorded at Izaña Observatory (IZO) by the Centro de Investigación Atmosférica de Izaña (CIAI) of the Spanish State Meteorological Agency (AEMet). Izaña (see Figure 1) is located on the island of Tenerife, ≈140 km from ORM, at approximately the same height and with the same mesoscale climatological conditions. The data recorded at the IZO were considered the most relevant and applicable to describe the long-term mid-tropospheric climate conditions at both the OT and the ORM (Castro-Almazán et al. 2015).
In this paper, we analyze 10 years of climate series recorded at the IZO and calculate normal values and thresholds for the parameters relevant to the CTA site selection, design, and operation. This paper is the first of two articles. In this first release, we analyze temperature (central tendency, maxima, minima, and ranges), precipitation (intensities per hour and daily, monthly, and yearly cumulative), and relative humidity. The main results are summarized and discussed in Sections 3 to 6. A discussion comparing the results with the last published WMO climatological normals for the station (period 1961-1990) is given in Section 7, and a study of the correlation between precipitation and the North Atlantic Oscillation index in Section 8. A data set of graphical results is given in extensive appendices (Appendices A and B), together with the full tables (Appendices C and D), for use as a catalog. In a second paper, we shall study the wind regime at Izaña, with emphasis on the identification of local features.

The Site
Izaña (see Figure 1) is located close to the Teide summit (3717 m), in the center of the island of Tenerife (Canary Islands, Spain), at 2395 m above sea level (the area covered by the ORM ranges from ≈2180 to 2423 m above sea level in the peak). Besides the IZO, the mountain also hosts the OT, an astronomical observatory belonging to the Instituto de Astrofísica de Canarias (IAC). The IZO station (CIAI-AEMet hereafter) has maintained a continuous climate record since 1916, and is one of the WMO's set of reference stations with code #60010 (see Figure 2 for details). The Observatory is also one of the high-altitude references of the Global Atmosphere Watch 5 (GAW) program of the WMO, designed to monitor the global chemical composition of the atmosphere.
Because of their latitude and eastern location in the North Atlantic Ocean, the Canary Islands exhibit a vertical tropospheric structure with a trade wind thermal inversion layer (IL), driven by subsiding cool air from the descending branch of the Hadley cell. The slowly descending air conserves the potential temperature and absolute humidity while it is warmed adiabatically, inducing low relative humidity (Graham 2017). The altitude of the IL ranges on average from 800 m in summer to 1600 m in winter, well below the altitude of the Observatories (Dorta-Antequera 1996; Carrillo et al. 2016). The IL separates the moist marine boundary layer and the dry free atmosphere, resulting in the very high stability of the atmosphere above.

Data Description and Treatment
The 10-year data series analyzed in this paper ranges from 2003 January 1 to 2012 December 31. The data have been obtained directly from the AEMet via an institutional agreement. The AEMet is the official meteorological service in Spain, as well as the air traffic meteorological authority. Because of its state agency status, AEMet is involved with standard international criteria in climate recording (quality controls, sensor calibration, and intercalibration after replacements, etc.). Statistical homogeneity was assumed for the series, so no further tests have been applied. This assumption is based on the near invariability of the environment, including obstacle-free surroundings in the more than 100-year history of the station and on AEMet's own quality controls.
The data, except for precipitation, were recorded with an automatic weather station (AWS) with a frequency of 1/10 min (10-minute series), which is suitable for separated nightdaytime analyses. All of the results are presented in triplicate, covering the whole series (all time), nighttime, and daytime. The day-night limit was calculated by means of the nautical twilight ephemeris (when the Sun is 12°below local horizon) corrected for the station's altitude. After a first inspection, some inconsistent values were detected in the AWS precipitation data (probably due to the presence of frozen-phase precipitation). For this reason, the precipitation data were obtained from the non-automated series that is available as hourly cumulative. All of the series were explored for the presence of outliers and inconsistencies, with nothing remarkable (apart from that mentioned for precipitation) to report.
The precipitation units in the original series are tenths of mm with a final resolution of 0.1(0.01 mm). The label "imperceptible", found when the precipitation detected by the observer falls below this level, was substituted by a default value of 0.01mm (upper limit) to include the data in the analyses. As the precipitation is a cumulative variable, the measured value depends on the integrated lapse of time, and different interpretations may be extracted from each. We have integrated the original (mm h −1 ) data to compound daily, monthly, and yearly cumulative series of precipitation. We have also separated the rainfall data as a function of the water phase into liquid and frozen precipitation. Additionally, in the hourly and daily series, we have separated the frozen data into snow and hail components. The units for temperature are Celsius (°C), with a precision of 0.1°C.
The statistical normals are provided in three different ranges: absolute 10-year period, monthly, and yearly values. The normals have been established by means of the central tendency and the dispersion of the distribution for all the data in the range. The central tendency estimates have been calculated as median values in a search for statistical robustness in the results (Wilcox & Keselman 2003;Arguez & Vose 2011). The median absolute deviation (MAD) has been used to estimate the dispersion. For normally distributed samples, 50% of the sample falls in the median±MAD or equally, MAD≈1/2·(P75-P25) and σ≈1.4826·MAD, where P75 and P25 are the 75th and 25th percentiles, respectively, and σ is the standard deviation. The mean á ñ x and σ are also given in the results.
It is not easy to estimate accurate sample errors in highly periodically correlated series such as climate series. Instead, we present the results with the coverage percentage, obtained from the potential data in each time interval. The following criterion was imposed on the data coverage to include the period in the analyses.
• A cutoff limit of 90% was imposed on the diurnal coverage to minimize the risk of missing the extremes when obtaining the temperature daily maxima, minima, and ranges. • For the yearly points, no values are considered when the coverage is below 82%; that is, ∼65 missing days (a random distribution in missing data is assumed.) • For the individual monthly points, no values are considered when the coverage is below 50%. This level ensures a sampling error of less than 10% (if s < á ñ x 4) or less than 20% (if s < á ñ x 2). • For the 10-year monthly behavior, no points are included for coverages below 50%. This level ensures a sampling error of less than 6% (if s < á ñ x 2) or less than 12% (if s < á ñ x ).
To estimate the errors defining the monthly cutoffs, we have assumed statistical homogeneity and a normal distribution within a standard month of 30 days. The confidence level was established at 95% (α = 0.05). The error is defined as where N is the sample size (30 for the individual months and 300 for the 10-year monthly behavior), n is the number of data, and a --t n 1;1 2 is the t-Student quantile for n−1 degrees of freedom and a confidence of 1−α.

Software Used
The data were collected and separated into day-night groups with a self-developed package of Linux shell scripts called DataCollector that uses the Python PyEphem package to compute the daily time of twilight. For the data analyses and plotting, we used the vector-oriented Interactive Data Language (IDL). Time gaps larger than 20 minutes were labeled to avoid line connection in plots. IDL scripts were also written to generate the appendix tables directly in LATEX format.

Temperature
Thermal stability is analyzed with four parameters: the temperature 10-minute records (T), with sampling frequency of 1/10 min (see Figure 3), and the daily maxima (T M ; see Figure 4), daily minima (T m ; see Figure 5), and temperature ranges (TR; see Figure 6), with a frequency of 1/24 h. TR is defined as the difference between T M and T m in the same day (or period, in the case of night/day times series). Figures 3 to 6 show the timeseries and the statistical behavior of each parameter and are also included in the Appendix A (Figures 11-14), along with other graphical results. In Table 14 in Appendix C, there is a full compilation of all the temperature statistical results.
Even though Izaña is a high-altitude observatory, temperature is dominated by a soft profile (see Figure 3), with very sporadic extremes events (Figures 4 and 5) and low TR ( Figure 6). From the point of view of an astronomical observatory, this is a desirable scenario. Extremely low temperatures make a harmless operation difficult, and put the installation at risk if it coincides with frozen precipitation. On the other hand, very high maxima involve excessive soilatmosphere heat interchange. This non-equilibrium system is the principal responsible of ground layer turbulence, by means . Temperature evolution in Izaña (CIAI-AEMet station). Top: timeseries. Monthly medians (white circles), dispersion (±median absolute deviation; shadow), and absolute monthly maximum and minimum (thin lines). Bottom: histogram (blue; left axis) and cumulative distribution (black; right axis). The main statistics are in the legend box (numerically) and in the thin lines of the cumulative plot. The coverage percentage is estimated from the sampling frequency and the full record length. "m.a.d." is the median absolute deviation (MAD). (A color version of this figure is available in the online journal.) of the fluctuations induced in the temperature field. Low TR implies rapid equilibrium recovering, particularly during the day-night transition.
The numerical results are summarized in the Table 1. The temperature at CIAI-AEMet shows a median value of 9.6°C, with T m and T M ranging from 6.3°C to 13.6°C. Attending to the extremes of the distribution, the 95% of time (95th percentile, P95) the temperature stays below 21°C. On the other hand, T is over 0°C also the 95% of time (P05). In particular, T M exceeds 0°C∼98% of days and therefore melting conditions are reached in an elevate percentage of days. TR also shows low values (median≈7°C) and a great stability, as a result of the very low dispersion (MAD≈1°C).

Night and Day Stability
Astronomical telescopes are normally designed to observe either the Sun during the day or other astronomical objects during the night with different techniques. For this reason, all of our results, including temperature maxima and minima, were recalculated using the night-and daytime series separately. Table 2 summarizes the night-day differences for T and TR for the whole 10-year data set (see also Figures 15 to 22 in Appendix A and Table 14 in Appendix C). The nighttime temperature is ∼4°C lower than during daytime, with a stable behavior for the different percentiles. As expected, the nighttime temperature distribution is less spread out than its daytime counterpart. Stability is also evident in the low . Maximum daily temperature evolution in Izaña (CIAI-AEMet station). Top: timeseries. Monthly medians (white circles), dispersion (±median absolute deviation; shadow), and absolute monthly maximum and minimum (thin lines). Bottom: histogram (blue; left axis) and cumulative distribution (black; right axis). The main statistics are in the legend box (numerically) and in the thin lines of the cumulative plot. The coverage percentage is estimated from the sampling frequency and the full record length. "m.a.d." is the median absolute deviation (MAD). (A color version of this figure is available in the online journal.) TR median values for the night-(2.7°C) and daytime series (6.5°C) together with the narrow dispersions (MAD≈1°C for both). The very low TR values at night also evidences a rapid reaching of thermal equilibrium in the night-day transition.  Tables 16 to 19 of Appendix C). The temperature shows a standard north hemisphere seasonal behavior with maximum in July (18.9°C) and minimum in February (3.8°C). Only moderate differences are found in dispersion for different months, the most stable month occurring in the summer.

Monthly Behavior
No big differences are found between TR extremes (∼2°C). The lowest values of TR are in autumn and winter, and the highest are in spring and summer.
In Table 4, we summarize the monthly night-day results for T and TR. More detailed information is given in Table 16 of Appendix C. The night and day temperature timeseries show the same behavior throughout the year, both with minima and maxima in February and July (see Table 4), respectively, as in the whole series (see Table 3). The higher daytime MAD values also dominate the dispersion in the full series. However, Figure 5. Minimum daily temperature evolution in Izaña (CIAI-AEMet station). Top: timeseries. Monthly medians (white circles), dispersion (±median absolute deviation; shadow), and absolute monthly maximum and minimum (thin lines). Bottom: histogram (blue; left axis) and cumulative distribution (black; right axis). The main statistics are in the legend box (numerically) and in the thin lines of the cumulative plot. The coverage percentage is estimated from the sampling frequency and the full record length. "m.a.d." is the median absolute deviation (MAD). (A color version of this figure is available in the online journal.) whereas the daytime TR also shows autumn-winter minima and spring-summer maxima, the nighttime series show a slightly reverse behavior, the most thermally stable nights occurring in August. Both night-and daytime TRs have very low dispersion all through the year.

Yearly Behavior
Finally, Table 5 summarizes the yearly T, T M , and T m results. The complete set of yearly results are shown in the  Figures 26 to 28 of Appendix A and Tables 20 to 23 of Appendix C. Some differences are found between median values and dispersion for different years. The warmest year in the sample was 2010, with a relative difference of 1.8°C with respect to the coolest year (2004). There is high correlation between T, T M , and T m . MAD values also show differences among different years (<1.5°C). These oscillations may be governed by the global climate oscillations on the hemispheric scale. We discuss this further in Section 7.

Precipitation and Relative Humidity
The precipitation (or rainfall; PCP hereafter) is a key factor in the climatic classification of any geographic area, and the relative humidity (RH hereafter), together with temperature, is one of the principal parameters governing the amount of Figure 6. Daily temperature range evolution in Izaña (CIAI-AEMet station). Top: timeseries. Monthly medians (white circles), dispersion (±median absolute deviation; shadow), and absolute monthly maximum and minimum (thin lines). Peaks outside the Y top range are specifically labeled. Bottom: histogram (blue; left axis) and cumulative distribution (black; right axis). The main statistics are in the legend box (numerically) and in the thin lines of the cumulative plot. The coverage percentage is estimated from the sampling frequency and the full record length. "m.a.d." is the median absolute deviation. (A color version of this figure is available in the online journal.) precipitation. RH is also important in understanding the water balance in the soil, which is related to the aridity. All of these factors together played an important role in the design and construction specifications for the CTA telescopes, involving the bases, mirror coatings, and the isolation of the instruments. From an astronomical point of view, the amount of rainfall is related to the downtime probability and safe operation of the telescopes. Relative humidity may also be related to the amount of precipitable water vapor (PWV) above the observatory, which is the principal cause of atmospheric extinction in the infrared, microwave, and submillimeter bands.
The principal statistical results of PCP and RH at CIAI-AEMet are shown in the Figures 7, 8, and 9. These figures are also included in Appendix B, along with the other graphical results (Figures 29-37). In Table 24 of Appendix D there is a full compilation of all the statistical results for PCP and RH.
The hourly PCP (see Figure 7) is related to the intensity of precipitation. Hourly PCP data are also an indicator of significant trends in precipitation and changes in rainfall patterns (Cooley & Chang 2017;Trenberth 1998). This feature is even more important for dry scenarios or periods. Complementary to the efficient identification of such periods, the daily precipitation (see Figure 8) is a useful tool. The daily PCP, together with the number of days with precipitation, indicates the total volume of rainfall.
In Table 6, we summarized the main results of hourly PCP and RH for the entire 10-year database. Rainfall episodes at CIAI-AEMet are extremely sporadic; therefore, all the percentiles less than the 97th are 0. P99 corresponds to a value of 1.1mmh −1 , and the fraction of time with precipitation ranging between 0.0 and 0.2mmh −1 is 97%.
The results of daily PCP (see Figure 8) show a clear absence of precipitation in Izaña, with ∼90% of days having no precipitation of any type. The PCP concentrates on some isolated days, frozen PCP producing the highest values. This is because, in terms of the amount of water, a volume of fallen snow is ∼100 times the equivalent volume of liquid rain (UKMET 2011). There is an exception in winter 2009-2010, with an episode of liquid PCP of ∼150mmday −1 and four days ranging between 50 and 100mmday −1 .
The RH (see Figure 9) shows a low median value of 29%, remaining below 15% for a quarter of the time. This is a consequence of the typical IL height that is usually below the Izaña summit (see Section 1.1). The operational limit for the astronomical facilities is normally established a few points before condensation, RH=90%, i.e., that is not reached in the 87% of the cases.

Monthly Behavior
The monthly behavior of PCP and RH is shown in the Figures 38 to 40 of Appendix B and in Tables 27 to 30 of Appendix D. Condensation episodes involving high RH values and PCP at Izaña usually coincide with either of two circumstances, the IL reaching observatory level (cloud overflow) or the arrival of certain types of Atlantic lowpressure systems (García- Herrera et al. 2001). Both scenarios tend to be concentrated in the autumn and winter months. More   2003). These values, taken with the dispersion ranges, correspond to an environment between the arid and the semi-arid levels in the arid zone classification published by the Food and Agriculture Organization of the United Nations (FAO) 6 (FAO 1989). This classification is coherent with the expected climate in a high-altitude observatory environment. The greatest dispersion relative to the central values is for the frozen phase of precipitation.

Night-Day Behavior
PCP is less frequent and intense during nighttime (see Table 8). The average number of days of PCP is 36a −1 at night and 42a −1 during daytime (see Table 32 in Appendix D). The average number of days for the full series is 51a −1 (note that PCP may coincide on a same day during both night-and daytime). Only for hail does the nighttime have a slight higher probability. No differences have been found between day and night for RH.

CTA Requirement Fulfillment
The CTA consortium established a set of requirements for all the parameters relating to site selection, including climatic thresholds (internal documentation of the project). In Table 9, we have included the requirements for the parameters analyzed in this study, together with the level of fulfillment, based on the climatic results described in the previous sections, figures, and tables where the results are found.
All of the requirements are fulfilled (see Table 9), with the marginal exception of the maximum allowed temperature (25°C), which is exceeded 0.6% of the time owing to some maximum recorded (the absolute maximum is 27.4°C). The nighttime TR limits (<  | | 7.5 C h and <  | | 30 C 24h) are far from being violated in the CIAI-AEmet series, with extreme values of  | | 1.2 C h and  | | 15.4 C 24h, and median values of  | | 0.3 C h and  | | 7.5 C 24h, respectively. This is indeed one of the most important strengths of the Canary Islands Observatories.
There is also a 100% fulfillment with regard to PCP requirements. The maximum allowed daily and hourly PCP values (200 mm/24h and 70 mm/h) are clearly fulfilled by the maximum recorded values at CIAI-AEMet: 145 mm/24h and 27 mm/h, respectively. There are two additional requirements not directly recorded in the series. The maximum ground snow accumulation should be 500 mm, and the hail episodes with stones larger than 20 mm in diameter should happen with a frequency <0.03 a −1 (1 in 30 year). To estimate the degree of fulfillment of the first of these limits, we have considered the absolute extreme value recorded in the entire series of frozen precipitation (360.2 mm/a in 2007), which shows that the threshold is not reached even when accumulating all the snow in a year in a single day.
To estimate the frequency of heavy hail episodes (>20 mm in diameter), we have searched for episodes with intensities exceeding 20 mm/h (the only case where hail stones larger than 20 mm could occur). Only three points, corresponding to the same event with the absolute maximum of the 10-year series (24.3 mm/h), exceed this value. This means that, even in the worst case, the requirement is fulfilled (in a 10-year-long series, 0.1 a −1 is the minimum measurable frequency). Moreover, the probability that this isolated episode involved 20 mm stones is close to zero.

Extreme Events, Thresholds, and Return Periods
The frequencies of occurrence by threshold and the different return periods for temperature are listed in Table 15 of Appendix C. As reflected by the P05 and P95 values given in Table 1, only moderately extreme events are recorded in the temperature series. The absolute maximum and minimum in the 10-year series are 27.9°C and −7.5°C, respectively. There are 2.5 days per year on average with T m <−5°C and 15.1 days per year with T M >25°C.
A full list of the different thresholds for hourly and daily PCP is given in Tables 25 and 26 of Appendix D. The most intense episode of PCP recorded corresponds to a rain event (liquid) of 27.0mmh −1 . This is the only point above 20mmh −1 recorded for rain in the whole series (the other episode is for hail). Following the UK Met Office synoptic scale (UKMET 2011) for rain showers, rain is classed as slight when it is 0-2mmh −1 , moderate when it is 2-10mmh −1 , heavy when it is 10-50mmh −1 , and violent when it is >50 mmh −1 . The same scale for snow showers is slight (0-5 mm h −1 ), moderate (5-40 mm h −1 ), and heavy (>40 mm h −1 ). Our results show that PCP at CIAI-AEMet is sparse and basically may be classified as slight. No violent rain showers have been recorded, and there is only one event (27.0 mm h −1 ) in the lower half of the heavy scale. Figure 7. Intensity of precipitation in Izaña (CIAI-AEMet station). Top: timeseries. Liquid phase (sky blue) includes rain and drizzle. The frozen phase is shown separately as snow (deep blue) and hail (green). Peaks outside the Y top range are specifically labeled. The intensity references follow the UK Met Office synoptic scale for rain showers UKMET (2011): slight (0-2 mm h −1 ), moderate (2-10 mm h −1 ), heavy (10-50 mm h −1 ), and violent (>50 mm h −1 ). Bottom: histogram (blue; left axis) and cumulative distribution (black; right axis). Y axes have been trimmed to detail structures other than precipitation 0mmh −1 . The main statistics are in the legend box. The coverage percentage is estimated from the sampling frequency and the full record length. "m.a.d." is the median absolute deviation. (A color version of this figure is available in the online journal.) Snow and hail occur 11.7 and 1.1days per year respectively. The number of days per year with PCP>2 mmh −1 (appreciable precipitation), is 14.7, of which 9.8d/a are liquid rain. The number of days per year with PCP (of any intensity or type) is 52.2.

Comparison with the WMO Climatological Normals
The latest available 30-year climatological normals for the CIAI-AEMet station are for the period 1961-1990(INM 1995WMO 1996, WMO30, hereafter). 7 In Table 10, we compare the normal average temperatures in WMO30 with the equivalent values obtained in this work. We have averaged the results by season, considering full-month seasons, with winter being the months January to March, and so on, the remaining seasons comprising three months each. Although a long-term analysis of trends is beyond the scope of this study, we have obtained decadal normalized variations to take into account the effect of GW. To estimate the trends we calculated the difference between our mean values and those from WMO30, normalized by a factor 3.2, which is the number of decades between the central year of both series (1975 for WMO30 and 2007 for this study). For comparison, we have included the trends obtained by Martín et al. (2012) in a detailed study of GW assessment on Tenerife. The authors The results in Table 10 show a very good concordance between our values and those of the WMO30 for the decadal GW estimated by Martín et al. (2012) (row C in the table). The calculated trends also fall inside the ranges defined by the trend and error estimates of Martín et al. (2012), with the exception of the winter period, where we found no significant warming. Figure 9. Relative humidity evolution at Izaña (CIAI-AEMet station). Top: timeseries. Monthly medians (white circles), dispersion (±median absolute deviation; shadow) and absolute monthly maximum and minimum (thin lines). Bottom: histogram (blue; left axis) and cumulative distribution (black; right axis). The main statistics are in the legend box (numerically) and in the thin lines of the cumulative plot. The coverage percentage is estimated from the sampling frequency and the full record length. "m.a.d." is the median absolute deviation. (A color version of this figure is available in the online journal.) Table 6 Precipitation Intensity and Relative Humidity for the Entire Database (See Table 24). The Values PXX are the Percentiles and MAD is the Median Absolute Deviation This good agreement is also shown in the averages of T M and T m (see Table 11). The results indicate a more pronounced effect of GW in the minima, where our estimate is higher than the trend obtained by Martín et al. (2012). In Table 11, we have also included the absolute maxima and minima recorded. There is a small difference when comparing the extreme values. Both the highest (29.6°C on 08/28/1988) and lowest (−8.0°C on 02/13/1983) temperature records in WMO30 are also the absolute maximum and minimum in the historical series of the station since 1916, thus increasing the differences.
In Table 12, we present the same comparison with the average normal values in WMO30 for RH and PCP. As in Table 10, we averaged the results by season. To homogenize the values, we have included the effect of the trend in precipitation found by García-Herrera et al. (2003). The authors obtained a decreasing tendency for the CIAI-AEMet amount of precipitation of −1.18% per year after analyzing the data for the period 1948-1998 (50 years). We have applied this correction factor year-by-year to the WMO30 series from 1975 to 2007 and calculated the differences with the average PCP values obtained in this study (see Tables 7 and 27). The results are in excellent agreement for all the seasons within the dispersion ranges. This dry tendency is also visible in the decrease of 10 points in á ñ RH . Table 13 compares the frequency of PCP episodes, both for the full series and for the frozen phases. In the first case, the different threshold considered to establish the minimum sensitivity must be taken into account to add a rainfall day (see Section 2). This circumstance explains the increase in the number of precipitation days from WMO30 to the series in this study, contrary to the recorded decrease in the absolute amount of precipitation shown in Table 12. A decrease in the frequency of PCP is detected for the snow and hail phases.

Correlation of Precipitation and the North Atlantic Oscillation Index
The North Atlantic Oscillation (NAO) index (Walker & Bliss 1932) is a well-studied climate index defined as the oscillation of the differences in the sea level pressure between the Icelandic low and the Azores high. The NAO's influence on different parameters governing the northern hemisphere climate is very well established (see Wanner et al. 2001 for a review). In particular, the NAO has been associated with the precipitation cycle in the Canary Islands (García- Herrera et al. 2001), one of the most southerly locations modulated by the NAO. Following the general atmospheric circulation model, negative phases of NAO (the Azores and Icelandic systems being weaker than normal) coincide with increases in the Canarian winter precipitation (anti-correlation). Nevertheless, the correlation is not homogeneous for all the islands and also depends on the type of atmospheric disturbance inducing the precipitation, as shown by García-Herrera et al. (2001), who found a correlation of ρ=−0.3 for the CIAI-AEMet station in different types of synoptic rainfall scenarios.
In Figure 10, we have plotted the NAO index series averaged for the winter months (January to March, the months when PCP is more significantly voluminous), and the cumulated precipitation for the same months (see Table 27 in Appendix D). We found a strong correlation between winter PCP and the negative phase of the NAO (ρ = −0.65). The value of ρ decreases to −0.12 when the same winter months are included separately. This result confirms the usefulness of the general atmospheric circulation model in the Canary Islands Observatories.

Summary and Conclusions
In this paper, we have performed a statistical analysis of 10year meteorological data series recorded at Izaña (CIAI-AEMet station) with the purpose of describing the mesoscale climate conditions governing the Canary Islands Observatories. The work was motivated by the need to provide necessary information for the site selection campaign of the CTA.
Here, we briefly summarize the main features found for temperature, precipitation, and relative humidity.  • T M exceeds 0°C∼98% of days; therefore, melting conditions are reached on an elevated percentage of days. • The average difference between T M and T m is ≈7°C.
• The nighttime T distribution is more concentrated than in daytime. • TR has low values (median=7.1°C) and a very low dispersion (MAD=1.1°C). Focusing on the nighttime, the TR is really low, with a median of 2.7°C and MAD=0.9°C. • T shows standard northern hemisphere seasonal behavior, with a maximum in July (18.9°C) and minimum in February (3.8°C) • The most thermally stable nights occur in August (lowest TR). • The warmest year in the 10-year timeseries was 2010 (10.8°C) and the coolest 2004 (9.0°C). The dispersion values show differences between different years being always less than 1.5°C. • Very good agreement has been found between the average temperature with the latest WMO climatological normals published for CIAI-AEMet within a GW trend of +0.09°C/ decade. The trends estimated for the whole series and the different seasons agree with the published values, except for winter, where no significant trends have been obtained.
410;420 <145 mm/24h 100% Figure 8; Table 25 PCP<70 mm/h 410;420 <27 mm/h 100% Figure 7; Table 26 Ground snow accumulation>500 mm 520 Estimation * 100% * Figure 41 Hail larger than 20 mm in diameter<0.03a −1 540 Estimation * 100% * Figure 7; Table 26 Table • PCP is even less frequent and intense during nighttime (by a factor of 1.6). • The PCP shows excellent agreement with the values published in the latest WMO climatological normals for CIAI-AEMet after being corrected for the published climate yearly trends. • A strong anti-correlation (ρ = −0.65) has been found between the PCP series and the NAO index for the winter months (the months with more precipitation). • All the requirements specified in the CTA project documentation are 100% fulfilled, except for the maximum temperature, which is 99.4% fulfilled.
In conclusion, the results obtained confirm that the Canary Islands Observatories are ideal locations to host telescopes or facilities such as CTA, and fulfill all meteorological and climatic requirements and thresholds. This work was funded by the Instituto de Astrofísica de Canarias (IAC) as part of its participation in the site selection campaign for the Cherenkov Telescope Array (CTA). We have used 10 years of data recorded at Izaña Observatory (IZO) by the Centro de Investigación Atmosférica de Izaña (CIAI) belonging to the Spanish Agencia Estatal de Meteorología (AEMet). We are grateful to them for kindly sharing their data with us. We particularly thank Ricardo Sanz of the AEMet Delegation in Santa Cruz de Tenerife for his constant help in preparing and sharing the data from CIAI-AEMet, for the facilitation of the original version of the report with the climatological normals for the period 1961-1990(INM 1995, and for his always useful suggestions. The North Atlantic Oscillation indices were downloaded from the webpage of the US National Center for Environmental Prediction (National Weather Service-NOAA). We also acknowledge Irene Puerto and Ramón García López of the CTA team at the IAC and Antonia M. Varela of the IAC Sky Quality Team for their great willingness in preparing the original report and for their productive discussions and revisions of the results. Finally, we thank Terry Mahoney (IAC) for his English language corrections.

Appendix A Temperature Plots
In this Section we show all the set of temperature plots in the study. The plots include the four analyzed parameters (T, T M , T m and TR; see Section 3) for the all time, nighttime, and daytime series, and different ranges.

A.1. Timeseries, Histograms, and Cumulative
The following plots (Figures 11 to 22) show the global statistical results through time series, histograms and cumulative curves for all the temperature parameters.

A.2. Monthly Behavior
The following plots (Figures 23 to 25) show the monthly statistical results for all the temperature parameters.

A.3. Yearly Behavior
The following plots (Figures 26 to 28) show the yearly statistical results for all the temperature parameters.

Appendix B Precipitation and Relative Humidity Plots
In this Section we show all the set of precipitation and relative humidity plots in the study. The plots include the analyzed parameters (hourly PCP, daily PCP and RH; see Section 4) for the all time, nighttime, and daytime series, and different ranges. The results also shown separated liquid and frozen precipitation.

B.1. Timeseries, Histograms, and Cumulative
The following plots (Figures 29 to 37) show the global statistical results in time series, histograms and cumulative curves for precipitation and relative humidity.

B.2. Monthly Behavior
The following plots (Figures 38 to 40) show the monthly statistical results for precipitation and relative humidity.

B.3. Yearly Behavior
The following plots (Figures 41 to 43) show the yearly statistical results for precipitation and relative humidity.

Appendix C Temperature Tables
This is the final compendium of all the numerical temperature statistical results (Tables 14 to 23).

Table 14
Temperature (T) Statistical Results from the whole Database. The Available Samples are "Daytime" (D), "Nighttime" (N) and "All Time" (A=D+N), and C is the Percentage of Coverage in the Period.

Appendix D Precipitation and Relative Humidity Tables
This is the final compendium of all the numerical PCP and RH statistical results (Tables 24 to 33).