• Barnett, A., , Hatton D. B. , , and Jones D. W. , 1998: Recent changes in thermometer design and their impact. Instrument and Observing Methods Rep. 66, WMO/TD-No. 871, 12 pp.

    • Search Google Scholar
    • Export Citation
  • Barnett, E. W., , and Suomi V. E. , 1949: Preliminary report on temperature measurement by sonic means. J. Meteor., 6, 273276.

  • Favier, V., , Agosta C. , , Genthon C. , , Arnaud L. , , Trouvilliez A. , , and Gallée H. , 2011: Interpreting an 18-month surface heat budget in the coastal blue ice area of Adelie Land, Antarctica. J. Geophys. Res., 116, F03017, doi:10.1029/2010JF001939.

    • Search Google Scholar
    • Export Citation
  • Genthon, C., , Town M. S. , , Six D. , , Favier V. , , Argentini S. , , and Pellegrini A. , 2010: Meteorological atmospheric boundary layer measurements and ECMWF analyses during summer at Dome C, Antarctica. J. Geophys. Res., 115, D05104, doi:10.1029/2009JD012741.

    • Search Google Scholar
    • Export Citation
  • Georges, C., , and Kaser G. , 2002: Ventilated and unventilated air temperature measurements for glacier-climate studies on a tropical mountain site. J. Geophys. Res., 107, 4775, doi:10.1029/2002JD002503.

    • Search Google Scholar
    • Export Citation
  • Georgiadis, T., , Argentini S. , , Mastrantonio G. , , Viola A. , , Sozzi R. , , and Nardino M. , 2002: Boundary layer convective-like activity at Dome Concordia, Antarctica. Nuovo Cimento, 25C, 425431.

    • Search Google Scholar
    • Export Citation
  • Grenfell, T. C., , Warren S. G. , , and Mullen P. C. , 1994: Reflection of solar radiation by the Antarctic snow surface at ultraviolet, visible and near-infrared wavelength. J. Geophys. Res., 99 (D9), 18 66918 684.

    • Search Google Scholar
    • Export Citation
  • Huwald, H., , Higgins C. W. , , Boldi M.-O. , , Bou-Zeid E. , , Lehning M. , , and Parlange M. B. , 2009: Albedo effects on radiative errors in air temperature measurements. Water Resou. Res., 45, W08431, doi:10.1029/2008WR007600.

    • Search Google Scholar
    • Export Citation
  • Lin, K., , Hubbard G. , , Walter-Shea E. A. , , and Brandle J. R. , 2001: Some perspectives on recent in situ air temperature obserations: Modeling the microclimate inside the radiation shields. J. Atmos. Oceanic Technol., 18, 14701484.

    • Search Google Scholar
    • Export Citation
  • Richardson, S. J., , Brock F. , , Semmer S. R. , , and Jirak C. , 1999: Minimizing errors associated with multiplate radiation shields. J. Atmos. Oceanic Technol., 16, 18621872.

    • Search Google Scholar
    • Export Citation
  • Warren, S. G., , Brandt R. E. , , and O’Rawe Hinton P. , 1998: Effect of surface roughness on bidirectional reflectance of Antarctic snow. J. Geophys. Res., 103 (E11), 25 78925 807.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    The (right) naturally ventilated and (left) mechanically aspirated radiation shields on the Dome C tower.

  • View in gallery

    (top) Temperature 3.5 m above surface, and (bottom) wind speed 41.9 m above surface, in 2009.

  • View in gallery

    Calculated incoming solar irradiation on a horizontal surface at (left) the top of the atmosphere and (right) the cosine of solar angle. The red curve is the 24-h running mean.

  • View in gallery

    The FV (black) and NV (red) (top) temperatures and (middle) difference FV − NV from the end of 2009 (negative days are before 1 January 2010) to mid-2010. Interruptions in January–February are due to datalogging failures. (bottom) The histogram of number of bias events as a function of bias range.

  • View in gallery

    (top) Temperature reports at the lowest tower level from FV (3.5 m, black), NV (3.5 m, red), and SO (8.4 m, green) at the end of 2009; and (bottom) the difference of NV and SO with respect to FV.

  • View in gallery

    Scatterplots of radiation bias (°C, y axis) against (a) air temperature, (b) incoming direct + diffuse solar radiation at the surface, (c) wind, and (d) cosine of solar zenith angle μ when diffuse is less than 10% of total solar radiation.

  • View in gallery

    Radiation bias (°C) as a function of wind speed (m s−1) and incoming (top) direct and (bottom) diffuse solar radiation (W m−2).

  • View in gallery

    Samples of automatic weather station (black, 170 cm above surface) and FV (red, 3.5 m above surface) atmospheric temperatures in (top) Jan and (bottom) May 2009.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 30 30 14
PDF Downloads 19 19 10

Atmospheric Temperature Measurement Biases on the Antarctic Plateau

View More View Less
  • 1 Laboratoire de Glaciologie et Géophysique de l’Environnement, CNRS/UJF, Saint Martin, d’Hères, France
  • 2 Antarctic Meteorological Research Center, University of Wisconsin—Madison, Madison, Wisconsin
© Get Permissions
Full access

Abstract

Observations of atmospheric temperature made on the Antarctic Plateau with thermistors housed in naturally (wind) ventilated radiation shields are shown to be significantly warm biased by solar radiation. High incoming solar flux and high surface albedo result in radiation biases in Gill (multiplate)-styled shields that can occasionally exceed 10°C in summer in cases with low wind speed. Although stronger and more frequent when incoming solar radiation is high, biases exceeding 8°C are found even when solar radiation is less than 200 W m−2. Compared with sonic thermometers, which are not affected by radiation but are too complex to be routinely used for mean temperature monitoring, commercially available aspirated shields are shown to efficiently protect thermistor measurements from solar radiation biases. Most of the available in situ reports of atmospheric temperature on the Antarctic Plateau are from automatic weather stations that use passive shields and are thus likely warm biased in the summer. In spite of low power consumption, deploying aspirated shields at remote locations in such a difficult environment may be a challenge. Bias correction formulas are not easily derived and are obviously shield dependent. On the other hand, because of a strong dependence of bias to wind speed, filtering out temperature reports for wind speed less than a given threshold (about 4–6 m s−1 for the shields tested here) may be an efficient way to quality control the data, albeit at the cost of significant data loss and records that are biased toward high wind speed cases.

Corresponding author address: Christophe Genthon, Laboratoire de Glaciologie et Géophysique de l’Envrironnement, UJF – Grenoble 1/CNRS, LGGE UMR 5183, Grenoble, F-38041, France. E-mail: genthon@lgge.obs.ujf-grenoble.fr

Abstract

Observations of atmospheric temperature made on the Antarctic Plateau with thermistors housed in naturally (wind) ventilated radiation shields are shown to be significantly warm biased by solar radiation. High incoming solar flux and high surface albedo result in radiation biases in Gill (multiplate)-styled shields that can occasionally exceed 10°C in summer in cases with low wind speed. Although stronger and more frequent when incoming solar radiation is high, biases exceeding 8°C are found even when solar radiation is less than 200 W m−2. Compared with sonic thermometers, which are not affected by radiation but are too complex to be routinely used for mean temperature monitoring, commercially available aspirated shields are shown to efficiently protect thermistor measurements from solar radiation biases. Most of the available in situ reports of atmospheric temperature on the Antarctic Plateau are from automatic weather stations that use passive shields and are thus likely warm biased in the summer. In spite of low power consumption, deploying aspirated shields at remote locations in such a difficult environment may be a challenge. Bias correction formulas are not easily derived and are obviously shield dependent. On the other hand, because of a strong dependence of bias to wind speed, filtering out temperature reports for wind speed less than a given threshold (about 4–6 m s−1 for the shields tested here) may be an efficient way to quality control the data, albeit at the cost of significant data loss and records that are biased toward high wind speed cases.

Corresponding author address: Christophe Genthon, Laboratoire de Glaciologie et Géophysique de l’Envrironnement, UJF – Grenoble 1/CNRS, LGGE UMR 5183, Grenoble, F-38041, France. E-mail: genthon@lgge.obs.ujf-grenoble.fr

1. Introduction

Surface meteorological measurements in the Antarctic are conducted at manned stations as well as via automatic weather stations (AWSs). Other than a few exceptions, almost all of the manned stations are found along the coast of the continent. Hence, automatic weather stations provide the bulk of the observations in the interior of the continent. Cold temperatures and high incident solar radiation make the measurement of atmospheric temperature on the Antarctic Plateau in summer particularly sensitive to radiation-induced biases. This is because the surface albedo of the Antarctic snow is very high (Grenfell et al. 1994). Thus, the temperature measurements may be affected not only by the downward incoming but also by the upward-reflected solar radiation. Most common radiation shields (e.g., Gill) tend to offer less protection to reflected upwelling than to incoming downwelling radiation (Richardson et al. 1999). Also, in the summer, incoming solar radiation may be very high and persistent. In fact, the largest daily mean is found at high latitudes in summer because of the high solar angle and permanent daylight.

Cold temperatures result in comparatively low thermal emission of temperature sensors, and thus a comparatively higher sensitivity to solar radiation. Sensors that directly measure the temperature of the air, for example, sonic anemothermometers (SOs), rather than a device that is expected to be at the same temperature as the air, are supposed to be little affected by solar radiation (e.g., Barnett and Suomi 1949). However, higher absolute accuracy is obtained and lower maintenance and energy supply are required by thermistors that are available at a much lower cost. Thus, thermistors are the most generally employed temperature sensors for automatic logging in such an environment.

Unshielded thermistors are known to be affected by radiation. The World Meteorological Organization (WMO) has evaluated various kinds of mainly passively ventilated radiation shields for robust temperature measurements (e.g., Barnett et al. 1998). Passively ventilated shields are naturally ventilated by the wind. In addition, forced ventilation shields have also been used to measure temperature in cold environments over snow and ice surfaces (e.g., Georges and Kaser 2002). However, apparently little has been done in this respect in the Antarctic environment, possibly because of limited energy resources and the logistical access required to operate and maintain ventilation. Here we present results of a comparison of temperatures recorded at Dome C on the Antarctic Plateau by temperature sensors that are housed in various kinds of shields and those that do not need shields. Large differences are recorded between mechanically aspirated and naturally ventilated shields.

2. Instruments

Genthon et al. (2010) analyzed the meteorological observations made in the summer of 2008 along a 45-m tower at Dome C, Antarctica (75°06′S, 123°20′E, and 3233 m ASL). The temperature sensors were Campbell HMP45Cs, housed in a Campbell Scientific URS1 passively ventilated shield. Genthon et al. (2010) reported occurrences of suspiciously warm events in low wind speed conditions, which were unlikely to reflect the real atmosphere and were most probably due to solar radiation. Corrections suggested by Huwald et al. (2009) were tested but were obviously inadequate. Such corrections are most probably shield dependent and possibly even site dependent. Additional sensors and new shields were deployed in early 2009, which now allow a systematic comparison of different kinds of shields in a large range of wind, temperature, solar radiation, and solar angle conditions over more than a full year.

Table 1 summarizes the various temperature sensors at Dome C used in the present study. For further general information about the overall setting at Dome C and the tower facility there, see Genthon et al. (2010). Three kinds of temperature sensors are available on the tower in 2009: Campbell Scientific HMP45C and Vaisala HMP155 thermohygrometers in passively ventilated Campbell URS1 Gill-styled radiation shields; PT100 thermistors in fan-aspirated Young 43502 radiation shields; and Applied Technology SAT-SX sonic anemothermometers. A PT100 thermistor is a platinum resistor, the resistance of which varies around 100 Ω depending on the temperature. The resistance is accurately measured using a Wheatstone bridge. Temperature measurements in the HMP155 sensor are actually also made with an internal PT100 resistor (PT1000 in HMP45C), so the main difference between the HMP and PT100 sensors here, besides sensor conditioning, is the kind of radiation shield that is used to house the sensors. Henceforth, they will be referred to as the naturally ventilated (NV) and force ventilated (FV) sensors, respectively. The accuracy of a PT100/1000 thermistor is typically ±(0.2°–0.5°)C depending on temperature, but that is the temperature of the sensor itself, not necessarily that of the air around. SOs measure the speed of sound, which depends, among other parameters, on the temperature of the air. Various other factors are involved including air pressure, air moisture, and wind speed. While the measurement of temperature fluctuations with the SO may be much more accurate, the absolute accuracy is not better than 1°. On the other hand, the SOs directly sense the temperature of the air, not that of an intermediate device (e.g., a piece of platinum in a PT100), the energy balance of which may be affected by radiation. No radiation shield is thus required for SO instruments.

Table 1.

Various temperature sensors at Dome C, shield type, location, and elevation above surface.

Table 1.

The NV and FV sensors are deployed side by side at the same levels on the tower (Fig. 1), at 3.5, 10.6, 18.0, 25.3, 32.7, and 41.9 m above the surface. Young 05106 aerovanes are also deployed at the same levels. Both NV and FV sensors are factory validated to −80°C for the HMP155 and PT100, and −40°C for the HMP45C. The Young 05106 aerovane clearly fails below −50°C (Genthon et al. 2010). However, we found that by removing the grease on the bearings, at the risk of increased wear, the aerovanes work fine at lower temperatures and provide a much more continuous record than the SO. However, we do not expect accuracy on the wind measurements to be quite the factory-stated nominal one (of ±0.3 m s−1). Mechanical aspiration in the FV is done by electric fans, with the airflow around the sensors being 5–10 m s−1. The airflow is bottom up, with the fan at the top, avoiding risk that a heating fan motor may affect the air temperature around the thermometer. The SOs are not quite at the same elevation as the other instruments (Table 1). In the case of a strong inversion, the fact that the sensors are not exactly at the same elevation should be taken into account when comparing the data; however, the strong inversions occur when incoming solar radiation is either low or nil, which is not in summer.

Fig. 1.
Fig. 1.

The (right) naturally ventilated and (left) mechanically aspirated radiation shields on the Dome C tower.

Citation: Journal of Atmospheric and Oceanic Technology 28, 12; 10.1175/JTECH-D-11-00095.1

The NV and FV sensors are interrogated every 10 s, then averaged over 30 min. The SO sampling rate is 10 Hz, and all of the data are retained for turbulence studies. Various problems have occasionally affected the operation of the instruments. However, the data used here sample an ample record of cases of radiation-induced warm biases for a large range of incoming radiation and wind conditions. Figure 2 displays the temperature recorded in 2009 by the lowermost FV. Figure 2 also shows wind at the upper level. They range from −73° to −23°C, and from 0 to 17 m s−1, respectively. The sensors have been variously affected by the extreme polar conditions at Dome C, and the records are not fully continuous. In particular, the SOs are very sensitive to frost deposition that frequently occurs at Dome C. Therefore, the SO records are highly discontinuous in spite of periodic heating. The SO data here can thus only be used to validate that the FV are little, if any, affected by solar radiation.

Fig. 2.
Fig. 2.

(top) Temperature 3.5 m above surface, and (bottom) wind speed 41.9 m above surface, in 2009.

Citation: Journal of Atmospheric and Oceanic Technology 28, 12; 10.1175/JTECH-D-11-00095.1

Downward solar radiation at the surface is available at Dome C from the Baseline Surface Radiation Network (BSRN; online at http://www.bsrn.awi.de/). BSRN monitoring at Dome C is operated by the Institute of Atmospheric Sciences and Climate (ISAC) of the Italian National Research Council, Bologna, Italy. Figure 3 shows the calculated cosine of the solar zenith angle and incoming solar radiation on a horizontal surface at the top of the atmosphere at Dome C, neglecting the weak contribution of the earth’s eccentricity. It may be expected that radiation biases on temperature measurements depend on both the solar intensity and solar angle resulting from the contribution of reflected radiation (Richardson et al. 1999). Obviously, both are zero during the polar night, when no radiation bias is expected on temperature measurements. Variability in summer is characterized by a strong diurnal cycle resulting from projection on a horizontal surface. Incoming radiation at the top of the atmosphere reaches 876 W m−2 at midday. The atmosphere at Dome C is very clean, and in cloud-free conditions the transfer of solar radiation to the surface is little affected. In 2009, according to BSRN data, solar incoming radiation at the surface reached 827 W m−2.

Fig. 3.
Fig. 3.

Calculated incoming solar irradiation on a horizontal surface at (left) the top of the atmosphere and (right) the cosine of solar angle. The red curve is the 24-h running mean.

Citation: Journal of Atmospheric and Oceanic Technology 28, 12; 10.1175/JTECH-D-11-00095.1

There are several other atmospheric temperature sensors deployed at Dome C, none of which is in forced-ventilated shields. The oldest meteorological station at Dome C was deployed by the Antarctic Meteorological Research Center (AMRC) at the University of Wisconsin (UW; online at http://amrc.ssec.wisc.edu) in 1980. As further topographical studies revealed that the actual top of the dome was ~50 km away, the station was moved in 1995. The overall design of the station (http://amrc.ssec.wisc.edu/aws/) did not change, and the UW Dome C AWS is one of the oldest automatic weather stations almost continuously operating on the Antarctic Plateau. It is located about 3 km from the tower, and the temperature sensor is currently about 1.7 m above the surface. It is a Weed platinum resistance thermometer of 1000 Ω shielded from radiation by a vertical piece of aluminum tube. The tube’s inner surface is black to avoid reflecting solar radiation toward the sensor. The outer surface is Mylar coated. This shield, designed by the UW AWS project, probably efficiently protects from lateral and some downward solar radiation and allows some natural ventilation. In addition to the Mylar covering, the sensor connections at the top of the shield are connected to the shield by a nylon-threaded coupling. The shield is mounted under the boom holding all of the sensors, so the shield is also shaded at the top by the boom. The open bottom of the shield does not protect from the upward-reflected solar radiation. This is a simple design used on many of the automatic meteorological stations in Antarctica. It is thus of interest to evaluate how temperature measurements in such conditions may be radiation biased. This design was an improvement over the original shield prototyped in the original Antarctic AWS by Stanford that was a very small (~8-cm diameter) Gill-styled shield with only three vents. Errors were possible in the temperature readings resulting from the sensor not being insulated from the shield, which resulted in the thermal conduction impacting measurements.

In the following section, we compare and analyze the temperature reports by the various temperature sensors on the Dome C tower. We also correlate the biases with temperature, wind speed, solar radiation, and solar zenith angle, and calculate linear multiple correlations to test for a simple empirical correction. Then, periods during which vertical mixing allows for comparing sensors at somewhat different elevations are selected to evaluate the UW AWS sensor radiation biases.

3. Observed radiation biases and regression

Figure 4 shows the temperatures reported by the NV and FV sensors on the lower tower level and the difference, from the last few days of 2009 to mid-2010. Differences occasionally reach more than 10°C. Temperature differences of more than 2°C, considered well above measurement accuracy, occur more than 6% of the time in summer. During the polar night, differences between the two sensors occasionally reach 2°C or more. Radiation biases are obviously minimal then. Thus, differences less than 2°C are not necessarily explained by radiation and should be considered as noise in the present study, although this is conservative in summer because thermistor accuracy is better at warmer temperature. On the other hand, even in late fall when daylight is weak, albeit not zero, biases that reach more than 4°C occur. It appears that it does not take large amounts of incoming solar radiation to affect the temperature measurements on the Antarctic Plateau.

Fig. 4.
Fig. 4.

The FV (black) and NV (red) (top) temperatures and (middle) difference FV − NV from the end of 2009 (negative days are before 1 January 2010) to mid-2010. Interruptions in January–February are due to datalogging failures. (bottom) The histogram of number of bias events as a function of bias range.

Citation: Journal of Atmospheric and Oceanic Technology 28, 12; 10.1175/JTECH-D-11-00095.1

Figure 5 compares NV, FV, and SO over a period of large differences between NV and FV. The SO and FV reports agree well, with a moderate diurnal cycle in the difference probably reflecting the fact that the SO sensors are off by 3–4 m with respect to the height of the other sensors. Beyond this uncertainty, Fig. 5 demonstrates that the FV data are much more correct than the NV measurements, which are prone to radiation-induced biases. Henceforth, we will refer to the differences between the FV and NV reports as radiation biases.

Fig. 5.
Fig. 5.

(top) Temperature reports at the lowest tower level from FV (3.5 m, black), NV (3.5 m, red), and SO (8.4 m, green) at the end of 2009; and (bottom) the difference of NV and SO with respect to FV.

Citation: Journal of Atmospheric and Oceanic Technology 28, 12; 10.1175/JTECH-D-11-00095.1

Figure 6 displays relationships between radiation bias and temperature, wind, solar radiation, and solar zenith angle. The observations are filtered to concentrate on the late spring, summer, and early fall months when incoming solar radiation is higher and biases are frequent. Cases of temperature below −39°C are left out for HMP45C reports. Finally, we only consider temperature biases above 2°C as discussed above. There are more bias cases at warmer temperature but bias amplitude is not significantly related to temperature (Fig. 6a). The relation to solar radiation is not straightforward (Fig. 6b). Although more frequent and stronger at higher irradiation, large biases occur for all values of solar radiation down to less than 200 W m−2.

Fig. 6.
Fig. 6.

Scatterplots of radiation bias (°C, y axis) against (a) air temperature, (b) incoming direct + diffuse solar radiation at the surface, (c) wind, and (d) cosine of solar zenith angle μ when diffuse is less than 10% of total solar radiation.

Citation: Journal of Atmospheric and Oceanic Technology 28, 12; 10.1175/JTECH-D-11-00095.1

The strongest dependence is found with respect to wind speed (Fig. 6c). The largest biases occur when there is no wind, and biases for winds above 6 m s−1 are insignificantly different from noise. Obviously, a very simple and straightforward way to filter out radiation biases in temperature records produced with the kind of naturally ventilated radiation shields used here is to cut out all reports for winds less than 4–6 m s−1, the latter value being a conservative threshold. In other windier places, for example, the coastal regions of Antarctica (Favier et al. 2011), this may be a very efficient way to process the data. On the Antarctic Plateau, on the other hand, this would leave aside the larger fraction of a dataset.

Solar angle may also be expected to modulate radiation biases in the case of direct exposure because radiation shields tend to protect more efficiently from downward than from upward radiation influx (Richardson et al. 1999). In fact, by construction (Fig. 1), one would expect protection to be minimal at a given upward angle, that is, in a range of solar angles considering the scattering properties and the directional reflectivity of the snow surfaces (Warren et al. 1998). Airborne ice particles and clouds also have scattering effects on radiation transfer in the atmosphere that result in the diffusive component of the incoming solar radiation. Figure 6d shows the temperature bias with respect to solar angle when the diffuse component is less than 10% of the total incoming solar (i.e., mostly clear sky), thus concentrating on the directional effect of the direct radiation. It shows that biases are more frequent and larger when solar zenith angles get smaller, that is, when the sun is higher on the horizon; yet significant biases are found for all zenith angles.

Figure 7 shows the multiple dependencies of the radiation biases with respect to wind speed and incoming solar radiation. The dependence on direct and diffuse solar radiation is shown separately because the two components are anticorrelated and cover relatively (but not fully) separate ranges of radiation values. As might be expected, the largest biases result from a combination of strong radiation exposure and low wind speed. However, similar biases occur for lesser values of diffuse radiation, suggesting that scattered light more efficiently penetrates the shields to affect the sensor inside.

Fig. 7.
Fig. 7.

Radiation bias (°C) as a function of wind speed (m s−1) and incoming (top) direct and (bottom) diffuse solar radiation (W m−2).

Citation: Journal of Atmospheric and Oceanic Technology 28, 12; 10.1175/JTECH-D-11-00095.1

Although separate linear regressions would be low (Fig. 6), multiple dependencies suggest that multiple regressions may be tested to tentatively derive an empirical bias correction. There are few places where direct and diffuse components of solar radiation are separately available or may be accurately estimated, so we only keep total incoming solar radiation. Multiple linear regression between the observed radiation bias and wind, downwelling solar radiation at the surface, and cosine of solar angle yields the parameters and statistics reported in Table 2. Considering the number of available samples (~5000) and values of the t test, all of the correlations are significant. However, the multiple square correlation is only 0.13. Thus, the fraction of variance reconstructed by the multilinear model in Table 2 is very low. In addition, a counterintuitive negative coefficient for incoming solar radiation, which probably reflects the complex contributions of solar angle and diffuse component, illustrates the limit of the approach. While radiation transfer and flow modeling both inside and around the shields would obviously be a much more defensible approach (Richardson et al. 1999; Lin et al. 2001), a simple regression illustrates that deriving a robust correction for radiation biases of temperature observations made in naturally ventilation shields is unlikely to be an easy task at a place like Dome C. Such correction is unlikely to be an alternative to using force-ventilated rather than passively ventilated shields when possible.

Table 2.

Parameters of multiple linear regression of bias B against wind W, incoming solar radiation at surface S and cosine of zenith solar angle A. The model is B = CwW + CsS + CaA + I.

Table 2.

Finally, Fig. 8 compares temperature reports by the UW AWS at Dome C with the FV observations at the lower level on the tower in January 2009. The temperature sensors are not quite at the same elevation above the surface (1.7 m for the UW AWS sensor, 3.5 m for the FV sensor). However, convection occurs in the early afternoon, which efficiently mixes the lower atmosphere (Georgiadis et al. 2002; Genthon et al. 2010). Therefore, although some of the differences between the sensor reports may be related to the building of an inversion at night, they mainly reflect the radiation bias in the afternoon. Because the terrain is very flat and homogeneous, comparing two sensors 3 km apart is valid. It is clear that the automatic station reports are seriously warm biased in summer. Figure 8 also shows a similar time sample for comparison in late spring when solar radiation is very low. This occasionally shows very good agreement; otherwise, the automatic weather station sensors report colder temperature, consistent with the fact that the sensor is closer to the surface in an environment characterized by increasingly strong temperature inversions.

Fig. 8.
Fig. 8.

Samples of automatic weather station (black, 170 cm above surface) and FV (red, 3.5 m above surface) atmospheric temperatures in (top) Jan and (bottom) May 2009.

Citation: Journal of Atmospheric and Oceanic Technology 28, 12; 10.1175/JTECH-D-11-00095.1

The different heights of the AWS and FV temperature sensors make quantitative direct comparison difficult. When the FV temperature is the same within 0.5°C all along the tower, the assumption is made that the air is well mixed by turbulence and/or convection, and the air temperature should be the same all the way down to the level of the automatic weather station. Only then can the AWS reports be directly compared to the FV reports on the tower to evaluate the radiation bias of the AWS measurements. A warm bias above 2°C occurs 80% of the time in such a situation. A bias above 5°C occurs 47% of the time at the AWS station, but only 11% of the time on the tower for the NV measurements. The naturally ventilated Gill-styled shields on the tower, although clearly inefficient to avoid all of the radiation biases, thus do a better job than the tubes that are used on the AWS. Other AWS stations in similar situations are likely to be similarly affected elsewhere.

4. Final remarks and conclusions

Atmospheric temperature measurements with thermistors housed in naturally ventilated radiation shields are affected by solar radiation biases on the Antarctic Plateau. The mean difference between naturally ventilated and aspirated measurements over the period of analysis here, excluding winter (with solar radiation less than 100 W m−2) and cases of recorded differences less than 2°C (to retain only unambiguously significant biases), is 3.5°C. The difference occasionally reaches more than 10°C. These are very significant biases that are bound to affect the use of such observations to analyze the local meteorology, build a climatology, or validate meteorological and climate models. It is thus strongly recommended that either sensors that are insensitive to solar radiation, for example, calibrated sonic thermometers, or sensors that are housed in radiation shields with forced ventilation are used in such environment. To our knowledge, very few of the meteorological systems that monitor the meteorology and climate of Antarctica use such devices. At Dome C, the only forced-ventilated systems are those cited in this work.

The automatic weather stations, an invaluable source of meteorological and climate data over the last 30 yr in an otherwise largely data-void region, also use naturally ventilated shields. When comparing the reports from the Dome C AWSs with the tower data when convective instability mixes the lower atmosphere, thus avoiding the effects of differences of sensors height above the surface, it is found that the AWS data are significantly warm biased as much as 80% of the time in low-speed winds during austral summer. AWSs are or have been operating at more than 100 sites in Antarctica. The reports are passed on to the Global Telecommunication System (GTS) to be potentially operationally used by weather analysis and forecasting services worldwide and to produce reanalyses. Obviously, there is no solar radiation bias when there is no radiation, so such biases are expected to be nil in winter. In addition, Antarctica is a windy place. Winds are weak on the plateau, particularly at the summit of a dome where there is no locally produced katabatic flow. Thus, one may expect that many of the automatic weather stations and, more generally, many of the observations of atmospheric temperature in Antarctica are less affected by radiation biases than shown here because they are made in winter and/or in windier places. However, one should be aware of potentially strong biases when using available temperature reports in Antarctica.

Obviously, existing and future meteorological systems in Antarctica should be adapted and planned in ways that avoid radiation biases on temperature measurements. The use of forced ventilation is customary in other regions of the world, but not in Antarctica, most obviously because the logistical and environmental constraints make it more difficult to implement ventilation. Historically power available to an AWS prohibited the deployment of such devices. The results presented here show that commercially aspirated radiation shields exist that do a good job at minimizing the radiation bias. Two years of operation have shown that the Young 43502 can withstand extreme conditions, such as on the Antarctica Plateau, reasonably well. However, this is next to a permanently manned Antarctic station, so power is not an issue. Although the power requirement is low for the shields used here (only 5 W), it may be difficult to consistently ensure this requirement in Antarctica. Autonomous systems generally work on solar panels. It may be an issue during the long polar night; however, this is also when ventilation is not expected to be required because there is no sunlight. On the other hand, it is found here that even low downwelling solar radiation in a number of cases induces significant warm biases. Twilight may thus be the most critical time to operated robust atmospheric temperature measurements in remote Antarctica. In the meantime, efforts are underway by the UW to replace all of the existing radiation shields with standard Gill shields that are expected to have better performance than the current shields.

Although no simple universal correction to temperature measurements made in passively ventilated shields could be extracted from the observations presented here, one way to minimize the bias is to retain only the observations that are made when the wind speed is larger than a given value. This value would be of the order of 4–6 m s−1 for the URS1 shields in the Dome C environment, but it is obviously shield dependent. Such filtering may be at the cost of significant data loss and may bias the observations against calm anticyclonic situations, thus affecting the climatological significance of the remaining data. This is likely to be a major limitation on the Antarctic Plateau, though probably much less so in coastal regions scoured by strong and frequent katabatic airflow. Considering how important the issue is, the UW AWS project plans to continue to experiment with different configurations of shields, temperatures sensors, and power for forced ventilation in order to find the most effective way to resolve the radiation bias problem on the high polar plateau.

Acknowledgments

Support in the field by the French (IPEV, Program CALVA-1013) and Italian (PNRA) polar institutes is acknowledged. This is a contribution to the CNES/INSU CONCORDIASI IPY Project. Meteorological data are obtained as part of the OSUG-CENECLAM observatory. We thank ISAC for access to the solar radiation data distributed by BSRN (http://www.bsrn.awi.de/). Some of this material is based upon work partially supported by the U.S. NSF-OPP under Grant ANT-0944018.

REFERENCES

  • Barnett, A., , Hatton D. B. , , and Jones D. W. , 1998: Recent changes in thermometer design and their impact. Instrument and Observing Methods Rep. 66, WMO/TD-No. 871, 12 pp.

    • Search Google Scholar
    • Export Citation
  • Barnett, E. W., , and Suomi V. E. , 1949: Preliminary report on temperature measurement by sonic means. J. Meteor., 6, 273276.

  • Favier, V., , Agosta C. , , Genthon C. , , Arnaud L. , , Trouvilliez A. , , and Gallée H. , 2011: Interpreting an 18-month surface heat budget in the coastal blue ice area of Adelie Land, Antarctica. J. Geophys. Res., 116, F03017, doi:10.1029/2010JF001939.

    • Search Google Scholar
    • Export Citation
  • Genthon, C., , Town M. S. , , Six D. , , Favier V. , , Argentini S. , , and Pellegrini A. , 2010: Meteorological atmospheric boundary layer measurements and ECMWF analyses during summer at Dome C, Antarctica. J. Geophys. Res., 115, D05104, doi:10.1029/2009JD012741.

    • Search Google Scholar
    • Export Citation
  • Georges, C., , and Kaser G. , 2002: Ventilated and unventilated air temperature measurements for glacier-climate studies on a tropical mountain site. J. Geophys. Res., 107, 4775, doi:10.1029/2002JD002503.

    • Search Google Scholar
    • Export Citation
  • Georgiadis, T., , Argentini S. , , Mastrantonio G. , , Viola A. , , Sozzi R. , , and Nardino M. , 2002: Boundary layer convective-like activity at Dome Concordia, Antarctica. Nuovo Cimento, 25C, 425431.

    • Search Google Scholar
    • Export Citation
  • Grenfell, T. C., , Warren S. G. , , and Mullen P. C. , 1994: Reflection of solar radiation by the Antarctic snow surface at ultraviolet, visible and near-infrared wavelength. J. Geophys. Res., 99 (D9), 18 66918 684.

    • Search Google Scholar
    • Export Citation
  • Huwald, H., , Higgins C. W. , , Boldi M.-O. , , Bou-Zeid E. , , Lehning M. , , and Parlange M. B. , 2009: Albedo effects on radiative errors in air temperature measurements. Water Resou. Res., 45, W08431, doi:10.1029/2008WR007600.

    • Search Google Scholar
    • Export Citation
  • Lin, K., , Hubbard G. , , Walter-Shea E. A. , , and Brandle J. R. , 2001: Some perspectives on recent in situ air temperature obserations: Modeling the microclimate inside the radiation shields. J. Atmos. Oceanic Technol., 18, 14701484.

    • Search Google Scholar
    • Export Citation
  • Richardson, S. J., , Brock F. , , Semmer S. R. , , and Jirak C. , 1999: Minimizing errors associated with multiplate radiation shields. J. Atmos. Oceanic Technol., 16, 18621872.

    • Search Google Scholar
    • Export Citation
  • Warren, S. G., , Brandt R. E. , , and O’Rawe Hinton P. , 1998: Effect of surface roughness on bidirectional reflectance of Antarctic snow. J. Geophys. Res., 103 (E11), 25 78925 807.

    • Search Google Scholar
    • Export Citation
Save