Geophysical Research Letters

Unprecedented upper-air dropsonde observations over Antarctica from the 2010 Concordiasi Experiment: Validation of satellite-retrieved temperature profiles

Authors


Corresponding author: Junhong Wang, National Center for Atmospheric Research, Boulder, CO, USA.

E-mail: junhong@ucar.edu

Abstract

[1] The 2010 Concordiasi field experiment took place over Antarctica from September to December 2010. During Concordiasi, for the first time, 13 National Center for Atmospheric Research Driftsonde systems were launched from McMurdo station, ascended to the stratosphere, and then drifted with the winds. The Driftsonde provides a unique platform to release dropsondes that measure the atmosphere from the lower stratosphere to the surface in otherwise difficult to reach parts of the globe. A total of 639 soundings were obtained and provided unprecedented high quality profiles over Antarctica. The sounding temperature profiles are compared with matched profiles from ten satellite products. All satellite products except The Constellation Observing System for Meteorology, Ionosphere, and Climate (COSMIC) are consistent colder than the sounding data, with larger discrepancies over the Antarctic continent than the coast and ocean. The COSMIC data are in agreement with the sounding data and display no degradation over the continent.

1 Introduction

[2] Antarctica plays an important role in the global climate system through teleconnections, Antarctic ice sheet changes, ozone depletion, and in other ways [SCAR, 2009]. However, it is very challenging to make measurements over Antarctica. Polar-orbiting passive satellites, GPS radio-occultation from COSMIC, and radiosondes are the only observing systems that provide routine upper-air observations over Antarctica. All of 16 operational radiosonde stations over Antarctica except Amundsen-Scott (at the South Pole) are along the coast, so they do not provide measurements over the deep interior of the Antarctic continent (Figure 1). Also, the performance of radiosonde sensors is degraded at cold temperatures [Skony et al., 1994]. Complex, heterogeneous surface conditions over Antarctica introduce significant errors to satellite temperature retrievals. As a result of the lack of in-situ upper-air measurements over Antarctica, satellite retrievals have not been well validated, especially over the Antarctic continent.

Figure 1.

The upper panel shows map of dropsonde locations (yellow squares), radiosonde stations (big purple balloons), and locations of 52 dropsondes (small colored balloons) from 21 October to 9 November 2010 on a single Driftsonde. The lower panel shows corresponding temperature profiles for the 52 sondes with a 3°C offset added to each profile from left to right.

[3] In order to fill gaps of upper-air observations over remote areas such as Antarctica, the National Center for Atmospheric Research (NCAR) has developed its Driftsonde system (S. A. Cohn, T. Hock, J. Wang, and others, Driftsondes: Providing In-Situ Dropsonde Observations over Remote Regions, submitted to Bull. Am. Meteorol. Soc., 2013). The objective of the Driftsonde system is to provide cost-effective upper-air observations over oceans and remote areas from days to months. The Driftsonde has promising science applications, including validating satellite remote sensing data and improving retrieval techniques. For example, the Driftsonde data collected during the THORPEX-Pacific Asian Regional Campaign (T-PARC) were used to validate satellite and global reanalysis products [Wang et al., 2010]. In this study, we present an unprecedented upper-air dataset over Antarctica collected from the NCAR Driftsonde system during the Concordiasi experiment in 2010 [Rabier et al., 2010]. The driftsonde data along with the radiosonde data are used to evaluate temperature profiles from ten satellite products, and the discrepancies between the sonde and satellite data are investigated.

2 Field Campaign, Data, and Analysis Method

[4] The NCAR Driftsonde system consists of a stratospheric balloon attached to a gondola that contains up to 56 Miniature In-situ Sounding Technology (MIST) dropsondes. The balloon is lifted up from the ground to the stratosphere and drifts with the wind. Sondes can be dropped either at a prescheduled time or by command from the ground. The Driftsonde system has been used in three large field projects, African Monsoon Multidisciplinary Analysis in 2006 [Redelsperger et al., 2006], T-PARC in 2008 [Parsons et al., 2008], and Concordiasi in 2010 [Rabier et al., 2010].

[5] The MIST sonde uses the same pressure/temperature/humidity sensor module as is used in the Vaisala RS92 radiosonde [Vaisala, 2012a], and the accuracy of this module, especially the temperature sensor, is well documented. Based on the manufacture's technical data, the RS92 capacitive wire temperature sensor has an accuracy of 0.5°C when the Vaisala Ground Check Set (GC25) is used to perform the ground check [Vaisala, 2012a]. The temperature measurement is subject to calibration, solar heating, and sensor response time errors. The GC25 is used to correct the calibration error by comparing the sonde temperature measurement on the ground with a reference sensor inside the GC25. The six-year (2006–2011) GC temperature data at Lindenberg, Germany show a consistent warm bias with a mean value of ~0.15°C [Holger Vömel, 2011, personal communication]. The NCAR dropsonde group also tested one MIST sonde in our calibration chamber and found a mean warm bias of 0.16°C. By applying the GC temperature correction, the warm bias is removed in the RS92 data. However, it is impractical to use the GC25 for dropsondes stored inside the unmanned driftsonde gondola. Therefore, the calibration bias of ~0.15°C exists in the Concordiasi dropsonde data. The capacitive wire is very small and thus responds quickly to temperature changes, with a response time of less than 1s at 100 hPa [Vaisala, 2012a]. The solar radiation error has a maximum value of 0.98°C at 1 hPa and 0° solar zenith angle and mainly depends on solar zenith angle and pressure [Vaisala, 2012b]. The Vaisala RS92 demonstrated its accurate performance in the WMO radiosonde intercomparison project in 2010 and showed a temperature accuracy of 0.3°C and 0.6°C from the surface to 100 hPa and from 100 hPa to 10 hPa, respectively [Nash et al., 2011]. Therefore, the MIST dropsonde temperature measurement can be considered as a reference to validate the satellite products.

[6] Concordiasi is a joint French-United States initiative that began during the International Polar Year. The Concordiasi field experiment in 2010 took place over Antarctica from September to December 2010 [Rabier et al., 2010]. The scientific objective of Concordiasi was to combine innovative measurements and modeling for better analysis and prediction of the weather over Antarctica. During Concordiasi, a total of 639 carefully quality controlled MIST soundings were collected during 13 Driftsonde flights launched from McMurdo station (Figure 1). The quality control process is described in Wang et al. [2011]. The 13 Driftsonde balloons remained operational for different periods of time ranging from 43 to 94 days. They achieved unprecedented spatial and temporal coverage of Antarctica, providing high quality atmospheric profiles (Figure 1). Fifty-two soundings released from one Driftsonde illustrate the detailed temperature structures observed from the surface to 60 hPa (Figure 1). The temperature profiles over land show very strong near-surface inversions, while those over the ocean and coast frequently show complex structure and inversions in the lower troposphere (Figure 1). Waves are seen in the upper troposphere and lower stratosphere in all profiles (Figure 1). In addition to temperature profiles, the relative humidity and wind speed/direction profiles are also available for further exploration.

[7] The NOAA PROducts Validation System (NPROVS) provides a centralized validation protocol for routine monitoring and comparison of derived atmospheric satellite products against in-situ observations (i.e., radiosonde and dropsonde) and Numerical Weather Prediction products [Reale et al., 2012]. In this work, NPROVS was used to identify in-situ Concordiasi soundings from either dropsondes or radiosondes that are collocated with ten satellite products from five different types of remote sensing instruments. A separation criterion of no more than 6 h temporally and 150 km spatially was used. The satellite temperature products used are from the Advanced InfraRed Sounder (AIRS) [Goldberg et al., 2003], the Infrared Atmospheric Sounding Interferometer (IASI), the Advanced TIROS Operational Vertical Sounder (ATOVS) [Reale et al., 2008], the Microwave Integrated Retrieval System (MIRS) [Boukabara et al., 2007], and GPS RO from COSMIC [Anthes et al., 2008]. Both the AIRS and IASI are hyperspectral instruments. Together, they provide 3 of the 10 products because IASI products are provided by both NOAA and European Organization for the Exploitation of Meteorological Satellites (EUMETSAT) but using different retrieval algorithms [Maddy et al., 2009; Schlussel et al., 2005]. The NOAA IASI product is available under both clear and cloudy conditions, while the EUMETSAT IASI is only for clear sky. Together, ATOVS and MIRS provide six products because they are both flown on the MetOp, NOAA-18, and NOAA-19 satellites. The tenth product, GPS RO, is available from the constellation of GPS satellites. In the following discussions, we group these ten products into two categories: radiance-derived (including all except COSMIC) and GPS-RO (i.e., COSMIC).

3 Intercomparisons of Temperature Profiles

[8] The mean bias and root-mean-square error (RMSE) of temperature differences between sondes and satellite products are shown in Figure 2. Note that Figure 2 only shows ATOVS and MIRS data from NOAA-19; the data from MetOp and NOAA-18 have similar features. We first focus on the results for the nine radiance-derived products and then the COSMIC data in the last paragraph of this section. All satellite data are consistently colder than the dropsonde data except in ~400–200 hPa for EUMETSAT_IASI and MIRS and below ~850 hPa for NOAA_IASI and ATOVS. This cold bias varies from one product to another one and changes with altitude. The RMSE generally increases with pressures and is consistent among different products except ATOVS, which has the largest RMSE. There are no significant discrepancies for different satellites using the same sounders, such as ATOVS on NOAA19, NOAA18, and MetOp. All radiance-derived satellite products are also shown to be colder than the radiosonde data collected from the stations shown in Figure 1. The cold bias with respect to dropsonde data has a larger magnitude than that relative to radiosonde data (see an example for NOAA_IASI in Figure 2).

Figure 2.

The left panel shows mean (solid line) and RMS (dashed line) differences between six satellite products and the dropsonde data. The right panel shows mean and RMS differences between NOAA_IASI or COSMIC and dropsonde (red and blue lines, denoted as “Drift” in the legend) or radiosonde data (black and green lines, denoted as “RAOB” in the legend).

[9] The reproducibility of the satellite data is presented as profiles of the correlation coefficients between the sonde and satellite data for all matched soundings and scatter plots of temperature comparisons at 500 hPa (Figure 3). The temperatures from the satellite and sonde data are highly correlated with statistically significant correlation coefficients (Figure 3). ATOVS has the smallest correlation coefficients, which is in concord with its large RMSE shown in Figure 2. The reproducibility is also clearly illustrated by individual matched profiles (see one example in Figure 3). The general structure of the AIRS profiles matches very well with the dropsonde data. However, these retrievals do not resolve the detailed structure near the surface and tropopause. The cold bias of the AIRS data is also evident in the scatter plot.

Figure 3.

The left panel shows correlation coefficient profiles of temperatures between satellite and dropsonde data for six satellite products. Scatter plot of matched temperatures at 500 hPa for NOAA_IASI versus dropsonde is in the upper right panel. The lower right panel shows temperature profiles for one matched sounding from dropsonde (black line) and AIRS (red line) data.

[10] The larger cold bias of the radiance-derived satellite data relative to the dropsonde data than relative to the radiosonde data remains unexplained. Only nine of the 16 radiosonde stations use the Vaisala RS92 radiosonde which has the same temperature thermistor as in the dropsondes, but there are no systematic differences seen in the satellite cold bias relative to different types of radiosondes (not shown). The radiosonde and dropsonde temperature datasets collected during Concordiasi do differ in spatial and temporal coverage. The Driftsonde (dropsonde) soundings cover a much larger area with greater spatial variability in temperature (Figure 1). In contrast, the radiosonde stations are all along the coast except the South Pole station, so the radiosonde profiles mainly characterize the variability along the coast. A map of the temperature differences at 500 hPa between the dropsonde data and the AIRS and IASI retrievals (Figure 4) shows that a cold bias over the continent prevails and has much larger magnitude than that along the coast and the surrounding ocean, where a warm bias is sometimes found. Histograms of the differences are also shown in Figure 4 for surface pressures smaller and larger 900 hPa, approximately representing the continent and the coastal/ocean region, respectively. The bias is clearly larger over the continent. Such contrast in the satellite bias between the continent and the ocean is also confirmed by the differences between the radiosonde stations along the coast and South Pole station (not shown). The same conclusion can be drawn for other radiance-derived satellite products.

Figure 4.

Left panels: Maps of 500 hPa temperature differences between NOAA_IASI/AIRS and dropsonde data at dropsonde locations. Right panels: histogram of temperature differences for soundings over the coast and ocean (black line) and continent (red lines). Number of samplings and mean and standard deviation of the differences are also shown in the legend. The “X” symbols on the maps are the radiosonde stations.

[11] Several factors can contribute to the temperature differences between the sonde and radiance-derived satellite data shown above, including the spatial and temporal separation between the sondes and the satellite overpasses, errors in the sonde temperature measurements, and deficiencies in the satellite retrievals. No correlation is found between the temperature bias and the spatial or temporal separations, suggesting that the first factor is not important. As discussed in section 2, the main errors in sonde temperature measurements are calibration error, solar heating, and sensor response time errors. The calibration error of ~0.15°C (warm bias) is much smaller than the differences between the dropsonde and satellite data shown in Figure 2. The sensor response error is expected to cause cold/warm biases in the dropsonde/radiosonde data in the troposphere. It should result the satellite data to have a smaller cold bias relative to the dropsonde data than to the radiosonde data, which is contrary to Figure 2. The dependence of solar radiation errors on solar zenith angle and pressure was not found in the data. Therefore, deficiencies in the radiance-derived satellite retrievals are suspected as the primary reason for the cold bias. These deficiencies are known to include the difference between surface skin temperature and surface air temperature, the complex and varied surface types over Antarctica, cloud contamination, and the ability to resolve complex temperature structures [cf., Rabier et al., 2010]. Detailed investigation of the causes for the cold bias in the satellite data is beyond the scope of this study and will require close collaboration with each satellite product developer.

[12] Comparing to the dropsonde data, the COSMIC performs better than or at least the same as the radiance-derived temperature data above 800 hPa and shows a mean cold bias of 0.48°C, which is within the 0.5°C uncertainty of the temperature sensor (Figure 2). The COSMIC above 900 hPa agrees very well with the radiosonde data with a mean difference close to zero (Figure 2). The COSMIC data also did not show the land/ocean contrast in the temperature bias relative to the dropsonde data displayed in Figure 4. This is likely due to the fact that the GPS RO technique is not affected by surface conditions and weather such as clouds. Our findings are inconsistent with the ~2°C cold bias found by Wang and Lin [2007] in the COSMIC temperature data comparing to the radiosonde data. This is speculated to be partially due to the slower response of Vaisala RS80 or RS90 temperature sensor used in Wang and Lin [2007] than Vaisala RS92 launched at nine out of 13 radiosonde stations during Concordiasi period. Note that the slow response causes a warm bias in the radiosonde data.

4 Conclusions

[13] Thirteen NCAR Driftsonde systems were deployed in the Concordiasi field experiment over Antarctica from September to December 2010. They collected 639 unprecedented pressure, temperature, humidity, and wind profiles from the stratosphere to the surface with high data quality, high vertical resolution, and large spatial coverage. The soundings cover the Antarctic continent, coast, and surrounding ocean, including areas where in-situ upper-air observations have never before been made. The Antarctic polar vortex provides ideal conditions for deploying the long duration stratospheric balloons carrying the Driftsondes. This study shows that the unique Concordiasi dataset is useful for validating satellite products over Antarctica, especially over the continent where the upper-air data are scarce. Many scientific applications of the Concordiasi dropsonde data remain to be discovered, such as studying Antarctic surface-based inversions [Zhang et al., 2011] and validating global reanalysis and model products. For example, previous studies have shown that the near-surface temperature is too warm in the weather models over the Antarctic plateau, but too cold over the surrounding sea ice due to the challenges in simulating the strong near-surface inversions in models [Rabier et al., 2010].

[14] NPROVS was used to co-locate the Concordiasi dropsonde and radiosonde data with ten satellite products. Comparisons of temperature profiles show consistent cold biases in all nine radiance-derived satellite products. The magnitude of the cold bias ranges from 0°C to 4°C varies from one product to another one and changes over altitude. The cold bias is larger relative to dropsonde measurements than radiosonde measurements for all radiance-derived products. This is attributed to the spatial coverage difference between the dropsonde and radiosonde data. All radiosonde stations but one are located along the coast, while the dropsondes cover the continent, coast, and surrounding ocean (Figure 1). All radiance-derived products exhibit larger cold biases over the Antarctic continent than over the coast and ocean. This finding would not be possible without the complete spatial coverage of the Concordiasi dropsonde data. Possible causes for the cold bias and larger bias over the continent are discussed, but investigation of deficiencies in satellite retrievals are left for future work. Collaboration with the satellite product developers is essential to understand these differences and improve the products. The COSMIC performs much better than other satellite products with very good agreement with the radiosonde data and a small cold bias comparing with the dropsonde data. It indicates that the GPS RO technique has advantages over traditional MW, IR and even hyper-spectral techniques. These are the absence of its dependence on surface properties and availability under all weather conditions. Finally, the satellite retrievals do reproduce the general structure of temperature profiles reasonably well in spite of the cold bias.

[15] The findings on the systematic errors of the satellite temperature retrievals over Antarctica have potential significant implications on past and future research on Antarctic weather and climate. Over Antarctica, the scarcity of in-situ observations especially over the interior continent makes it more important to assimilate satellite data in weather and climate models and to analyze satellite data to study the Antarctic climate changes. Previous studies also show that satellite data have much larger impact on the forecasts and reanalyses in Antarctica than in areas, such as the Arctic, where more in-situ measurements are available [cf., Rabier et al., 2010]. As a result, any systematic error in the satellite retrievals would translate in greater errors in model forecasts and reanalyses. Bracegirdle and Marshall [2012] found that four current global reanalysis products show a domination of cold biases over the period 1999–2008 when compared with the radiosonde data, with the largest bias at Amundsen-Scott station. This result is consistent with the characteristics of satellite biases found in this study. Further investigation is required to understand whether such a cold bias in the reanalyses is a result of biases in the input satellite data. In spite of significant efforts to reconstruct in-situ Antarctic temperature records and study trends [Screen and Simmonds, 2012; references therein], there are still few studies using the satellite data other than Johanson and Fu [2007]. Further improvement of satellite temperature retrievals over Antarctica is needed to reconstruct consistent long-term records for climate studies.

Acknowledgments

[16] The Concordiasi driftsonde data have been obtained through cooperation between UCAR and CNES, under the sponsorship of the NSF and the CNES. NSF Office of Polar Program supported the Concordiasi Driftsonde deployment through the grant ANT-0733007. We are grateful to all NCAR/EOL and French CNAS staff that developed Driftsonde system, and all of people who participated in Concordiasi to deploy and operate the system and collect the data. We also would like to thank Florence Rabier and other Concordiasi principle investigators for leading the project. Comments from Jordan Powers and Bill Brown have been very helpful. The National Center for Atmospheric Research is sponsored by the National Science Foundation.