SEARCH

SEARCH BY CITATION

Keywords:

  • tropopause;
  • radio occultation;
  • climate change

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Statistical Evaluation
  5. 3. LRT Height Trends
  6. 4. Correlations Between Temperatures and LRT Height
  7. 5. Summary and Outlook
  8. Acknowledgments
  9. References
  10. Supporting Information

[1] This study discusses global tropopause height variabilities and trends based on zonal monthly mean GPS radio occultation data from the German CHAMP satellite mission for the period May 2001–December 2007 (80 months). A data gap of missing CHAMP data in July 2006 was filled with radio occultation data of the US-German GRACE mission. A global increase of the tropopause height between 26–44 m during the observation period (4–7 m/yr) is found depending on the binning method (5° or 10° latitude bands) and the used tropopause detection algorithm. The corresponding trend errors vary between 19–21 m. The inclusion of the quasi-biennial oscillation in the regression model leads to a global increase of the tropopause height from 1–5 m (2–12%) during the time period depending on the binning method. Global tropopause height variations are positively correlated with upper tropospheric (500–100 hPa) and anti-correlated with lower stratospheric (100–30 hPa) temperature variations.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Statistical Evaluation
  5. 3. LRT Height Trends
  6. 4. Correlations Between Temperatures and LRT Height
  7. 5. Summary and Outlook
  8. Acknowledgments
  9. References
  10. Supporting Information

[2] The tropopause layer is one of the key regions of the atmosphere with important links to the stratosphere-troposphere exchange as well as climate research. The global mean tropopause height shows an increase in re-analyses and radiosonde observations during the last decades. Tropopause height changes are caused by different forcing mechanisms [Santer et al., 2004]. One mechanism leading to an increase of the tropopause height is a warming of the troposphere (due to more CO2) and a cooling of the lower stratosphere (due to less stratospheric ozone), both observed during the last decades. Thus, the tropopause height could be considered as a parameter for the detection of climate change processes and therefore the continuous identification and monitoring of the tropopause height is an important goal in climate research.

[3] The most important data source for the determination of tropopause parameters are radiosonde data whereas model analyses suffer from lower vertical resolution. Despite of good vertical resolution of radiosonde measurements a global coverage is impossible. Global Positioning System (GPS) radio occultation (RO) enables precise refractivity and temperature profiles with high vertical resolution (<1 km in the tropopause region). The GPS RO technique requires no active calibration, is weather independent, and the occultations are almost uniformly distributed over the globe [Melbourne et al., 1994]. Another important characteristic is the long-term stability of the system, excluding problems like discontinuities in the time series.

[4] The CHAMP (CHAllenging Minisatellite Payload) RO experiment provides data continuously since mid-2001 [Wickert et al., 2005]. For data consistency we only use results from an uniform processing software developed at GeoForschungsZentrum (GFZ) Potsdam starting at the raw data level (version 005 of atmospheric profiles from CHAMP).

[5] For the determination of the tropopause different definitions and concepts exist. In this study we use the classical definition of the World Meteorological Organization (WMO) for the first lapse rate tropopause (LRT) derived from a temperature profile [WMO, 1957].

[6] In our study we determine trends on the basis of a (climatological) relatively short time scale (80 months from May 2001 to December 2007). Nevertheless we will show that our results agree with results from radiosondes if our achievements are related to a decade.

[7] The general potential of GPS RO data for climate monitoring has been shown with simulation studies [e.g., Leroy et al., 2006]. A 5-year climatology of CHAMP RO temperature data in the upper troposphere and lower stratosphere was performed by Foelsche et al. [2007]. This study can be considered as a reference for the discussion of the different errors relevant for climatological investigations with GPS RO data.

2. Data and Statistical Evaluation

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Statistical Evaluation
  5. 3. LRT Height Trends
  6. 4. Correlations Between Temperatures and LRT Height
  7. 5. Summary and Outlook
  8. Acknowledgments
  9. References
  10. Supporting Information

[8] The CHAMP mission generates the first long-term GPS RO data set. Beside one complete month of missing data (July 2006) CHAMP delivers continuously 150–200 global and uniform distributed temperature profiles daily (more than 439,000 between 2001–2007). The data gap was bridged with RO data from the GRACE (Gravity Recovery And Climate Experiment) mission which has the same GPS receiver and error characteristics as CHAMP [Wickert et al., 2005].

[9] We calculate monthly zonal means of LRT heights if at least for 50% of the days per month CHAMP RO data are available. Additional, from each temperature profile the temperature at nine standard pressure levels (500–20 hPa) was determined to correlate the LRT height with upper tropospheric and lower stratospheric temperatures.

2.1. Tropopause Height

[10] The WMO definition was applied to each individual GPS RO temperature profile between 500–70 hPa to determine the first LRT height. In our study we use the GFZ temperature profiles version 005 with a 200 m vertical resolution and define the final LRT height twice: (1) The LRT is defined exactly at that profile altitude zi where the WMO criterion is fulfilled for the first time (called the simple method here). (2) The LRT is defined at the heights between zi and zi−1 as the intersection of two regression lines using the data points 〈zi, zi+1, zi+2〉 from above and 〈zi−1, zi−2, zi−3〉 below (called the extended method here). This two LRT heights were determined to examine the influence of the tropopause algorithm to the absolute LRT height trend values.

2.2. Binning Methods

[11] Monthly zonal means were determined for the LRT height and all temperatures at the nine pressure levels. Here also two conditions were considered to analyze the influence of the binning method on the LRT height trends. First 10° latitude bands centered at 85°N–85°S (18 latitudes) and second 5° latitude bands centered at 77.5°N–77.5°S were applied. Due to the poorer monthly data coverage at the poles two 10° latitude bands centered at 85°N and 85°S were used for the second example leading to 34 latitudes in that case. In both cases non-overlapping zonal bands were used.

2.3. Statistical Method

[12] The determination of the monthly zonal means follows the biweight mean estimation [Lanzante, 1996]. The parameter for removing outliers is selected here to exclude values more than three standard deviations from the mean. From the monthly zonal mean values the annual cycle is calculated. Finally, monthly zonal anomalies are estimated by subtracting the annual cycle from each individual monthly mean.

[13] The LRT height anomalies are the basis for linear trend calculations, whereas we use the median of pairwise slopes regression as a non-parametric technique [Lanzante, 1996, equations B37–B39]. This is called the simple regression model in the following. To study the influence of the quasi-biennial oscillation (QBO) on the LRT height trend results a multiple regression was applied with the 70 hPa monthly mean zonal wind data from Singapore radiosonde station (1°N, 104°E) as a proxy for the QBO. For the determination of the trend uncertainties standard errors were calculated under consideration of an effective sample size neff based on the 1-lag auto-correlation function (r1) of the time series with neff = n(1 − r1)/(1 + r1) and n the number of months [Karl et al., 2006, pp. 129–139] (include all used formulas). The statistical significance of the observed trends will be discussed below.

[14] To be able to compare our results easier with those obtained from longer observation periods we have related the estimated trends per 80 months as well as the according errors to trends/errors per year and decade.

3. LRT Height Trends

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Statistical Evaluation
  5. 3. LRT Height Trends
  6. 4. Correlations Between Temperatures and LRT Height
  7. 5. Summary and Outlook
  8. Acknowledgments
  9. References
  10. Supporting Information

3.1. Zonal Means

[15] Figure 1 summarizes the results for the LRT height trends for the different binning and tropopause detection algorithms.

image

Figure 1. Annual LRT height trends (simple regression) deduced with different tropopause detection algorithms (black/green, simple LRT; blue/red, extended LRT) and different binning (black/blue, 10°; green/red, 5°). Error bars denote the ±2-sigma confidence intervals. The annual trends (errors) are derived from trends (errors) based on the complete time interval from May 2001–December 2007.

Download figure to PowerPoint

[16] Generally all methods are consistent with positive LRT height trends in the extra-tropics (exceptions between 50°S–60°S and 65°N–75°N) and negative trends in the tropics (20°N–20°S). One can recognize at most latitudes that the LRT height trend is more sensitive to the binning than the used LRT method.

[17] The best agreement between all methods is observed in the tropics. This can be explained with the relatively stable tropopause height distribution between 20°N–20°S leading to a small binning effect. From ∼50°–55° to the poles on both hemispheres all methods deliver also similar trend results, whereas the transition zone between the tropical and polar tropopause shows the largest differences between the methods considered here. Especially between 25°N–45°N the choice of the latitude band width is crucial for the final trend estimate. Here, the results for the LRT height trend can differ up to 15–20 m/yr. In this transition zone the location of the subtropical jet stream strongly influences the height of the first LRT depending on whether the observation is performed on the tropical or polar side of the subtropical jet stream [Randel et al., 2007; Schmidt et al., 2006]. This leads to much more variable monthly LRT height anomalies depending on the chosen latitudinal resolution compared with other geographical regions.

[18] The general picture of Figure 1 for the extra-tropics agrees very well with findings from Seidel and Randel [2006] for radiosonde station trends based on the time interval 1980–2004 [Seidel and Randel, 2006, Figure 8]. They found also a LRT height trend maximum around 30°S and 30°N–50°N. For the tropics (20°N–20°S) discrepancies occur. It should be noted here that for the tropics a slightly positive LRT height trend (∼4 m/56 months) is observed in the GPS RO data if only data from May 2001 to December 2005 were considered. This behavior leads to the critical point of deriving trends from short time series with arbitrary end points. On the other hand there is no a priori reason that trends from 2001–2007 must agree with trends from pre-periods, but the negative tropical LRT height trend observed here should be in the focus for further investigations.

3.2. Trends in Different Geographical Regions and Globally

[19] The single monthly zonal means are the basis for calculating trends at different geographical regions and for the globe. For this reason we have defined seven latitude bands according to Seidel and Randel [2006], but with small differences in band widths due to the better latitudinal data coverage of CHAMP RO compared with radiosondes: N Polar (>60°N), N Midlat (40°N–60°N), N Subtrop (20°N–40°N), Tropics (20°N–20°S), S Subtrop (20°S–40°S), S Midlat (40°S–60°S), S Polar (<60°S). For the determination of the regional monthly means and for the globe the single monthly zonal means falling into the according latitude band are cosine-weighted.

[20] Figure 2 shows the resulting mean LRT height anomalies for the different binning and tropopause algorithms. One can clearly see that the different methods give nearly the same sign of the LRT anomalies. Figure 3 presents the corresponding trends calculated with the simple and multiple regression model. Table 1 specifies the trends including uncertainties for the simple regression model more exactly. Note that the global trends per year are not the simple averages of the trends above in the according column. With each method we find the maximum trend in the N Subtrop and S Subtrop and S Polar region (21–34 m/yr) and the minimum trend (−17 m/yr) in the tropics. Only the tropics and S Midlat give a negative LRT height trend with all methods. The estimated errors are largest in the S Subtrop and S Polar region reflecting the large variability of the first LRT there. The last column of Table 1 shows the difference between the maximum and minimum trend for the different calculation methods which can be interpreted as an uncertainty of the resulting trend due to the usage of different methods for the trend determination. Largest values occur in the N Subtrop mainly due to the binning method.

image

Figure 2. Monthly anomalies of LRT heights for different geographical regions and global. Each tick mark on the vertical axis represents 1 km (0.1 km for the global time series).

Download figure to PowerPoint

image

Figure 3. LRT height trends [m/yr] based on GPS RO data from May 2001–December 2007 (80 months) and for different binning methods (5° and 10° resolution) and different tropopause algorithms (simple and extended). The solid/dashed (squares/crosses) lines denote results based on the simple/multiple regression model. (right) Error bars (±2-sigma confidence intervals) and (left) SNR are shown for the 10°/simple LRT results. The dotted lines refer to the one-sided 80% and 90% confidence intervals.

Download figure to PowerPoint

Table 1. LRT Height Trends for Different Binning Methods and Tropopause Algorithmsa
 10° Simple10° Extended5° Simple5° ExtendedMax-Min
  • a

    Trends are in m/yr. The errors denote the ±2-sigma confidence intervals.

Simple Regression
N Polar6 ± 175 ± 160 ± 171 ± 166
N Midlat21 ± 1218 ± 1117 ± 1215 ± 126
N Subtrop32 ± 1034 ± 921 ± 1222 ± 1213
Tropics−17 ± 8−17 ± 8−16 ± 8−16 ± 81
S Subtrop25 ± 1722 ± 1625 ± 1822 ± 183
S Midlat−2 ± 16−3 ± 15−3 ± 15−4 ± 142
S Polar21 ± 3822 ± 3521 ± 3621 ± 341
Global [m/yr]7 ± 36 ± 35 ± 34 ± 33
Global [m/80 months]44 ± 2041 ± 1931 ± 2126 ± 2118
Global [m/decade]66624739 
Multiple Regression
Global [m/yr]7 ± 37 ± 35 ± 34 ± 33
Global [m/80 months]49 ± 2046 ± 1932 ± 2127 ± 2122
Global [m/decade]74694841 

[21] The statistical significance of the linear trends expressed here by the signal-to-noise ratio SNR = T/N (T: linear trend value, N: standard deviation of the anomaly time series) for the different geographical regions are included in Figure 3 (left). In cases of a clear trend value the highest SNR is observed in the northern hemisphere. From the N Midlat to the S Subtrop region and globally the SNR is larger than the one-sided 80% confidence interval (C.I.), the N Subtrop zone even reaches the 90% C.I. We assess this as a weak statistically significance of the derived linear trends. Due to the general small SNR values the already mentioned problem of determining trends from short time series arises again. But this study is a first preliminary attempt to assign LRT height trends from the longest GPS RO data set available at the moment.

[22] If we compare our results (Figure 3) with Seidel and Randel [2006, Figure 10a] there is a similar course of the trend curve and the global LRT height trend value they found (64 m/decade) is in excellent agreement with our study (62–66 m/decade) if we extrapolate our results to a decade (10° bins).

[23] Globally, the simple tropopause algorithm trend is 1 m/yr (3–5 m/80 months) greater than the trend values deduced from the extended tropopause detection method independent from the binning method. On the other hand, the binning method shows LRT height trend differences of 2 m/yr or 13–15 m/80 months (higher values with lower bin resolution). Largest differences occur in the N Subtrop band.

[24] Generally, the multiple regression model is in good agreement with the simple model (Figure 3). From this we conclude that the LRT height trends are not importantly affected by the QBO, which is more or less visible in the monthly LRT height anomalies (Figure 2). Globally, the inclusion of the QBO leads to LRT height trend differences between 1–5 m/80 months (2–12%) in comparison to the simple regression model, whereas the largest values are found for the 10° binning (Table 1).

4. Correlations Between Temperatures and LRT Height

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Statistical Evaluation
  5. 3. LRT Height Trends
  6. 4. Correlations Between Temperatures and LRT Height
  7. 5. Summary and Outlook
  8. Acknowledgments
  9. References
  10. Supporting Information

[25] The LRT height and its variability is related to tropospheric and lower stratospheric temperatures. To study this correlations monthly mean temperature anomalies at different standard pressure levels were identified with the same methods as for the LRT height. Finally from the LRT height and temperature anomalies at different pressure levels correlations were designated. Figure 4 shows the corresponding results for the different latitude bands and globally for 10° latitude resolution and the extended tropopause method. The other methods are suppressed here because the results are very similar.

image

Figure 4. Correlation coefficients between monthly anomalies of LRT height and temperatures at different standard pressure levels for different latitude bands and global. The results are based on 10° binning and the extended tropopause algorithm. The correlations vary between −1 and +1 on the vertical axis.

Download figure to PowerPoint

[26] One can recognize that in all regions and globally the LRT height anomalies are positively correlated with tropospheric temperature anomalies and negative correlated with most of the stratospheric temperatures (exceptions are 20 and 30 hPa temperatures). One should keep in mind that the climatological tropopause is located at about 100 hPa in the tropics and at about 300 hPa in the polar regions.

[27] We find the strongest tropospheric correlations in both the N and S Subtrop and Midlat. In the tropics the correlations are smallest. The strongest stratospheric anti-correlations are observed in the polar regions.

5. Summary and Outlook

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Statistical Evaluation
  5. 3. LRT Height Trends
  6. 4. Correlations Between Temperatures and LRT Height
  7. 5. Summary and Outlook
  8. Acknowledgments
  9. References
  10. Supporting Information

[28] In this study we have discussed LRT height trends from GPS RO data determined with different methods. We have shown that the binning method is crucial, especially in the subtropical regions. We have shown that the binning method is more important for the absolute values of LRT height trends than the tropopause detection algorithm. The LRT height trends are not importantly (2–12%) influenced by the QBO.

[29] Global LRT height trends between 39–66 m/decade are in good agreement with radiosonde observations. For all geographical regions we found positive tropospheric and negative stratospheric correlations between LRT height anomalies and temperature anomalies at different pressure levels, whereas the extra-tropical correlations are higher compared with the tropics.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Statistical Evaluation
  5. 3. LRT Height Trends
  6. 4. Correlations Between Temperatures and LRT Height
  7. 5. Summary and Outlook
  8. Acknowledgments
  9. References
  10. Supporting Information

[30] The authors would like to thank the CHAMP and GRACE team for providing the RO data and orbits. We would also like to thank Bill Randel and the two anonymous referees for helpful comments and suggestions. The radiosonde data were provided by the Free University Berlin.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Statistical Evaluation
  5. 3. LRT Height Trends
  6. 4. Correlations Between Temperatures and LRT Height
  7. 5. Summary and Outlook
  8. Acknowledgments
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Statistical Evaluation
  5. 3. LRT Height Trends
  6. 4. Correlations Between Temperatures and LRT Height
  7. 5. Summary and Outlook
  8. Acknowledgments
  9. References
  10. Supporting Information
FilenameFormatSizeDescription
grl24628-sup-0001-t01.txtplain text document1KTab-delimited Table 1.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.