Contributors: UNIVIE : Leopold Haimberger, Federico Ambrogi, Michael Blaschek, Ulrich Voggenberger, Susanna Winkelbauer
Issued by: UNIVIE / Leopold Haimberge
Issued Date: 07/02/2021
Ref: C3S_ DC3S311c_Lot2.2.2.2_ATBD_Wind_Adjustments
Official refence number service contract: 2019/C3S_311c_Lot2_UNIVIE/SC1
1. Introduction
This document describes the rationale and implementation of the algorithms used for adjustments of wind data from radiosondes and PILOT balloons available in CUON.
Biases in wind data, particularly directional biases, may lead to spurious flow patterns in reanalyses which assimilate such data. Such biases were highlighted by Hollingsworth et al. (1986), who were among the first to use background departure statistics for diagnosing problems in the observing system. The results of this paper prompted the maintainers of the affected stations to fix problems with north alignment of their ground base. However, the biased measurements still exist in historical upper air archives. Gruber and Haimberger (2008) were the first to systematically check and correct wind direction biases in long radiosonde and PILOT wind time series using ERA-40 data. The work was extended by Ramella-Pralungo et al. (2014a,b) to very early data, going back to 1920, and vertical wind shear from adjusted wind records was used by Ramella-Pralungo Haimberger (2015) to get independent estimates of tropical temperature trend amplification.
Ramella-Pralungo et al. (2015) found biases mostly in wind direction and found very little evidence of wind speed biases. Ingleby et al. (2018) confirmed that there is no hint of wind speed biases, at least when using GPS radiosondes. There may be some chance of underestimating wind speed peaks in cores of jet streams, due to undersampling, but since in general the maximum of the wind speed vertical profile triggers designation of a significant level, even that is unlikely.
The inertia of balloons is very small (2kg). If we calculate the dynamic pressure of a relative wind of 5m/sec impinging on the balloon in the mid-troposphere, it causes a force of roughly 40N on a 100m2 balloon cross section. Thus, it accelerates or decelerates faster than the Earth's acceleration to achieve equilibrium speed. Also, the RADAR location method (since ~1955) and theodolite location method are not a source of speed biases. As has been extensively discussed by Ramella-Pralungo et al. (2014b), apparent monthly mean wind speed biases can almost entirely be traced back to sampling errors when balloons were lost already at low altitudes under high wind conditions in the days before RADAR tracking was available. So, while the individual measurements were accurate, high wind speed cases were not sampled. All these arguments justify that we have restricted the bias adjustments for this service to wind direction as well.
As for temperature and specific humidity, the observation location and timing information has been improved by taking into account the balloon drift (Voggenberger et al. 2024). This reduction of representation errors helped to significantly reduce the RMS departures between gridded ERA5 reanalysis data and observations. In very recent times, balloons report data at high (1 second) frequency, which leads to new opportunities for wind measurements, i.e. sampling high frequency signals, but also increases the noise due to pendulum motions of the payload. It has been tried to keep the high resolution information if possible. Lower resolution profiles have been used only if the high resolution profiles were incomplete for some reason. All those aspects are elaborated more in the following sections.
2. Adding pressure to height level data
Figure 1: Number of radiosonde and tracked balloon platforms per year. The platform information was taken from the ERA5 reportype value or, if that was not available (e.g. before 1940), from the (non)availability of observed temperatures in the records.
Upper air balloon data are mostly available as TEMP and PILOT reports; for simplicity we considered all reports with wind observations but no temperature or humidity data as PILOT reports. One can see from Fig. 1 that PILOT ascents contribute significantly to the overall amount of wind data, particularly before the 1950s. Since the 1980s the fraction of PILOT ascents has decreased and is below 20% in the 2020s. Unless the balloon is equipped with a GNSS receiver, its height can only be estimated from the time since launch and an assumed ascent rate of 5 m/s. For PILOT ascents, pressure is typically estimated from height using a standard atmosphere profile, which is also the method used in ERA5. More accurate pressures could be derived using geopotential forecasts from reanalyses. Future datasets may adopt ERA5 background forecasts for improved accuracy of the derived pressures. For TEMP, where temperature and pressure is measured, the situation is different. Geopotential height can be calculated accurately using piecewise linear temperature profiles.
3. Taking into account actual launch time and balloon drift
Once the height data are available, it is possible to reconstruct the balloon drift by integrating the balloon trajectory using the observed wind and time (assuming constant ascent rates). Balloon drift was calculated for all PILOT and radiosonde ascents, if at least two height levels were available. The balloon drift calculation accuracy depends on the quality and frequency of the available wind data, but even with coarse information it leads to significantly better observation location estimates compared to assuming straight vertical ascents (Voggenberger et al. 2024).
The reconstructed time since launch also leads to more accurate estimates of observation time, particularly if the launch time is known (as is the case for many stations originating from the IGRA archive). If launch time is known only for a certain period, the average difference between launch time and nominal time is used to get a best estimate for the launch time for those ascents where it is not available. This estimate, or even the assumption of a launch time half an hour prior to the nominal time and then calculating the actual time from the height of the balloon appears to reduce the representation error. The impact of reduced representation error can be seen in Figure 2, which shows Hovmöller diagrams of the ratio between archived obs-bg departures from ERA5 and offline calculated departures, taking balloon drift and actual launch time into account. A clearly positive impact is visible, because of using accurate launch times over the US (as available from IGRA) and accounting for balloon drift. Removal of a pervasive wind direction bias over the US in the 1940s (Ramella-Pralungo and Haimberger, 2014) also helped reduce the background departures. Figure 3 shows this ratio for the period 1960-2024. A weak positive effect is noticeable there as well, although the offline calculated departures have a larger interpolation error, indicating the balloon drift error is larger than the interpolation error also in this period. An exception are the most recent years where balloon drift estimates no longer have an advantage over GPS position data from high-resolution BUFR, which were available at this time.
Figure 2: Hovmöller diagram of ratios between offline-calculated wind component background departures (rms over all stations within a 3 month period at standard pressure levels). Left panels: archived online-calculated background departures from ERA-5. Middle panels: offline-calculated wind component background departures taking balloon drift into account. Right panels like center panels after adjustment of wind direction
Figure 3: Same as Figure 2, but for time period 1960-2024.
4. Calculation of the adjustments
Besides the drift information, CUON also provides bias estimates for wind direction. The wind direction bias adjustment method applied for CUON data relies on the availability of observation minus background departure (obs-bg) time series. These are available for radiosonde and PILOT records that have been assimilated by ERA5. If the records were not assimilated, the obs-bg differences can be calculated offline on standard pressure levels from the available gridded ERA5 forecast data. For the pre-1940 period the NOAA 20CR v3 (Slivinski et al. 2019) is used as background. The interpolation error is larger than during online assimilation because of larger time and space deviations (0.25 deg grid, 3-hourly analyses/forecasts for ERA5, 1 deg grid, 6 hours for NOAA 20CRv3), but particularly in the early days, this error appears to be not larger than the forecast errors. In order to avoid any possible inhomogeneities in the obs-bg departures when switching from archived to offline calculated obs-bg departures, it has been decided to always use offline calculated obs-bg differences for bias estimation. In addition, the offline calculated departures are available also for records after 1940 that were not assimilated into ERA5. The offline calculated departures are available in the era5/fg_depar@offline variable, which is not part of CDM-OBS-core but can be requested using the CDSAPI.
The calculation of the balloon drift as well as offline calculation of background departures with respect to ERA5 is performed in the CUON software repository script resort/convert_faster_with_recarray_plus_fb.py from the CEUAS software repository https://github.com/MBlaschek/CEUAS/tree/master/CEUAS/public. This important step is performed not only for wind but also for temperature and humidity variables.
After this step, one can calculate the background wind direction from the observations and departures and compare it with the observed wind direction. The resulting wind direction difference time series can then be analyzed for breakpoints and the changes of the bias estimates at the changepoints can be calculated as well. Breakpoints are detected using the Standard Normal Homogeneity Test (Alexandersson, 1986), adapted to work with moving averages (Haimberger, 2007). Only step-like breakpoints are detected; it is not attempted to adjust for creeping changes of the bias. The time window used is 4 years (2 years before, 2 years after a suspected breakpoint), a minimum of 80 data points is required on each side of the breakpoint. The wind direction difference between obs and background, being circular, is required to be between -180 and 180 degrees. Only data points with minimum wind speed of 2 m/sec are used, since for lower wind speeds the wind direction was found to be unreliable.
A breakpoint must be found at least at three standard pressure levels and the break size in the obs-bg angle difference time series must be at least three degrees. The break size is calculated as average of the break size estimates at the 200-700 hPa standard pressure levels. Since the direction bias, practically always resulting from wrong north alignment, is constant in the vertical, the breaksize is constant in the vertical as well. In this respect the adjustment procedure follows closely the one used by Gruber and Haimberger (2008). Direction differences of a few degrees are quite common, given that wind direction is usually reported at an accuracy of only 10 degrees.
A spectacular example of changing wind direction bias is station Aktobe (35229), whose northward wind component background departure time series (obs-bg) from ERA5 is shown in Figure 4. One can clearly see (blue line) that there is a period of enhanced background departures from the late 1970s to early 1990s. The reason is wrong north alignment during this time. If one calculates wind direction departures from the observed winds and background departures, one gets a wind direction departure time series (shown in figure 5). This time series is analyzed and a sequence of constant adjustments is calculated. The beneficial impact of the adjustments on the northward wind departures can be seen in Figure 4. The standard deviation of the departures is a lot smaller after adjustment.
As documentation at the source code level, the python script used for calculating the adjustments can be found at adjust/Wind_adjustment/adjust_wind_from_cds.py.
Figure 4: Observation – ERA5 background northward wind departures, before(blue) and after adjustment of wind direction at Aktobe (35229).
Figure 5: Time series of wind direction difference between observed wind and ERA5 background wind at 200 hPa , as well as calculated bias adjustment for radiosonde station Aktobe (35229).
5. Global plots of trends
The bias adjustments for wind direction mostly have a moderate impact on long term regional trends, since they are relatively rare. The spatial heterogeneity of trends, measured as a ''trend cost'' function (see Haimberger, 2007) decreases when applying the homogeneity adjustments. Assuming predominantly westerly winds, the northward component of wind is most strongly affected by shifts in wind direction. One can see reduced trend heterogeneity of the v-component of the wind after the adjustments in periods 1940-1960 and 1990-2010 (Figure 6). The most remarkable signal can be seen in the US over the very early period (1940-1960), this has already been shown by Ramella-Pralungo et al. (2014) using NOAA 20CR data and is reproducible also from ERA5. The better trend consistency is a sign that observation uncertainty is reduced by this adjustment procedure. At the later period (lower panels) the trend heterogeneity is only slightly reduced.
|
|
|
|
Figure 6: v-component of wind trends at radiosonde/PILOT stations in units m/s/10a in intervals 1940-1960 and 1990-2010 at 700hPa, from unadjusted observations (left panels) and observations with component bias estimate subtracted (right panels). Periods considered are 1940-1960 (top panels), and 1990-2010 (lower panels)
6. Scripts specific for wind adjustments
The harvesting and merging of CUON data is described in the Product User Guide . In those steps no specific processing is applied to wind data except that ascents that have only height as coordinate are converted into pressure as coordinate. The calculation of pressure at height levels, if not yet available from e.g. the ERA5 odb data is performed in the script resort/convert_faster_with_recarray_plus_fb.py. Also the balloon drift as well as the offline background departures with respect to ERA5 are estimated in the script resort/convert_faster_with_recarray_plus_fb.py. After this step, one can calculate the background wind direction from the observations and departures and compare it with the observed wind direction. The resulting wind direction difference series are analyzed to find changepoints and to calculate adjustments. This happens in the script adjust_wind_from_cds.py. This script is best applied to the station time series data of CUON as they are available before uploading them to the CDS repository. One could also apply it to data retrieved from CDS, though with significant speed penalty.
7. References
Alexandersson, H. (1986). A homogeneity test applied to precipitation data. Journal of Climatology, 6(6), 661–675. https://doi.org/10.1002/joc.3370060607
Gruber, C. and Haimberger, L. 2008: On the homogeneity of radiosonde wind time series. Meteorol. Z. 17, 631-643. https://doi.org/10.1127/0941-2948/2008/0298
Haimberger, L., 2007: Homogenization of radiosonde temperature time series using innovation statistics. J. Climate 20, 1377-1403. https://doi.org/10.1175/JCLI4050.1
Hollingsworth, A., D.B. Shaw, P. Lonnberg, L. Illari, A.J. Simmons, 1986: Monitoring of observation and analysis quality by a data assimilation system. – Mon. Wea. Rev. 114, 861–879. https://doi.org/10.1175/1520-0493(1986)114<0861:MOOAAQ>2.0.CO;2
Ingleby, B., Isaksen, L., Kral, T., Haiden, T. and Dahoui, M., 2018: Improved use of atmospheric in situ data. ECMWF Newsletter 155. https://www.ecmwf.int/en/newsletter/155/meteorology/improved-use-atmospheric-situ-data
Ramella Pralungo, L , Haimberger, L, Stickler, A. and S. Brönnimann, 2014a, 'A global radiosonde and tracked balloon archive on 16 pressure levels (GRASP) back to 1905 – Part 1: Merging and interpolation to 00:00 and 12:00 GMT', Earth System Science Data. https://doi.org/10.5194/essd-6-185-2014
Ramella Pralungo, L. and Haimberger, L 2014b.: A "Global Radiosonde and tracked-balloon Archive on Sixteen Pressure levels" (GRASP) going back to 1905 – Part 2: homogeneity adjustments for pilot balloon and radiosonde wind data, Earth Syst. Sci. Data, 6, 297–316, https://doi.org/10.5194/essd-6-297-2014
Ramella Pralungo, L & Haimberger, L 2015, 'New estimates of tropical mean temperature trend profiles from zonal mean historical radiosonde and pilot balloon wind shear observations', Journal of Geophysical Research: Atmospheres, Jg. 120, Nr. 9, S. 3700-3713. https://doi.org/10.1002/2014JD022664
Slivinski, L. C., Compo, G. P., Whitaker, J. S., Sardeshmukh, P. D., Giese, B. S., McColl, C., Allan, R., Yin, X., Vose, R., Titchner, H., Kennedy, J., Spencer, L. J., Ashcroft, L., Brönnimann, S., Brunet, M., Camuffo, D., Cornes, R., Cram, T. A., Crouthamel, R., Domínguez‐Castro, F., Freeman, J. E., Gergis, J., Hawkins, E., Jones, P. D., Jourdain, S., Kaplan, A., Kubota, H., Le Blancq, F., Lee, T., Lorrey, A., Luterbacher, J., Maugeri, M., Mock, C. J., Moore, G. K., Przybylak, R., Pudmenzky, C., Reason, C., Slonosky, V. C., Smith, C., Tinz, B., Trewin, B., Valente, M. A., Wang, X. L., Wilkinson, C., Wood, K. and Wyszyński, P. (2019), Towards a more reliable historical reanalysis: Improvements for version 3 of the Twentieth Century Reanalysis system. Q J R Meteorol Soc. 145, 2876-2908. https://doi.org/10.1002/qj.3598
Voggenberger, U., Haimberger, L., Ambrogi, F., Poli, P., 2024: Balloon drift estimation and improved position estimates for radiosondes, GMD, https://doi.org/10.5194/gmd-17-3783-2024
