• Barker, D. M., , Huang W. , , Guo Y.-R. , , Bourgeois A. J. , , and Xiao Q. N. , 2004: A three-dimensional variational data assimilation system for MM5: Implementation and initial results. Mon. Wea. Rev., 132, 897914, doi:10.1175/1520-0493(2004)132<0897:ATVDAS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Barker, D. M., and et al. , 2012: The Weather Research and Forecasting (WRF) Model’s Community Variational/Ensemble Data Assimilation System: WRFDA. Bull. Amer. Meteor. Soc., 93, 831843, doi:10.1175/BAMS-D-11-00167.1.

    • Search Google Scholar
    • Export Citation
  • Chen, F., , and Dudhia J. , 2001: Coupling an advanced land surface–hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129, 569585, doi:10.1175/1520-0493(2001)129<0569:CAALSH>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Courtier, P., , Thépaut J.-N. , , and Hollingsworth A. , 1994: A strategy for operational implementation of 4D-Var, using an incremental approach. Quart. J. Roy. Meteor. Soc., 120, 13671387, doi:10.1002/qj.49712051912.

    • Search Google Scholar
    • Export Citation
  • Davies, T., , Cullen M. J. P. , , Malcolm A. J. , , Mawson M. H. , , Staniforth A. , , White A. A. , , and Wood N. , 2005: A new dynamical core for the Met Office’s global and regional modelling of the atmosphere. Quart. J. Roy. Meteor. Soc., 131, 17591782, doi:10.1256/qj.04.101.

    • Search Google Scholar
    • Export Citation
  • Dee, D. P., and et al. , 2011: The ERA-Interim reanalysis: Configuration and performance of the data assimilation system. Quart. J. Roy. Meteor. Soc., 137, 553597, doi:10.1002/qj.828.

    • Search Google Scholar
    • Export Citation
  • Dudhia, J., 1989: Numerical study of convection observed during the Winter Monsoon Experiment using a mesoscale two-dimensional model. J. Atmos. Sci., 46, 30773107, doi:10.1175/1520-0469(1989)046<3077:NSOCOD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Hong, S.-Y., , and Lim J.-O. , 2006: The WRF single-moment 6-class microphysics scheme (WSM6). J. Korean Meteor. Soc., 42, 129151.

  • Hong, S.-Y., , Noh Y. , , and Dudhia J. , 2006: A new vertical diffusion package with an explicit treatment of entrainment processes. Mon. Wea. Rev., 134, 23182341, doi:10.1175/MWR3199.1.

    • Search Google Scholar
    • Export Citation
  • Huang, X.-Y., and et al. , 2009: Four-dimensional variational data assimilation for WRF: Formulation and preliminary results. Mon. Wea. Rev., 137, 299314, doi:10.1175/2008MWR2577.1.

    • Search Google Scholar
    • Export Citation
  • Kain, J. S., 2004: The Kain–Fritsch convective parameterization: An update. J. Appl. Meteor., 43, 170181, doi:10.1175/1520-0450(2004)043<0170:TKCPAU>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Keyser, D., 2012: RAOB/PIBAL balloon drift latitude, longitude, and time calculation in PREPBUFR. NCEP/NMC, accessed 22 April 2014. [Available online at www.emc.ncep.noaa.gov/mmb/data_processing/prepbufr.doc/balloon_drift_for_TPB.htm.]

  • Laroche, S., , and Sarrazin R. , 2013: Impact of radiosonde balloon drift on numerical weather prediction and verification. Wea. Forecasting, 28, 772782, doi:10.1175/WAF-D-12-00114.1.

    • Search Google Scholar
    • Export Citation
  • MacPherson, B., 1995: Radiosonde balloon drift—Does it matter for data assimilation? Meteor. Appl., 2, 301305, doi:10.1002/met.5060020402.

    • Search Google Scholar
    • Export Citation
  • Martin, J. E., 2006: Mid-Latitude Atmospheric Dynamics: A First Course. J. Wiley and Sons, 324 pp.

  • McGrath, R., , Semmler T. , , Sweeney C. , , and Wang S. , 2006: Impact of balloon drift errors in radiosonde data on climate statistics. J. Climate, 19, 34303442, doi:10.1175/JCLI3804.1.

    • Search Google Scholar
    • Export Citation
  • Mlawer, E. J., , Taubman S. J. , , Brown P. D. , , Iacono M. J. , , and Clough S. A. , 1997: Radiative transfer for inhomogeneous atmospheres: RRTM, a validated correlated-k model for the longwave. J. Geophys. Res., 102, 16 66316 682, doi:10.1029/97JD00237.

    • Search Google Scholar
    • Export Citation
  • Parrish, D. F., , and Derber J. , 1992: The National Meteorological Center’s spectral statistical-interpolation analysis system. Mon. Wea. Rev., 120, 17471763, doi:10.1175/1520-0493(1992)120<1747:TNMCSS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Seidel, D. J., , Sun B. , , Pettey M. , , and Reale A. , 2011: Global radiosonde balloon drift statistics. J. Geophys. Res., 116, D07102, doi:10.1029/2010JD014891.

    • Search Google Scholar
    • Export Citation
  • Skamarock, W. C., and et al. , 2008: A description of the Advanced Research WRF version 3. NCAR Tech. Note TN–475+STR, 113 pp. [Available online at http://www2.mmm.ucar.edu/wrf/users/docs/arw_v3.pdf.]

  • Zhang, X., , Huang X.-Y. , , and Pan N. , 2013: Development of the upgraded tangent linear and adjoint of the Weather Research and Forecasting (WRF) Model. J. Atmos. Oceanic Technol., 30, 11801188, doi:10.1175/JTECH-D-12-00213.1.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    (a) Geographical areas of domains 1–3 with topographic height (m; shading) and (b) trajectories of balloons launched from the Boseong (red), Changwon (blue), and Seongsan (green) stations over domain 3. Max drift distances (km) are indicated. Locations of the KMA regular upper-air stations are denoted by black dots.

  • View in gallery

    Vertical profile of (a) drift distance (km), (b) elapsed time (s), and (c) ascent speed (m s−1) for the Boseong (red), Changwon (blue), and Seongsan (green) stations, and their mean value (black). Drift distances, elapsed times, and ascent speeds during the observation period from 20 Jun to 5 Jul 2013 are averaged.

  • View in gallery

    Time series of daily rainfall amount (mm day−1) from 20 Jun to 4 Jul 2013. Daily rainfall amount from the KMA AWSs between 33° and 36°N are averaged.

  • View in gallery

    Synoptic environments at 0000 UTC 4 Jul 2013. (a) Geopotential height [m; black solid contours with a contour interval (CI) of 30 m], temperature (°C; red dashed contours with a CI of 3°C), and water vapor mixing ratio (kg kg−1; shaded for values greater than 0.012 kg kg−1) at 850 hPa. (b) Conditional instability between 850 and 500 hPa (K hPa−1; red solid contours with a CI of 0.005 K hPa−1) and equivalent potential temperature (K; shaded) at 850 hPa.

  • View in gallery

    (a) Observed 12-h accumulated rainfall distribution [mm (12 h)−1] from 0000 to 1200 UTC 4 Jul 2013. Locations of the KMA AWSs are also shown. (b) Hourly rainfall amount (mm h−1) at Hampyeong (or the corresponding grid points in the case of model experiments) for the observations (black), and the NoDA (blue), All_BD (green), and No_BD (red) experiments.

  • View in gallery

    RMSEs of each cycle’s analysis (forecasts inside the assimilation window) verified against radiosonde observations (assimilated with or without considering balloon drift information) from the Boseong, Changwon, and Seongsan stations for the NoDA (blue), No_BD (red), and All_BD (green) experiments. The positions and elapsed times of the balloons are considered in the verification, and the average over all cycles for each experiment is also shown next to its experiment name: (a) zonal wind (m s−1), (b) meridional wind (m s−1), (c) temperature (K), and (d) dewpoint temperature (K).

  • View in gallery

    As in Fig. 6, but for each cycle’s forecast (forecasts outside the assimilation window) and radiosonde observations, which are not assimilated.

  • View in gallery

    Threat and bias scores of 3- and 12-h accumulated rainfall for the NoDA (blue), No_BD (red), and All_BD (green) experiments. Threat and bias scores are calculated during the cycling period and averaged. (a) Threat score with a threshold value of 0.1 mm, (b) threat score with a threshold value of 10.0 mm, (c) bias score with a threshold value of 0.1 mm, and (d) bias score with a threshold value of 10.0 mm.

  • View in gallery

    Simulated 12-h accumulated rainfall distribution [mm (12 h)−1] from 0000 to 1200 UTC 4 Jul 2013 for the (a) NoDA, (b) No_BD, (c) All_BD, (d) BD_Position, and (e) BD_Time experiments.

  • View in gallery

    RMSEs of 6- or 12-h accumulated rainfall [mm (6 h)−1 or mm (12 h)−1] for the NoDA (blue), No_BD (red), and All_BD (green) experiments.

  • View in gallery

    Analysis increments of CAPE (J kg−1; red contours with negative values denoted by dashed contours) for the All_BD experiment at 0000 UTC, and the difference in CAPE between the NoDA and All_BD experiments (J kg−1; shaded) at 0400 UTC 4 Jul 2013.

  • View in gallery

    (a) MPV (PVU; shaded) and water vapor mixing ratio (kg kg−1; green solid contours with a CI of 0.001 kg kg−1) at the 800-hPa level for the All_BD experiment at 0700 UTC 4 Jul 2013. (b) Conditional instability between 850 and 400 hPa (K hPa−1; shaded) and vertical motion (m s−1; red solid contours with only positive values plotted) at the 850-hPa level for the No_BD experiment at 0100 UTC 4 Jul 2013.

  • View in gallery

    Vertical distribution of RMSEs for the NoDA (blue), No_BD (red), and All_BD (green) experiments. Radiosonde observations from the Boseong, Changwon, and Seongsan stations are used for verification, and the positions and elapsed times of the balloon are considered. Forecasts between 0500 (1100) UTC and 0700 (1300) UTC are verified against radiosonde observations targeted for 0600 (1200) UTC, and the RMSEs from the two periods are averaged: (a) zonal wind (m s−1), (b) meridional wind (m s−1), (c) temperature (K), and (d) dewpoint temperature (K).

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 52 52 10
PDF Downloads 48 48 7

Investigation of the Effects of Considering Balloon Drift Information on Radiosonde Data Assimilation Using the Four-Dimensional Variational Method

View More View Less
  • 1 School of Earth and Environmental Sciences, Seoul National University, Seoul, South Korea
  • | 2 Forecast Research Division, National Institute of Meteorological Research, Jeju, South Korea
  • | 3 School of Earth and Environmental Sciences, Seoul National University, Seoul, South Korea
© Get Permissions
Full access

Abstract

Effects of balloon drift information (i.e., position and elapsed ascent time of the balloon) on the assimilation of radiosonde observations are investigated by using the Weather Research and Forecasting (WRF) Model and its data assimilation (WRFDA) system. Special radiosonde observations over the Korean Peninsula, which include the exact position and elapsed time of the balloon, are used instead of estimating the balloon drift information. To consider the balloon drift information appropriately, the four-dimensional variational data assimilation (4DVAR) and a high horizontal resolution (6 km) are used. Cycling experiments over the observation period from 20 June to 4 July 2013 are carried out to obtain the statistical robustness of the effects of considering the balloon drift information, and a single-case experiment is also conducted to show further details about the effects. The verification results of cycling experiments, such as root-mean-square errors (RMSEs) for meteorological-variable forecasts verified against the radiosonde observations and threat and bias scores for rainfall forecasts, show the positive impacts of considering the balloon drift information. Results of the single-case experiment also reveal that the simulated rainfall distribution, time series of hourly rainfall, and quantitative precipitation forecast (QPF) skills are improved through the assimilation of radiosonde observations while considering the balloon drift information. Additionally, forecasts of meteorological variables such as horizontal wind components, temperature, and dewpoint temperature are also improved by considering the balloon drift information in the single-case experiment.

Corresponding author address: Gyu-Ho Lim, Atmospheric Science Program, School of Earth and Environmental Sciences, Seoul National University, Seoul 151-742, South Korea. E-mail: gyuholim@snu.ac.kr

Abstract

Effects of balloon drift information (i.e., position and elapsed ascent time of the balloon) on the assimilation of radiosonde observations are investigated by using the Weather Research and Forecasting (WRF) Model and its data assimilation (WRFDA) system. Special radiosonde observations over the Korean Peninsula, which include the exact position and elapsed time of the balloon, are used instead of estimating the balloon drift information. To consider the balloon drift information appropriately, the four-dimensional variational data assimilation (4DVAR) and a high horizontal resolution (6 km) are used. Cycling experiments over the observation period from 20 June to 4 July 2013 are carried out to obtain the statistical robustness of the effects of considering the balloon drift information, and a single-case experiment is also conducted to show further details about the effects. The verification results of cycling experiments, such as root-mean-square errors (RMSEs) for meteorological-variable forecasts verified against the radiosonde observations and threat and bias scores for rainfall forecasts, show the positive impacts of considering the balloon drift information. Results of the single-case experiment also reveal that the simulated rainfall distribution, time series of hourly rainfall, and quantitative precipitation forecast (QPF) skills are improved through the assimilation of radiosonde observations while considering the balloon drift information. Additionally, forecasts of meteorological variables such as horizontal wind components, temperature, and dewpoint temperature are also improved by considering the balloon drift information in the single-case experiment.

Corresponding author address: Gyu-Ho Lim, Atmospheric Science Program, School of Earth and Environmental Sciences, Seoul National University, Seoul 151-742, South Korea. E-mail: gyuholim@snu.ac.kr

1. Introduction

The radiosonde is one of the most important observation platforms for numerical weather prediction (NWP). Through the assimilation of radiosonde observations into NWP models, the predictability of NWP models can be improved. Globally, there are about 1300 upper-air stations where radiosondes make measurements of pressure, horizontal wind components, temperature, and humidity from just above ground to heights of up to 30 km. At over two-thirds of the stations, radiosonde observations are made at 0000 and 1200 UTC every day. Generally, it is assumed in NWP applications that radiosonde observations represent the vertical profile of the atmosphere at the upper-air station and at the synoptic hour. However, strictly speaking, because of balloon drift, this assumption is not valid. Recently, information on the exact position and elapsed ascent time of the balloon has become available because of a progressive transition from the alphanumeric code to the Binary Universal Form for the Representation of Meteorological Data (BUFR) code on the Global Telecommunications System (GTS). Therefore, we can eliminate a source of uncertainties and improve the analyses and forecasts by considering balloon drift information when we assimilate radiosonde observations into NWP models.

Seidel et al. (2011) calculated a comprehensive global climatology of radiosonde balloon drift distance and ascent time based on 2 years of data from 419 stations in the Global Climate Observing System (GCOS) Reference Upper-Air Network (GRUAN). Typical drift distances are a few kilometers in the lower troposphere, ~5 km in the midtroposphere, ~20 km in the upper troposphere, and ~50 km in the lower stratosphere. Drift distances tend to increase with height, be larger in midlatitudes than in the tropics, and to be greater in winter than in summer. Estimated elapsed times from balloon launch to various pressure levels have median values ranging from about 5 min at 850 hPa to approximately 1.7 h at 10 hPa, with ranges of about 20% of the median values.

Radiosonde data are a valuable resource in the detection of climate change in the upper troposphere. McGrath et al. (2006) examined the impact of balloon drift error in radiosonde data on climate statistics using simulated radiosonde data generated from the European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis archive. The results suggested that the temperature errors, while generally small in the troposphere, are locally significant in the stratosphere, particularly in the jet stream areas. However, the impact of drift error on global climate statistics is very small. Although the mean errors related to radiosonde balloon drift are small from a climatological perspective, individual errors may be large and of significance for NWP applications.

MacPherson (1995) conducted a sensitivity experiment with the Met Office (UKMO) mesoscale model to investigate the impact of balloon drift on a short-period forecast of a small-scale baroclinic development. The balloon trajectory was calculated from the observed horizontal winds with an assumed rise speed of 5 m s−1, and a launch time of 30 min before the synoptic hour was assigned to the trajectory data. He found that the impact of balloon drift can indeed be detected. However, the difference between the experiments with and without considering balloon drift is not significant. This may be attributed to the coarse horizontal resolution (~17 km) of the model and the rather simple data assimilation method (analysis correction scheme). Therefore, with the increasing resolution of NWP models, it is now worthwhile to take into account the accurate positions and times of radiosonde observations.

Laroche and Sarrazin (2013) examined the impact of balloon drift on forecasts with the Environment Canada (EC) global forecast system for January 2009. They developed a method for estimating the balloon drift position from reported horizontal wind components and a representative elapsed ascent time profile. They showed that the retrieved position error using the method is roughly 10% of the drift distance from the upper-air station, which represents a significant error reduction with respect to taking the station location as the radiosonde position on all reporting levels. They carried out two data assimilation experiments over a 1-month period. The horizontal positions and times from launch of radiosonde data were estimated in one experiment, and the horizontal positions and times of radiosonde data on all levels were set to the launching locations and times in the other experiment. The observations were assimilated using the EC global four-dimensional variational data assimilation (4DVAR) system over a 6-h assimilation window in both experiments. The impact of neglecting the balloon position in data assimilation and verification systems was shown to be significant in short-range forecasts in the upper troposphere and stratosphere, especially for the zonal wind field in the Northern Hemisphere winter season. Medium-range forecasts were also improved overall when the horizontal position of the radiosonde data was considered.

Radiosonde observations are generally disseminated through the GTS in alphanumeric codes and these codes do not include information on the position and elapsed ascent time of the balloon. In this study, we assimilated radiosonde observations from three upper-air stations over the southern part of the Korean Peninsula. These observations included the exact position and time of the balloon, and hence we could consider balloon drift information directly without estimating that information using horizontal wind components and assumed rise speed (or a precalculated elapsed ascent time profile) like the previous studies. We used the 4DVAR method to assimilate radiosonde observations. The purpose of this study is to investigate the effect of considering balloon drift information on radiosonde data assimilation and subsequent forecasts using the Weather Research and Forecasting (WRF) Model and its data assimilation (WRFDA) system. To the best of our knowledge, this study is the first attempt to assimilate radiosonde observations while considering their exact positions and times using the 4DVAR method.

The characteristics of radiosonde data used in this study are provided in section 2. Section 3 includes a description of the weather for an observation period and the experimental designs for single-case and cycling experiments. In section 4, we discuss the results from the single-case and cycling experiments. A summary and conclusions are presented in section 5.

2. Radiosonde observations

In South Korea, radiosonde observations from seven upper-air stations (Baengnyeongdo, Osan, Kwangju, Heuksando, Cheju, Sokcho, and Pohang; refer to Fig. 1b for their locations) are regularly disseminated through the GTS. However, they do not include information on the position and elapsed ascent time of the balloon. The Korea Meteorological Administration (KMA) made special radiosonde observations to understand the internal structure of the changma front (a part of the quasi-stationary front of the East Asian summer monsoon; it is called mei-yu and baiu in China and Japan, respectively) in 2013. Balloons were launched at three observing stations (Boseong, Changwon, and Seongsan; refer to Fig. 1b for their locations) over the southern part of the Korean Peninsula twice a day (at 0000 and 1200 UTC; the observation interval was decreased to 6 or 3 h when rainfall occurred) from 20 June to 5 July 2013. The observed variables are pressure, wind speed, wind direction, temperature, and dewpoint temperature from near the surface to above 10 hPa (~30 km). The exact position and elapsed ascent time of the balloon are also available.

Fig. 1.
Fig. 1.

(a) Geographical areas of domains 1–3 with topographic height (m; shading) and (b) trajectories of balloons launched from the Boseong (red), Changwon (blue), and Seongsan (green) stations over domain 3. Max drift distances (km) are indicated. Locations of the KMA regular upper-air stations are denoted by black dots.

Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00161.1

Figure 1b shows trajectories of balloons launched from Boseong, Changwon, and Seongsan stations at 0010 UTC 4 July, 2330 UTC 3 July, and 0010 UTC 4 July 2013, respectively. All balloons drifted eastward in the troposphere and back to the station in the lower stratosphere according to the environmental wind. Maximum drift distances from the station were about 57.6, 65.8, and 47.0 km at Boseong, Changwon, and Seongsan stations, respectively. Figures 2a and 2b show vertical profiles of drift distance and elapsed ascent time for Boseong, Changwon, and Seongsan stations, and their mean. Drift distances and elapsed times of the balloon launches from 20 June to 5 July 2013 were averaged. The mean drift distances were a few kilometers in the lower troposphere, about 10–20 km in the midtroposphere, approximately 25–40 km in the upper troposphere, and greater than 40 km in the lower stratosphere. Drift distance increased rapidly near the 200-hPa level owing to the jet stream, and it decreased at levels above 50 hPa owing partly to vertical shear in the wind direction (i.e., wind direction is changed from westerly to easterly). The drift distances were the greatest at Boseong station, and they were the smallest at Seongsan station during the observation period. The mean elapsed times from balloon launch to various pressure levels were about 5 min at 850 hPa, 17 min at 500 hPa, 26 min at 300 hPa, 46 min at 100 hPa, and 1.11 h at 10 hPa. Compared to the global climatology of balloon drift statistics presented in Seidel et al. (2011), the maximum drift distance was similar to the climatological value and the elapsed time at 10 hPa was smaller than the climatological value. Vertical variations of the ascent speeds of the balloons launched at Boseong, Changwon, and Seongsan stations, along with their means, are shown in Fig. 2c. The mean ascent speed was fairly constant with height and close to the value of 6 m s−1. This value was greater than the mean ascent rate (~5 m s−1) reported in Laroche and Sarrazin (2013) and the ascent rate (5 m s−1) used at the National Centers for Environmental Prediction (NCEP) for estimating the horizontal drift position of radiosonde data (Keyser 2012). The ascent speed at Boseong (Changwon) station was smaller (greater) than the mean value.

Fig. 2.
Fig. 2.

Vertical profile of (a) drift distance (km), (b) elapsed time (s), and (c) ascent speed (m s−1) for the Boseong (red), Changwon (blue), and Seongsan (green) stations, and their mean value (black). Drift distances, elapsed times, and ascent speeds during the observation period from 20 Jun to 5 Jul 2013 are averaged.

Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00161.1

3. Case description and experimental design

a. Case description

Radiosonde observations including the position and elapsed ascent time of the balloon over the period from 20 June to 4 July 2013 were available in this study. We carried out a single-case experiment using a heavy rainfall case that occurred on 4 July 2013, and we also conducted cycling experiments for the whole observation period. Figure 3 shows time series of daily rainfall amount averaged over the Automatic Weather Stations (AWSs) between 33° and 36°N. The changma front usually influences the Korean Peninsula from the last 10 days of June to the middle of July, and hence rainfall is concentrated over this period. However, in 2013, the changma front was located south of the Korean Peninsula until the end of June, and it began to affect the Korean Peninsula from the first day of July. Therefore, there was little rainfall in June, and the rainfall on 25 June was caused by a midlevel trough passing across the Korean Peninsula. On 1 July 2013, the changma front induced heavy rainfall over the central part of the Korean Peninsula. Afterward, it moved southward and stayed over the southern part of the Korean Peninsula for about 5 days. We selected 2300 UTC 3 July 2013 as the analysis time for the single-case experiment because on that day there was considerable rainfall over the southern part of the Korean Peninsula, where special radiosonde observations including the balloon drift information were made.

Fig. 3.
Fig. 3.

Time series of daily rainfall amount (mm day−1) from 20 Jun to 4 Jul 2013. Daily rainfall amount from the KMA AWSs between 33° and 36°N are averaged.

Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00161.1

Figure 4 shows the synoptic environments over East Asia at 0000 UTC 4 July 2013. As previously mentioned, the changma front was located over the southern part of the Korean Peninsula (near 35°N) at that time. At the 850-hPa level, warm and moist air was transported to the Korean Peninsula by a westerly or southwesterly flow between the North Pacific high and a low pressure system (Fig. 4a). This increased the conditional instability defined as the partial derivative of the equivalent potential temperature with respect to pressure near the Korean Peninsula. There was also a large meridional gradient of equivalent potential temperature over the southern part of the Korean Peninsula, which was related to the changma front, and this made the atmosphere over the Korean Peninsula baroclinically unstable (Fig. 4b). This synoptic environment was favorable for the development of the heavy rainfall system associated with the changma front.

Fig. 4.
Fig. 4.

Synoptic environments at 0000 UTC 4 Jul 2013. (a) Geopotential height [m; black solid contours with a contour interval (CI) of 30 m], temperature (°C; red dashed contours with a CI of 3°C), and water vapor mixing ratio (kg kg−1; shaded for values greater than 0.012 kg kg−1) at 850 hPa. (b) Conditional instability between 850 and 500 hPa (K hPa−1; red solid contours with a CI of 0.005 K hPa−1) and equivalent potential temperature (K; shaded) at 850 hPa.

Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00161.1

Figure 5a shows the observed 12-h accumulated rainfall distribution made by using the observations from approximately 450 AWSs over the Korean Peninsula. Rainfall was concentrated over the southern part of the Korean Peninsula, where the changma front was located. The 12-h accumulated rainfall amount at the maximum point (at Hampyeong) was about 107.5 mm. There was also localized rainfall near Seoul with a maximum 12-h accumulated rainfall amount of about 70 mm. This rainfall was caused by the combination of convective instability and forced upward motion, and its spatial scale was very small. The observed hourly rainfall amount at Hampyeong from 0000 to 1200 UTC 4 July 2013 is shown in Fig. 5b. There were two rainfall peaks in the time series: one at 0900 UTC 4 July and the other at 1100 UTC 4 July 2013. The hourly rainfall amounts at 0900 and 1100 UTC were approximately 41.0 and 30.5 mm, respectively.

Fig. 5.
Fig. 5.

(a) Observed 12-h accumulated rainfall distribution [mm (12 h)−1] from 0000 to 1200 UTC 4 Jul 2013. Locations of the KMA AWSs are also shown. (b) Hourly rainfall amount (mm h−1) at Hampyeong (or the corresponding grid points in the case of model experiments) for the observations (black), and the NoDA (blue), All_BD (green), and No_BD (red) experiments.

Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00161.1

b. Experimental design

In this study, we conducted a single-case experiment initialized at 2300 UTC 3 July 2013 in cold-start mode, and then a 15-day experiment from 20 June to 4 July 2013 in cycling mode. The statistical robustness of the effects of considering balloon drift information can be obtained from the cycling experiments, and further details about the effects can be deduced from the single-case experiment. We used the Advanced Research version of the Weather Research and Forecasting (WRF) Model (ARW; Skamarock et al. 2008), version 3.5.1, for both the single-case and cycling experiments. We used triply nested domains with horizontal resolutions of 54, 18, and 6 km, respectively (Fig. 1a). The resolution of the innermost domain was determined by considering both balloon drift distances in Fig. 2a and computational cost. The 54-km domain covers East Asia including the Korean Peninsula, Japan, Taiwan, and eastern China. The 18-km domain covers the Korean Peninsula and the surrounding areas, and the 6-km domain focuses on South Korea. The numbers of horizontal grids for the 54-, 18-, and 6-km domain are 120 × 102, 121 × 103, and 121 × 127, respectively. In the vertical, there are 40 levels, and the model top is at 10 hPa for all domains.

The following physical parameterizations were used for all domains except for turning off the cumulus parameterization scheme on the 6-km domain: the WRF single-moment 6-class microphysics scheme (WSM6; Hong and Lim 2006), the Kain–Fritsch cumulus parameterization scheme (Kain 2004), the Yonsei University (YSU) planetary boundary layer scheme (Hong et al. 2006), the Rapid Radiative Transfer Model (RRTM) longwave radiation scheme (Mlawer et al. 1997), the Dudhia shortwave radiation scheme (Dudhia 1989), and the Noah land surface model (Chen and Dudhia 2001). For tangent linear and adjoint model runs related to the minimization of the 4DVAR cost function, the simplified linear physics schemes introduced in Zhang et al. (2013) were utilized. Initial and lateral boundary conditions for the 54-km domain were from the ECMWF interim reanalysis (ERA-Interim; Dee et al. 2011) dataset, and the initial and boundary conditions for the 18 (6)-km domains were from model outputs of the 54 (18)-km domain (i.e., one-way nesting was used). This configuration was applied to both the single-case and cycling experiments, except for the initial conditions of the cycling experiments. All data assimilation experiments were conducted on the 6-km domain, and the experimental period for the cycling runs was from 20 June to 4 July 2013.

We used the 4DVAR method included in WRFDA, version 3.5.1 (Huang et al. 2009; Barker et al. 2012). The WRF 4DVAR takes the incremental formulation (Courtier et al. 1994) in model space, and a control variable transform (Barker et al. 2004) is implemented as the preconditioning of the cost function. In the incremental 4DVAR approach, the cost function is minimized using the tangent linear and adjoint models derived from a simplified nonlinear model in the inner loop, and the nonlinear model trajectories are updated using the full nonlinear model in the outer loop. In this study, we employed a single outer loop, and in the inner loop, we used grid spacing of 6 km for the tangent linear and adjoint models, which was the same as that for the nonlinear model. We also used the conjugate gradient algorithm as an iterative minimization method, and the stopping criterion of the reduction in the gradient norm was set to 0.01 of its initial value. The background was from the 18-km domain run in the cold-start mode, and it was from the 12-h forecast starting from the analysis of the previous cycle in the cycling mode (except for the first cycle). The background error covariance was calculated using the National Meteorological Center (NMC; now known as the NCEP) method (Parrish and Derber 1992), where the background error statistics were derived from the differences between the 24- and 12-h forecasts. We used 24- and 12-h forecast differences (from 0000 and 1200 UTC to remove the diurnal cycle) in June 2013 to compute the background error covariance. The 24- and 12-h forecasts of the 54-km domain were conducted by using the ERA-Interim data as initial and boundary conditions, and the forecasts of the 18- and 6-km domains were nested in sequence. It is noted that only the 24- and 12-h forecasts of the 6-km domain were utilized in the calculation of the background error covariance. The radiosonde observations described in section 2 were assimilated using the 4DVAR method, and the observation error standard deviations were from prespecified errors for conventional observations included in the WRFDA system.

From 20 June to 4 July 2013, the 4DVAR analyses were made twice a day (at 2300 and 1100 UTC), and the 12-h forecast from the previous cycle was used as the background of the current cycle (i.e., 12-h cycling). The assimilation window for data assimilation experiments starts 1 h before the synoptic hour (i.e., 0000 or 1200 UTC) because of the typical launching time of the balloon (40–60 min before the synoptic hour) and it ends 1 h after the synoptic hour in consideration of the elapsed time of the balloon (Fig. 2b). In other words, for the 0000 UTC analyses, the assimilation window extends from 2300 (of the previous day) to 0100 UTC, and for the 1200 UTC analyses, it extends from 1100 to 1300 UTC. Although the frequency of the radiosonde observations used in this study was 2 min, the observations were binned every 10 min within the assimilation window when considering the elapsed ascent time of the balloon. A total of five experiments were conducted for the single-case run, and three experiments were carried out for the cycling runs (Table 1). In the NoDA experiment, no data assimilation was conducted. In the All_BD (No_BD) experiment, radiosonde data were assimilated using the 4DVAR method when considering all (no) balloon drift information (position and elapsed ascent time). In the BD_Position (BD_Time) experiment, only the position (elapsed time) of the balloon was considered when assimilating the radiosonde data.

Table 1.

Summary of numerical experiments.

Table 1.

4. Results and discussion

Results of the cycling experiments are presented in section 4a and those of the single-case experiment are given in section 4b. Verification results of the cycling experiments can give some statistical evidence for the positive impact of considering balloon drift information on the analyses and forecasts when assimilating radiosonde observations.

a. Cycling experiment

To verify analyses (initial conditions in the case of the NoDA experiment) or forecasts from the NoDA, No_BD, and All_BD experiments, radiosonde observations from the Boseong, Changwon, and Seongsan stations are used, and balloon drift information is considered for verification. The RMSEs of zonal wind, meridional wind, temperature, and dewpoint temperature are calculated using radiosonde observations targeted for 0000 UTC (launched before 0000 UTC) or 1200 UTC (launched before 1200 UTC), as well as the WRF forecasts inside the assimilation window (i.e., the forecasts between 0 and 2 h from the analysis time) of the NoDA, No_BD, and All_BD experiments (Fig. 6). As expected, the RMSEs of the All_BD experiment are smaller than those of the other experiments because the position and elapsed time of the balloon are considered in the verification. The RMSEs of all variables are increased sharply between 2 and 3 July 2013 (cycle numbers 24 and 27), especially those of zonal and meridional wind. Compared to the NoDA experiment, in the All_BD experiment, the RMSEs of zonal wind, meridional wind, temperature, and dewpoint temperature are reduced by 14.6%, 24.2%, 23.8%, and 25.3%, respectively. Although the radiosonde observations are assimilated without considering balloon drift information in the No_BD experiment, the RMSEs of zonal wind, meridional wind, temperature, and dewpoint temperature are reduced by 12.3%, 19.2%, 13.1%, and 21.4%, respectively, when compared to the NoDA experiment. These results indicate that radiosonde observations in consideration of balloon drift information are successfully assimilated in the All_BD experiment during the cycling period.

Fig. 6.
Fig. 6.

RMSEs of each cycle’s analysis (forecasts inside the assimilation window) verified against radiosonde observations (assimilated with or without considering balloon drift information) from the Boseong, Changwon, and Seongsan stations for the NoDA (blue), No_BD (red), and All_BD (green) experiments. The positions and elapsed times of the balloons are considered in the verification, and the average over all cycles for each experiment is also shown next to its experiment name: (a) zonal wind (m s−1), (b) meridional wind (m s−1), (c) temperature (K), and (d) dewpoint temperature (K).

Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00161.1

To check whether the above aspect (i.e., the RMSEs of the All_BD experiment are the smallest) is maintained when the WRF forecasts outside the assimilation window (i.e., the forecasts between 12 and 14 h from the analysis time) are considered in verification, the RMSEs of zonal wind, meridional wind, temperature, and dewpoint temperature are computed (Fig. 7). For all variables, the RMSEs of the All_BD experiment are the smallest among the experiments during most of the cycling period, and the improvement is noticeable in the RMSEs of zonal wind and temperature. Overall, the RMSEs of zonal wind and temperature (meridional wind and dewpoint temperature) for the No_BD experiment are smaller (larger) than those for the NoDA experiment. The RMSEs of the All_BD experiment are smaller than the other experiments even when considering the WRF forecasts outside the assimilation window, and this implies that the positive effect of data assimilation is sustained at least for 12–14 h in the All_BD experiment.1

Fig. 7.
Fig. 7.

As in Fig. 6, but for each cycle’s forecast (forecasts outside the assimilation window) and radiosonde observations, which are not assimilated.

Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00161.1

Although the daily rainfall amount is not large during the cycling period, except for on some days (e.g., 25 June and 2–4 July 2013), threat and bias scores are calculated to investigate the effects of radiosonde data assimilation with or without considering balloon drift information on the quantitative precipitation forecasts (QPFs). Threat and bias scores of the 3 (12)-h accumulated rainfall amount for a 0.1 (10.0)-mm threshold value are computed during the cycling period and averaged (Fig. 8). For a threshold value of 0.1 mm (i.e., light rainfall), the threat scores of the No_BD and All_BD experiments are greater than those of the NoDA experiment, and this shows the positive impacts of data assimilation on QPF skill. Furthermore, the threat score of the All_BD experiment is greater than that of the No_BD experiment, and this suggests that considering balloon drift information adds additional value to the QPF skill. Implications obtained from the threat scores can be confirmed by analyses of the bias scores. For a threshold value of 10.0 mm (i.e., moderate rainfall), the relative performance of each experiment varies according to a forecast ranges (0–3, 3–6, 6–9, 9–12, or 0–12 h). On average, the threat (bias) scores of the All_BD experiment are greater (closer to one) than for the other experiments. The threat (bias) scores of the No_BD experiment are greater (closer to one) than those of the NoDA experiment except for some forecast ranges (3–6 and 6–9 h). Therefore, regardless of the threshold values, the QPF skills are improved by assimilating radiosonde observations, and they are additionally improved by considering balloon drift information.

Fig. 8.
Fig. 8.

Threat and bias scores of 3- and 12-h accumulated rainfall for the NoDA (blue), No_BD (red), and All_BD (green) experiments. Threat and bias scores are calculated during the cycling period and averaged. (a) Threat score with a threshold value of 0.1 mm, (b) threat score with a threshold value of 10.0 mm, (c) bias score with a threshold value of 0.1 mm, and (d) bias score with a threshold value of 10.0 mm.

Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00161.1

b. Single-case experiment

A total of five experiments are conducted for the heavy rainfall case that occurred on 4 July 2013. The assimilation window is from 2300 UTC 3 July to 0100 UTC 4 July 2013, and a 13-h forecast from 2300 UTC 3 July to 1200 UTC 4 July 2013 is run using the nonlinear WRF Model. The positive impact of considering balloon drift information in the cycling experiments is shown in section 4a, and further details on the impact of considering balloon drift information in the single-case experiment are presented in this section.

Table 2 shows the RMSEs of the difference between the observation and background (OB) and the difference between the observation and analysis (OA) for zonal wind U, meridional wind V, temperature T, and water vapor mixing ratio Q. The number of assimilated observations for each variable is also shown. The RMSEs of OB and OA are derived by using all observations within the assimilation window. The number of assimilated observations can be different among the experiments and variables because the observations whose innovations (OB) are larger than a threshold value, which is defined as a multiple of the observation error, are rejected during the assimilation in the WRFDA system. In all data assimilation experiments, the RMSE of OA is reduced compared to the RMSE of OB as a result of the assimilation of radiosonde data. This implies that the radiosonde observations are successfully assimilated in all experiments. Although the specifics of assimilated observations vary in different experiments (depending on whether the position and/or elapsed time of the balloon are considered), the RMSE of OA for the All_BD experiment is relatively small compared to the other experiments, which indicates that the analysis of the All_BD experiment is closest to the observations.

Table 2.

RMSEs of OB and OA for zonal wind, meridional wind, temperature, and water vapor mixing ratio. The number of assimilated observations is also indicated.

Table 2.

Figure 9 shows the simulated 12-h accumulated rainfall amount from 0000 to 1200 UTC 4 July 2013. In the observations, the rainfall was concentrated mainly over the southwestern part of the Korean Peninsula, and the 12-h accumulated rainfall amount at Hampyeong was about 107.5 mm. There was also localized rainfall near Seoul with a 12-h accumulated rainfall amount of about 70 mm (Fig. 5a). In the NoDA experiment, the rainfall is distributed over the southern part of the Korean Peninsula, which coincides with the location of the simulated changma front in this experiment (not shown). The overall distribution of the simulated rainfall is similar to that of the observations. However, the 12-h accumulated rainfall amount at the grid point corresponding to Hampyeong (the closest grid point to Hampyeong; ~167.3 mm) is highly overestimated compared to the observations. In addition, two spurious rainfall maxima are simulated east of the observed rainfall, and the 12-h accumulated rainfall amounts at each maximum are approximately 146.1 and 156.6 mm, respectively (Fig. 9a). When the radiosonde observations are assimilated without considering the position and elapsed time of the balloon (No_BD experiment), the simulated rainfall distribution, compared to the observations, is qualitatively similar to the distribution of the NoDA experiment. However, the 12-h accumulated rainfall amount at the grid point corresponding to Hampyeong (~66.4 mm) is underestimated compared to the observations, and excessive rainfall [more than 180 mm (12 h)−1] is simulated near Boseong (one of the radiosonde stations; Fig. 9b). The rainfall forecast is improved when the radiosonde observations are assimilated with the position and elapsed time of the balloon (All_BD experiment) taken into consideration. The simulated rainfall distribution is very similar to the observations although the rainfall over the southwestern part of the Korean Peninsula is shifted slightly southward compared to the observations. The 12-h accumulated rainfall amount at the grid point corresponding to Hampyeong (~101.6 mm) is close to the observations, and the simulated rainfall along the southeastern coast corresponds well to the observations (Fig. 9c). It is also noted that the observed rainfall near Seoul is simulated only in the No_BD experiment although the rainfall distribution is displaced southwestward compared to the observations.

Fig. 9.
Fig. 9.

Simulated 12-h accumulated rainfall distribution [mm (12 h)−1] from 0000 to 1200 UTC 4 Jul 2013 for the (a) NoDA, (b) No_BD, (c) All_BD, (d) BD_Position, and (e) BD_Time experiments.

Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00161.1

To investigate the relative importance of the two types of balloon drift information, only the position (elapsed ascent time) of the balloon is considered in the BD_Position (BD_Time) experiment. In the BD_Position experiment, the main feature of the observed rainfall distribution is not simulated, and spurious rainfall near Boseong and along the southeastern coast is simulated (Fig. 9d). In the BD_Time experiment, the simulated rainfall distribution is similar to the observations, but excessive rainfall is incorrectly simulated south of the observed rainfall maximum point (i.e., Hampyeong; Fig. 9e). On the basis of the analyses of the simulated rainfall, it is concluded that the position and elapsed time of the balloon should be considered simultaneously when assimilating radiosonde observations. The results from only three experiments (i.e., NoDA, No_BD, and All_BD) will be considered for the remainder of this section.

Figure 5b shows hourly rainfall amount at Hampyeong (the point of maximum rainfall in the observations) from 0000 to 1200 UTC 4 July 2013. There was little rainfall until 0900 UTC, and then two peaks appeared in the observations, at 0900 UTC (~41.0 mm) and 1100 UTC (~30.5 mm) 4 July 2013. When no radiosonde observation is assimilated (NoDA experiment), rainfall is concentrated between 0400 and 0600 UTC 4 July 2013 (earlier than the observations), and the 3-h accumulated rainfall amount is nearly 113.8 mm (overestimated). When the radiosonde observations are assimilated without considering balloon drift information (No_BD experiment), the rainfall peak appears at 0400 UTC (earlier than in the observations), and the hourly rainfall amount is increased steadily from 0700 to 1200 UTC 4 July 2013. Although the rainfall peak appears earlier than the observations (at 0400 UTC) like in the other experiments, the overall pattern (i.e., bimodal shape) and rainfall amount of the peaks (~47.8 and ~31.2 mm) simulated in the All_BD experiment are more similar to the observations than in the other experiments.

To evaluate the rainfall forecast quantitatively, RMSEs of 6-h accumulated (from 0000 to 0600 UTC and from 0600 to 1200 UTC) and 12-h accumulated rainfall amount (from 0000 to 1200 UTC 4 July 2013) are calculated (Fig. 10). The RMSE of rainfall for the first 6-h forecast is the largest in the No_BD experiment. In the No_BD experiment, a large amount of rainfall is incorrectly simulated near the Boseong station during the first 6 h of simulation, which leads to the large RMSE value. During the second 6 h of simulation, the RMSE of rainfall in the NoDA experiment is the greatest. This is because excessive rainfall is simulated east of 127°E between 0600 and 1200 UTC 4 July 2013 in the NoDA experiment. Finally, the RMSE of 12-h accumulated rainfall in the All_BD experiment is the smallest among the experiments. It can be concluded that rainfall forecast of the All_BD experiment is better than those of the NoDA and No_BD experiments quantitatively as well as qualitatively.

Fig. 10.
Fig. 10.

RMSEs of 6- or 12-h accumulated rainfall [mm (6 h)−1 or mm (12 h)−1] for the NoDA (blue), No_BD (red), and All_BD (green) experiments.

Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00161.1

Analysis increments (difference between the analysis and background) of the No_BD and All_BD experiments are analyzed to explain the differences in the simulated rainfall. Figure 11 shows analysis increments of convective available potential energy (CAPE) for the All_BD experiment at 0000 UTC and the differences in CAPE between the NoDA and All_BD experiments at 0400 UTC 4 July 2013. Analysis increments of CAPE in most regions of the Korean Peninsula are negative, especially over the southwestern part of the Korean Peninsula. The decrease in CAPE found in the All_BD experiment compared to the NoDA experiment is more clearly shown at 0400 UTC 4 July 2013, when spurious rainfall begins in the NoDA experiment. This implies that convective instability over the southwestern part, where excessive rainfall is simulated in the NoDA experiment, is reduced in the All_BD experiment, and finally, overestimation of rainfall in that region is reduced in the All_BD experiment. The overall pattern of the analysis increment of CAPE for the No_BD experiment is similar to the All_BD experiment (not shown). Figure 12a shows the 800-hPa moist potential vorticity (MPV) and the water vapor mixing ratio of the All_BD experiment at 0700 UTC 4 July 2013. Although the atmosphere is stable with respect to vertical and horizontal displacements, it can be unstable with respect to displacement of the slant path. This instability, known as the conditional symmetric instability (CSI), can be diagnosed using the sign of MPV [i.e., unstable if the sign of the MPV is negative; Martin (2006)]. Over the southern part of the Korean Peninsula, the sign of the MPV is negative, and hence the atmosphere in that region is unstable in terms of the CSI. However, to have the CSI in some regions does not guarantee a large amount of rainfall over those regions. Over the regions near 35°N, where a large amount of rainfall is simulated in the All_BD experiment, the water vapor mixing ratio is also large, and there exist updrafts related to the changma front (not shown). This results in considerable rainfall near Hampyeong (the maximum rainfall point in the observations) in the All_BD experiment. Figure 12b shows the conditional instability between 850 and 400 hPa (defined as the partial derivative of the equivalent potential temperature with respect to pressure; a positive value denotes instability) and the 850-hPa vertical velocity of the No_BD experiment at 0100 UTC 4 July 2013. Some regions over the Korean Peninsula (e.g., west of Seoul, east of 128°E, and the southwestern part of the Korean Peninsula) are conditionally unstable. Over those regions, if there is updraft in place to lift warm and moist air to its level of free convection (LFC), a convective system may develop. Areas of convective instability coincide with areas of upward motion near the Boseong station, and this causes the simulated rainfall amount over those areas to be overestimated in the No_BD experiment. This indicates that the assimilation of radiosonde observations without considering balloon drift information in the No_BD experiment leads to an incorrect vertical structure of the atmosphere near the Boseong station.

Fig. 11.
Fig. 11.

Analysis increments of CAPE (J kg−1; red contours with negative values denoted by dashed contours) for the All_BD experiment at 0000 UTC, and the difference in CAPE between the NoDA and All_BD experiments (J kg−1; shaded) at 0400 UTC 4 Jul 2013.

Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00161.1

Fig. 12.
Fig. 12.

(a) MPV (PVU; shaded) and water vapor mixing ratio (kg kg−1; green solid contours with a CI of 0.001 kg kg−1) at the 800-hPa level for the All_BD experiment at 0700 UTC 4 Jul 2013. (b) Conditional instability between 850 and 400 hPa (K hPa−1; shaded) and vertical motion (m s−1; red solid contours with only positive values plotted) at the 850-hPa level for the No_BD experiment at 0100 UTC 4 Jul 2013.

Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00161.1

Short-range model forecasts of the NoDA, No_BD, and All_BD experiments are verified using the independent radiosonde observations (i.e., not assimilated). Figure 13 shows the vertical distribution of RMSEs of zonal wind, meridional wind, temperature, and dewpoint temperature for the NoDA, No_BD, and All_BD experiments. The RMSEs are calculated using radiosonde observations from the Boseong, Changwon, and Seongsan stations, and the position and the elapsed ascent time of the balloon are considered in the verification. Forecasts between 0500 (1100) UTC and 0700 (1300) UTC are verified against radiosonde observations launched before 0600 (1200) UTC 4 July 2013, and the RMSEs from the two periods are averaged. The RMSEs of zonal wind for the All_BD experiment are the smallest among the experiments, especially at upper levels. The RMSEs of zonal wind for the No_BD experiment are greater than those for the NoDA experiment above the 70-hPa pressure level. The RMSEs of meridional wind for the All_BD experiment are smaller than the NoDA and No_BD experiments at almost all levels although the differences are not substantial compared to the RMSEs of zonal wind. The RMSEs of temperature for the All_BD experiment are less than for the other experiments, and this feature is significant above the 300-hPa level. It is also noted that the RMSEs of temperature for the No_BD experiment are larger than for the NoDA experiment at upper levels. The differences in the RMSEs of dewpoint temperature are not comparatively large among the experiments. The RMSEs of dewpoint temperature for the All_BD experiment are slightly smaller than those for the other experiments between the 700- and 200-hPa levels. The RMSEs for the All_BD experiment, where the position and elapsed time of the balloon are considered in the assimilation, are the smallest, and the improvement in the All_BD experiment is substantial in the RMSEs of zonal wind and temperature at upper levels. The short-range (7 and 13 h) forecasts (valid at 0600 and 1200 UTC) of the NoDA, No_BD, and All_BD experiments are verified against the KMA operational Unified Model (UM; Davies et al. 2005) analyses, in which balloon drift information is not considered. Vertical distributions of RMSEs for zonal wind, meridional wind, temperature, and relative humidity are qualitatively similar to Fig. 13 except for some minor differences (not shown). In summary, the results of the single-case experiment reveal that not only rainfall forecasts but also meteorological-variable forecasts for the heavy rainfall case are improved by assimilating radiosonde observations with consideration of balloon drift information.

Fig. 13.
Fig. 13.

Vertical distribution of RMSEs for the NoDA (blue), No_BD (red), and All_BD (green) experiments. Radiosonde observations from the Boseong, Changwon, and Seongsan stations are used for verification, and the positions and elapsed times of the balloon are considered. Forecasts between 0500 (1100) UTC and 0700 (1300) UTC are verified against radiosonde observations targeted for 0600 (1200) UTC, and the RMSEs from the two periods are averaged: (a) zonal wind (m s−1), (b) meridional wind (m s−1), (c) temperature (K), and (d) dewpoint temperature (K).

Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00161.1

5. Summary and conclusions

Radiosonde observations are one of the crucially important components of numerical weather prediction (NWP). In general, it is assumed in NWP applications that radiosonde observations represent the vertical profile of the atmosphere for a fixed location (an observation station) and time (usually, the synoptic hour). However, in reality, balloons drift horizontally during their flights, and it takes about 1–2 h to finish their flights. Therefore, it is necessary to consider balloon drift information (i.e., position and elapsed ascent time of the balloon) when assimilating radiosonde observations into NWP models. Recently, Laroche and Sarrazin (2013) developed a method for estimating balloon drift position using horizontal wind components and a representative elapsed ascent time profile, and they assimilated radiosonde observations using the 4DVAR method to show the effects of balloon drift on analyses and forecasts of the EC global forecast system.

In this study, we used radiosonde observations from three observing stations (Boseong, Changwon, and Seongsan; originally targeted for understanding the changma front), and radiosonde observations from these stations included the exact positions and elapsed times of the balloons. Therefore, balloon drift information did not need to be estimated, unlike in Laroche and Sarrazin (2013). Moreover, we used the 4DVAR method and high horizontal resolution (6 km) to fully take into account balloon drift information when assimilating the radiosonde observations. We selected a heavy rainfall case that occurred on 4 July 2013 to investigate the impact of balloon drift information on forecasts of heavy rainfall over the Korean Peninsula. This case was related to the changma front over the southern part of the Korean Peninsula and led to a 12-h accumulated rainfall amount of more than 100 mm. We also conducted cycling experiments for the period from 20 June to 4 July 2013 to obtain the statistical robustness of the effects of balloon drift information on assimilating radiosonde observations.

The analysis and forecast of each cycle are verified against the radiosonde observations with consideration of balloon drift information. The RMSEs of all variables are reduced compared to the control experiment through the assimilation of radiosonde observations with consideration of balloon drift information at each analysis time, and the positive effect of data assimilation is retained during the 13-h forecast. Threat and bias scores during the cycling period show that assimilating radiosonde observations with consideration of balloon drift information leads to improved QPF skills.

The simulated rainfall distribution for the heavy rainfall case is improved by assimilating radiosonde observations with consideration of balloon drift information, and some degradation compared to the control experiment (i.e., no data assimilation) is observed when the radiosonde observations are assimilated without consideration of balloon drift information. From the additional experiments, it is concluded that it is necessary to consider both the position and the elapsed time of the balloon simultaneously in order to have a positive influence on the forecasts, and assimilating the KMA regular radiosonde observations with the estimated balloon drift information is not effective because of the limited number of observations (results not shown). Considering balloon drift information in the assimilation of radiosonde observations also leads to improvements in the simulated time series of hourly rainfall and the RMSEs of rainfall amount. Examination of the differences between the control and data assimilation experiments reveals that thermodynamic and dynamic features of the data assimilation experiment are modified through assimilation of the radiosonde observations with consideration of balloon drift information, and this modification results in improved rainfall forecasts. The RMSEs of the forecasted zonal wind, meridional wind, temperature, and dewpoint temperature are calculated using the radiosonde observations (which are not assimilated). Based on the analyses of the RMSEs, it can be deduced that forecasts of meteorological variables as well as forecasts of rainfall are improved through the assimilation of radiosonde observations with consideration of balloon drift information.

In conclusion, the positive impacts of the assimilation of radiosonde observations with consideration of balloon drift information are shown in the verification results of the cycling experiments, and further details about the positive impacts are shown in the forecasts of heavy rainfall over the Korean Peninsula. With the increasing resolution of NWP models and the use of sophisticated data assimilation methods, it is necessary to consider balloon drift information (both position and elapsed ascent time of the balloon) in the assimilation of radiosonde observations. A progressive transition of a transmission method through the GTS (i.e., from the alphanumeric code to the BUFR code) will add value to considering balloon drift information.

Acknowledgments

This work was funded by the Korea Meteorological Administration Research and Development Program under Grant CATER 2012-6080. This research was supported by the Brain Korea 21 Plus Project in 2014. This work was also supported by the National Institute of Supercomputing and Network/Korea Institute of Science and Technology Information with supercomputing resources including technical support (KSC-2014-C2-011). Radiosonde data used in this study were from the project Development and Application of Technology for Weather Forecast (NIMR-2013-B-1).

REFERENCES

  • Barker, D. M., , Huang W. , , Guo Y.-R. , , Bourgeois A. J. , , and Xiao Q. N. , 2004: A three-dimensional variational data assimilation system for MM5: Implementation and initial results. Mon. Wea. Rev., 132, 897914, doi:10.1175/1520-0493(2004)132<0897:ATVDAS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Barker, D. M., and et al. , 2012: The Weather Research and Forecasting (WRF) Model’s Community Variational/Ensemble Data Assimilation System: WRFDA. Bull. Amer. Meteor. Soc., 93, 831843, doi:10.1175/BAMS-D-11-00167.1.

    • Search Google Scholar
    • Export Citation
  • Chen, F., , and Dudhia J. , 2001: Coupling an advanced land surface–hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129, 569585, doi:10.1175/1520-0493(2001)129<0569:CAALSH>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Courtier, P., , Thépaut J.-N. , , and Hollingsworth A. , 1994: A strategy for operational implementation of 4D-Var, using an incremental approach. Quart. J. Roy. Meteor. Soc., 120, 13671387, doi:10.1002/qj.49712051912.

    • Search Google Scholar
    • Export Citation
  • Davies, T., , Cullen M. J. P. , , Malcolm A. J. , , Mawson M. H. , , Staniforth A. , , White A. A. , , and Wood N. , 2005: A new dynamical core for the Met Office’s global and regional modelling of the atmosphere. Quart. J. Roy. Meteor. Soc., 131, 17591782, doi:10.1256/qj.04.101.

    • Search Google Scholar
    • Export Citation
  • Dee, D. P., and et al. , 2011: The ERA-Interim reanalysis: Configuration and performance of the data assimilation system. Quart. J. Roy. Meteor. Soc., 137, 553597, doi:10.1002/qj.828.

    • Search Google Scholar
    • Export Citation
  • Dudhia, J., 1989: Numerical study of convection observed during the Winter Monsoon Experiment using a mesoscale two-dimensional model. J. Atmos. Sci., 46, 30773107, doi:10.1175/1520-0469(1989)046<3077:NSOCOD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Hong, S.-Y., , and Lim J.-O. , 2006: The WRF single-moment 6-class microphysics scheme (WSM6). J. Korean Meteor. Soc., 42, 129151.

  • Hong, S.-Y., , Noh Y. , , and Dudhia J. , 2006: A new vertical diffusion package with an explicit treatment of entrainment processes. Mon. Wea. Rev., 134, 23182341, doi:10.1175/MWR3199.1.

    • Search Google Scholar
    • Export Citation
  • Huang, X.-Y., and et al. , 2009: Four-dimensional variational data assimilation for WRF: Formulation and preliminary results. Mon. Wea. Rev., 137, 299314, doi:10.1175/2008MWR2577.1.

    • Search Google Scholar
    • Export Citation
  • Kain, J. S., 2004: The Kain–Fritsch convective parameterization: An update. J. Appl. Meteor., 43, 170181, doi:10.1175/1520-0450(2004)043<0170:TKCPAU>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Keyser, D., 2012: RAOB/PIBAL balloon drift latitude, longitude, and time calculation in PREPBUFR. NCEP/NMC, accessed 22 April 2014. [Available online at www.emc.ncep.noaa.gov/mmb/data_processing/prepbufr.doc/balloon_drift_for_TPB.htm.]

  • Laroche, S., , and Sarrazin R. , 2013: Impact of radiosonde balloon drift on numerical weather prediction and verification. Wea. Forecasting, 28, 772782, doi:10.1175/WAF-D-12-00114.1.

    • Search Google Scholar
    • Export Citation
  • MacPherson, B., 1995: Radiosonde balloon drift—Does it matter for data assimilation? Meteor. Appl., 2, 301305, doi:10.1002/met.5060020402.

    • Search Google Scholar
    • Export Citation
  • Martin, J. E., 2006: Mid-Latitude Atmospheric Dynamics: A First Course. J. Wiley and Sons, 324 pp.

  • McGrath, R., , Semmler T. , , Sweeney C. , , and Wang S. , 2006: Impact of balloon drift errors in radiosonde data on climate statistics. J. Climate, 19, 34303442, doi:10.1175/JCLI3804.1.

    • Search Google Scholar
    • Export Citation
  • Mlawer, E. J., , Taubman S. J. , , Brown P. D. , , Iacono M. J. , , and Clough S. A. , 1997: Radiative transfer for inhomogeneous atmospheres: RRTM, a validated correlated-k model for the longwave. J. Geophys. Res., 102, 16 66316 682, doi:10.1029/97JD00237.

    • Search Google Scholar
    • Export Citation
  • Parrish, D. F., , and Derber J. , 1992: The National Meteorological Center’s spectral statistical-interpolation analysis system. Mon. Wea. Rev., 120, 17471763, doi:10.1175/1520-0493(1992)120<1747:TNMCSS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Seidel, D. J., , Sun B. , , Pettey M. , , and Reale A. , 2011: Global radiosonde balloon drift statistics. J. Geophys. Res., 116, D07102, doi:10.1029/2010JD014891.

    • Search Google Scholar
    • Export Citation
  • Skamarock, W. C., and et al. , 2008: A description of the Advanced Research WRF version 3. NCAR Tech. Note TN–475+STR, 113 pp. [Available online at http://www2.mmm.ucar.edu/wrf/users/docs/arw_v3.pdf.]

  • Zhang, X., , Huang X.-Y. , , and Pan N. , 2013: Development of the upgraded tangent linear and adjoint of the Weather Research and Forecasting (WRF) Model. J. Atmos. Oceanic Technol., 30, 11801188, doi:10.1175/JTECH-D-12-00213.1.

    • Search Google Scholar
    • Export Citation
1

The RMSEs of each cycle’s analysis and forecast for zonal and meridional wind are comparatively large. This may be because of the consideration of balloon drift in verification, an insufficient assimilation of radiosonde observations, and a model bias. However, this is beyond the scope of this study.

Save