Comparisons in the global planetary boundary layer height obtained from COSMIC radio occultation, radiosonde, and reanalysis data

Jie Gu ,Yehui Zhng ,N Yng ,Rui Wng

a School of Hydrology and Water Resources, Nanjing University of Information Science and Technology, Nanjing, China

b Rugao Meteorologic Bureau, Rugao, China

c State Oceanic Administration, Key Laboratory for Polar Science, Polar Research Institute of China, Shanghai, China

Keywords:Planetary boundary layer height Radiosonde COSMIC Reanalysis

ABSTRACT The global planetary boundary layer height (PBLH) estimated from 11 years (2007–17) of Integrated Global Radiosonde Archive (IGRA) data, Constellation Observing System for Meteorology, Ionosphere and Climate (COSMIC) soundings, and European Center for Medium-Range Weather Forecasts (ECMWF) interim reanalysis (ERAInterim) data, are compared in this study.In general, the spatial distribution of global PBLH derived from ERAInterim is consistent with the one from IGRA, both at 1200 UTC and 0000 UTC.High PBLH occurs at noon local time, because of strong radiation energy and convective activity.There are larger differences between the results of COSMIC and the other two datasets.PBLHs derived from COSMIC are much higher than those from radiosonde and reanalysis data.However, PBLHs derived from the three datasets all exhibit higher values in the low latitudes and lower ones in the high latitudes.The latitudinal difference between IGRA and COSMIC ranges from − 1700 m to − 500 m, while it ranges from − 500 m to 250 m for IGRA and ERA-Interim.It is found that the differences among the three datasets are larger in winter and smaller in summer for most studied latitudes.

1.Introduction

The planetary boundary layer (PBL) comes into direct contact with the surface and dominates the vertical exchange of momentum, heat,moisture, and trace substances, such as aerosols, on an order of one hour or less ( Stull, 1988 ).It is important to accurately estimate the PBL height (PBLH) –which is strongly influenced by physical factors,including evaporation, sensible heat flux, wind, and frictional drag –to promote weather prediction, climate monitoring, and air quality monitoring ( Lee and Kawai, 2011 ).The PBLH is a significant parameter in weather and pollutant forecasting models since it determines the volume into which pollutants can disperse ( Jordan et al., 2010 ).A higher PBLH causes a faster vertical turbulent exchange with a reduction in aerosol and contaminant concentrations.

At present, there are many instruments and datasets used to estimate the PBLH.Each kind of observation, with a specific method, in fact detects the PBLH according to one of several definitions ( Shi et al.,2020 ), each with its own advantages and disadvantages.One of the most common type of data in determining the PBLH is long-term global radiosonde data.Such datasets are widely used because they can describe the thermodynamics and dynamics of the lower atmosphere simultaneously.The methods and algorithms used to estimate the PBLH based on radiosonde observations have been discussed in various studies.Seidel et al.(2012) discussed the uncertainties in the PBLH calculated by different methods and found that the bulk Richardson (Ri) number (BRN) method can be used as a consistent method to estimate the stable and convective boundary layer height (BLH) when using large datasets.Besides radiosonde data, other data, such as surfaced-based remote sensing data including measurements from wind profile lidars,ceilometers, and SODAR (sonic detection and ranging), can also be used for PBLH estimation.In addition, with the rapid development of Global Positioning System radio occultation (GPS RO) technology, some new datasets for global PBLH monitoring have been introduced, such as the new generation of GPS occultation missions represented by the Constellation Observing System for Meteorology, Ionosphere and Climate (COSMIC).The minimum gradient method and wavelet covariance transform(WCT) method are commonly used to calculate the PBLH in COSMIC( Ratnam and Basha, 2010),and the WCT method can effectively detect small boundary layer transitions.Previous studies have tended to analyze PBLH at a single site or in an individual area.On the contrary, the present paper compares global PBLH data across three different types of data (radiosonde data, GPS RO data, and reanalysis data), at two times of day, and the differences in the results are analyzed systematically.

More specifically, we compare the global PBLH at 1200 UTC and 0000 UTC derived from radiosonde, satellite, and reanalysis data covering the period 2007–17.In addition, the latitudinal differences among the three datasets are discussed.In Section 2, the data and methods are introduced, the details are listed in Table 1. Section 3 reports the characteristics of the PBLH, such as the global distribution and the latitudinaland seasonal-averaged PBLH differences among the three datasets.A summary follows in Section 4.

Table 1 Descriptions of the three datasets in terms of their resolutions and estimation methods.

2.Data and methods

In this section, the three different types of data sets and the corresponding methods for PBLH estimations used in this paper are described in detail.The key information is summarized in Table 1 .

2.1. Radiosonde

Integrated Global Radiosonde Archive (IGRA) data provided by the National Climate Data Center are used (available at http://www1.ncdc.noaa.gov/pub/data/igra/ ).This radiosonde dataset includes more than 2700 observational stations around the world and two soundings per day (1200 UTC and 0000 UTC).IGRA provides profiles of pressure, temperature, relative humidity, and wind, amongst other variables.The dataset has a good integrity and high spatial and temporal density.Climatological statistics in this paper are derived from twice-daily sampling of the data from 1973 to 2017, since the data after 1973 are relatively better quality ( Wang and Wang, 2016 ).The BLH is reported to be less affected by data inhomogeneities after 1973( Zhang et al., 2013 ).

2.2. COSMIC

COSMIC is a cooperative project of several institutions in Taiwan of China and the United States ( http://cosmicio.cosmic.ucar.edu/cdaac/ ).It was launched in April 2006 for the Global Navigation Satellite System occultation constellation service detection test ( Anthes et al., 2008 ).The dataset is composed of dry profile and wet profile data.Because the existence of water vapor is not considered in the dry profile data, the deviation of air pressure and temperature is large below 10 km.In the study of boundary layer meteorology, most scholars use wet profile data( Ho et al., 2015 ; Kishore et al., 2016 ).In the wet profile, COSMIC provides four meteorological parameters: height, temperature, air pressure,and refractivity up to 40 km.The vertical resolution is 100 m and over 1000 occultations are distributed globally in one day.In this study, refractivity in COSMIC from 2007 to 2017 has been used for estimating the PBLH.

2.3. Reanalysis

The third-generation reanalysis data from the European Center for Medium-Range Weather Forecasts (ECMWF) interim reanalysis (ERAInterim) (available at https://apps.ecmwf.int/datasets/data/interimfull-daily/levtype = sfc/ ) have been used for PBLH estimation from 1979( Paul et al., 2009 ).Compared with previous reanalysis data, this generation of reanalysis data uses more advanced 4DVAR and 3DVAR-GSI assimilation technology, which improves the spatial resolution.In this study, we directly use the BLH in ERA-Interim (2007–17), which is estimated by the parcel method ( Troen and Mahrt, 1986 ).The spatial resolution of the data is 1°×1°and the time step is 3 h.

2.4. Methods

To determine the PBLH from the IGRA radiosonde data, the BRN method is used, which has been evaluated as an appropriate method to estimate the PBLH for climatological analyses in large datasets( Seidel et al., 2012 ).This method is not strongly dependent on the vertical resolution of the dataset and can identify a non-negative height in all cases ( Zhang et al., 2013 ).Soundings with fewer than four levels of wind speed records and seven levels of records with temperature profiles below 5 km have been discarded.The formula of the BRN is as follows:

where

z

is height and s denotes the surface;

g

is gravitational acceleration;

θ

z

and

θ

are the virtual potential temperature at

z

and

z

s ;

u

and

v

are zonal and meridional wind speeds, respectively;

b

is a constant;and

u

is the surface friction velocity.As a parameter in the denominator,

u

∗ is much smaller than the wind shear items and not significant in stable and neutral conditions, so we ignore surface friction terms( Vogelezang and Holtslag, 1996 ).Because winds at the surface (2-m level) are not included in radiosonde data, we calculated BRN profiles with the wind speed as 0 m s1 at the surface and the wind speed at

z

= 10 m (the first level reported in the wind records).The results were similar.In this paper, the surface winds are set to zero for consistency with other radiosonde data studies.We scanned the profile from the surface upwards and obtained the first level with Ri ≥ 0.25.This level and the next lowest level can derive

z

(Ri = 0.25) by linear interpolation, as the PBLH.Seidel et al.(2012) discussed the uncertainties between the interpolation-estimated PBLH and the closest reported data level.For low PBLH in the IGRA data, the interpolation causes a large percentage uncertainty, but for PBLH

>

1 km, the uncertainty is generally well below 20%.

Fig.1.(a) Monthly variations of PBLH derived from the BRN and parcel methods at station OSAN AB (37.08°N, 127.03°E, 11.8 m) from IGRA.Scatterplot between the BRN and (b) parcel methods of IGRA, and (c) WCT methods of IGRA.(d) Scatterplot between COSMIC and IGRA data using the WCT method.

To verify the results derived from the BRN method, we also used the parcel method ( Seibert et al., 2000 ), which is also used for the ERAInterim PBLH product, to estimate PBLH.The parcel method can determine a PBLH by comparing the virtual potential temperature at the surface to the values aloft and evaluating the height at which the virtual potential temperature is the same as at the surface.

To compare the results of these two methods, the comparison at one station is shown.Monthly averaged PBLH from 1973 to 2017 at OSAN AB (37.08°N, 127.03°E, 11.8 m) is shown in Fig.1 (a).The results of the two methods are highly consistent and their correlation is 0.9 in Fig.1 (b).Similar results can be found at most other stations (not shown).According to the comparisons, the PBLH results estimated by the BRN and parcel method are similar.

The BRN method cannot be used in the COSMIC dataset since no wind observations are provided.Here, we use the WCT method based on the Haar function to calculate the PBLH from COSMIC.The Haar wavelet function

h

,as defined by Brooks (2003) is:

where

z

is the height,

a

is the dilation, and

c

is the center of the Haar function.The covariance transform of the Haar function

W

is defined as:

where

f

(

z

) is the profile of refractivity (profiles with no value below 0.5 km have been discarded), and

z

and

z

are the upper and lower limits of the profile, respectively.The selection of an appropriate

a

in Eq.(2) is a key challenge in the retrieval of the true PBLH.Since

f

(

z

) is not a continuous profile in COSMIC, Eq.(2) can simplify as a discrete form, as clearly mentioned in Basha et al.(2018) :

where

a

=

n

Δ

z

,

n

= 2, 4, 6, 8, ……, and the center position

c

should be selected between two discrete data points,

c

−

a

/2 and

c

+

a

/2, to make sure that the number of data points in each integral is equal.The BLH is determined by the substantial change in backscatter, which is detected by the location of the maximum in

W

(

a,

c

).A dilation of

a

= 200 m has been selected in this paper since this value avoids the need for subjective bounds on the PBLH ( Wiegner et al., 2006 ).

To compare the PBLH using the WCT method over two datasets, we interpolate IGRA profiles to a 100-m vertical resolution.Taking Abu Dhabi station as an example, Fig.1 (c) shows the scatterplot of IGRA data calculated by the BRN and WCT methods.The correlation coefficient in the daytime (1200 UTC, red) is about 0.45, while at nighttime (0000 UTC, blue) it is 0.29.It seems that in the daytime these two methods have a fair match, while at nighttime they have a poor match.Fig.1 (d) shows the scatterplot of IGRA- and COSMIC-derived PBLH by using the same WCT method.COSMIC data with latitude and longitude within 2° and time difference within 2 h are selected near the ground station of Abu Dhabi from 2007 to 2017.There are 77 soundings found.The results show poor correlations between the two datasets, and the PBLH estimated by COSMIC is much higher than that of IGRA.

Fig.2.Global averaged PBLH distribution in June–July–August from 2007 to 2017 at (a, c, e) 1200 and (b, d, f) 0000 UTC derived from (a, b) IGRA, (c, d)ERA-Interim, and (e, f) COSMIC.

3.Results and discussion

3.1. Global distribution of PBLH

The seasonal spatial distribution of global PBLH at 1200 UTC and 0000 UTC has been studied.Fig.2 shows the results in June, July, and August (JJA) from 2007 to 2017.For IGRA, where the number of profiles involved in the calculation is less than 50 in each season, these data have been discarded.The total qualifying number of IGRA stations at 1200 UTC and 0000 UTC is 575 and 318, respectively.The spatial resolution of ERA-Interim shown here is 1°×1°, and that of COSMIC is 2.5°×2.5°.

It is found that the global spatial distribution of PBLH from IGRA is consistent with that of ERA-Interim both at 1200 UTC and 0000 UTC.At 1200 UTC, high PBLHs estimated from IGRA and ERA-Interim appear in similar areas (such as Europe and northern and southern parts of Africa) where the local time is around noon.Clearly, the PBLH estimated from COSMIC is larger than that derived from the other two datasets in most areas.Similar results can be found in Europe and northern and southern parts of Africa.In Europe and western Asia, the average PBLH is lowest in IGRA (~1200–1500 m) and highest in COSMIC(~1500–1800 m).In northern and southern parts of Africa, however,the PBLH estimated from COMISC (~2100–2700 m) is lower than from ERA-Interim (~2400–2700 m).The results at each grid point of ERAInterim are determined by the parcel method, and this method usually derives a convective BLH, which is generally higher than the neutral and stable layer height.The results from the grid of COSMIC are the average of three boundary layer regimes (stable boundary layer, neutral residual layer, and convective boundary layer), which may lead to a relative lower value.At 0000 UTC, the larger PBLH values occur in southern North America, where it is also noon local time, with strong radiation energy and convective activity.The PBLH at low latitudes is generally higher than that at high latitudes in both IGRA and ERA-Interim.Due to the stronger solar radiation at low latitudes and a longer exposure time to the sun, more radiation energy can be transported upwards from the surface, which may lead to a higher PBLH.In addition, the higher PBLHs in Africa may be associated with the higher near-surface temperature and lower near-surface relative humidity ( Zhang et al., 2013 ).

Fig.2 (e, f) shows the averaged PBLH distribution using the COSMIC dataset.This is consistent with the averaged PBLH spatial distribution (derived from all the observation times) in Basha et al.(2018) .The 1200 UTC PBLH is larger and covers a wider region than that at 0000 UTC.The PBLH is not shown in the Antarctic region owing to the sparse availability of data.The PBLHs over land and ocean are different, being higher over the former (reaching ~2700 m or more) than the latter (mainly reaching ~1200–1800 m).This is consistent with the results of Ho et al.(2015).Due to the complex factors of the land surface and ocean environment, there will be a considerable impact on the near-surface atmosphere, and thus the top of the boundary layer will be affected.

Fig.3.Scatterplots of latitudinally and monthly averaged (a) IGRA and COSMIC PBLH at 1200 UTC, (b) IGRA and ERA-Interim PBLH at 1200 UTC, (c) IGRA and COSMIC PBLH at 0000 UTC, and (d) IGRA and ERA-Interim PBLH at 0000 UTC.

The PBLHs over the three datasets are all higher at low latitudes and lower at high latitudes.Results from the radiosonde and reanalysis data agree well.However, there are larger differences between the results of COSMIC and the other two datasets.The PBLHs derived from COSMIC are much higher than those from the radiosonde and reanalysis data.To clarify the specific differences among the three datasets, latitudinally averaged PBLHs are compared in the following section.

3.2. Latitudinally averaged PBLH comparison

Fig.3 shows scatterplots of the latitudinally and monthly averaged PBLH.A total of 36 latitudinal bins with widths of 5° were selected to estimate the latitudinally and monthly averaged PBLH.Fig.3 (a, c) compares IGRA and COSMIC.At 1200 UTC, the two sets of observations show relatively better agreement (correlation reaching 0.526) than at 0000 UTC (correlation is only 0.236).The results of COSMIC at both 0000 and 1200 UTC are much larger than those of IGRA.The lower PBLH values cannot be estimated in COSMIC because the closer to the surface, the more inaccurate the observed profiles.In addition, the vertical resolution of COSMIC is 100 m and the bottom of the profiles cannot be observed, which may lead to an overestimation.

The differences between IGRA and ERA-Interim are displayed in Fig.3 (b, d).The correlation is good and relatively consistent.The coefficients are close to 0.6 at both 1200 UTC and 0000 UTC.It is found that the estimation of the ERA-Interim is slightly larger.Because the PBLH derived from ERA-Interim is mainly estimated by the parcel method,which is suitable for the height of the convective boundary layer, thus the result is larger than that of IGRA.In addition, it is ignored in the data assimilation process, which will also cause deviation between the results of ERA-Interim and IGRA.

Furthermore, IGRA’s stations are basically all on land or islands,while the latitudinally averaged PBLHs from COSMIC and ERA-Interim both include results over ocean.This may also bring discrepancies into the comparisons.

3.3.Latitudinally and seasonally averaged PBLH comparison

To further quantify the differences in the three datasets, the latitudinal difference in the PBLH among the three datasets in four seasons(December–January–February (DJF), March–April–May (MAM), June–July–August (JJA), and September–October–November SON)) were analyzed (not shown).

According to Fig.3 (a, c), the IGRA-estimated PBLH is generally lower than that from COSMIC.The difference between IGRA and COSMIC ranges from − 1700 m to − 500 m.Regarding the seasonal variations,it is found that the differences are quite close for MAM, JJA, and SON at most latitudes at 1200 UTC.This phenomenon is only found in the equatorial regions in the Southern Hemisphere at 0000 UTC.Interestingly, the difference shows obvious seasonal variations at midlatitudes in the Southern Hemisphere at 0000 UTC.The differences are larger in winter and smaller in summer.

The seasonal PBLH differences between IGRA and ERA-Interim are also analyzed.In general, the PBLH estimated from ERA-Interim seems slightly larger than that from IGRA.The results show that the seasonal difference near the Antarctic region is the largest.However, the poor level of data both quantitatively and qualitatively in this region may indicate there is considerable uncertainty in that result.Except in the Antarctic, the difference between the two datasets ranges from − 500 m to 250 m.The seasonal variations in the differences between IGRA and ERA-Interim are clear.The difference in the equatorial regions at both times of the day and in northern midlatitudes at 1200 UTC are similar,ranging from − 400 m to 0 m.For most regions in the Southern Hemisphere, the differences are larger in winter and smaller in summer.Similar seasonal changes are also found in the regions within 40°–70°N at 1200 UTC and 10°–60°N at 0000 UTC.

In summary, the PBLH is generally higher in summer and with quite low values in winter, while the differences among the three datasets are larger in winter and smaller in summer.We propose several possible reasons for why this is the case.According to the uncertainty test devised in Seidel et al.(2012),the uncertainties in the BRN method used for radiosonde data can be large ( > 50%) for low PBLHs and quite small ( <20%) for high PBLHs ( > 1 km).On the other hand, the coarse vertical resolution of COSMIC and the poor data quality in the bottom of the profiles may lead to a large uncertainty in the low PBLHs.Moreover,the parcel method used in ERA-Interim is basically good for determining the convective boundary layer, but not for the lower PBLH.The IGRA station locations may also contribute to these discrepancies.

4.Conclusion

In this paper, the global distribution of the PBLH and latitudinally and seasonally averaged PBLH differences among three datasets (IGRA,COSMIC, and ERA-Interim) from 2007 to 2017 have been investigated.We found that the global spatial distribution of IGRA is consistent with that from ERA-Interim, both at 1200 UTC and 0000 UTC.Higher PBLHs occur in areas where it is noon local time, with strong radiation energy and convective activity.There are larger differences between the results of COSMIC and the other two datasets.PBLHs from COSMIC are much higher than those from the radiosonde and reanalysis data.These differences may be caused by fewer stations over ocean in IGRA and the PBLH derived from ERA-Interim being more effective over land.However, the PBLHs derived from the three datasets all exhibit higher values at low latitudes and lower ones at high latitudes.

The latitudinal difference between IGRA and COSMIC ranges from− 1700 m to − 500 m, while it ranges from − 500 m to 250 m between IGRA and ERA-Interim.It was found that the differences among the three datasets are larger in winter and smaller in summer for most studied latitudes.This may be associated with (1) the large uncertainties ( >50%) in the BRN method used for radiosonde data for the low PBLH; (2)the coarse vertical resolution of COSMIC and the poor data quality in the bottom of the profiles; and (3) the parcel method used in ERA-Interim not being suitable for the lower PBLH.Overall, the findings improve our understanding of PBLH differences derived from different datasets.It serves as a reference for studies of the atmospheric boundary layer.

Funding

This work was supported by the Meteorological Research Open Foundation of Huaihe Basin [grant number HRM201604 ].

Acknowledgments

We are grateful to the COSMIC Data Analysis and Archive Center,the National Climatic Data Center, and the ECMWF for providing the observed data.