water vapor variability and comparisons in the subtropical...

Water vapor variability and comparisons in the subtropical Pacificfrom The Observing System Research and PredictabilityExperiment‐Pacific Asian Regional Campaign (T‐PARC)Driftsonde, Constellation Observing System for Meteorology,Ionosphere, and Climate (COSMIC), and reanalyses

Junhong Wang,1 Liangying Zhang,1 Po‐Hsiung Lin,2 Mark Bradford,1 Harold Cole,1

Jack Fox,1 Terry Hock,1 Dean Lauritsen,1 Scot Loehrer,1 Charlie Martin,1

Joseph VanAndel,1 Chun‐Hsiung Weng,2 and Kathryn Young1

Received 14 May 2010; revised 22 July 2010; accepted 2 August 2010; published 4 November 2010.

[1] During the THORPEX (The Observing System Research and Predictability Experiment)Pacific Asian Regional Campaign (T‐PARC), from 1 August to 30 September 2008, ∼1900high‐quality, high vertical resolution soundings were collected over the Pacific Ocean.These include dropsondes deployed from four aircrafts and zero‐pressure balloons in thestratosphere (NCAR’s Driftsonde system). Thewater vapor probability distribution and spatialvariability in the northern subtropical Pacific (14°–20°N, 140°E–155°W) are studied usingDriftsonde and COSMIC (Constellation Observing System for Meteorology, Ionosphere, andClimate) data and four global reanalysis products. Driftsonde data analysis shows distinctdifferences of relative humidity (RH) distributions in the free troposphere between the Easternand Western Pacific (EP and WP, defined as east and west of 180°, respectively), very drywith a single peak of ∼1% RH in the EP and bi‐modal distributions in the WP with one peaknear ice saturation and one varying with altitude. The frequent occurrences of extreme dryair are found in the driftsonde data with 59% and 19% of RHs less than or equal to 5% andat 1% at 500 hPa in the EP, respectively. RHwith respect to ice in the free troposphere exhibitsconsiderable longitudinal variations, very low (<20%) in the EP, but varying from 20% to100% in theWP. Inter‐comparisons ofDriftsonde, COSMIC and reanalysis data show generallygood agreement among the Driftsonde, COSMIC, ECMWFReanalysis‐Interim (ERA‐Interim)and Japanese Reanalysis (JRA) below 200 hPa. The ERA‐Interim and JRA are approved to besuccessful on describing RH frequency distributions and spatial variations in the region. Thecomparisons also reveal problems in Driftsonde, two National Center for EnvironmentalPrediction (NCEP) reanalyses and COSMIC data. The moist layer at 200–100 hPa in the WPshown in the ERA‐Interim, JRA and COSMIC is missing in Driftsonde data. Major problemsare found in the RH means and variability over the study region for both NCEP reanalyses.Although the higher‐moisture layer at 200–100 hPa in theWP in the COSMIC data agrees wellwith the ERA‐Interim and JRA, it is primarily attributed to the first guess of the 1‐Dimensional(1D) variational analysis used in the COSMIC retrieval rather than the refractivitymeasurements. The limited soundings (total 268) of Driftsonde data are capable of portrayingRHprobability distributions and longitudinal variability. This implies thatDriftsonde systemhasthe potential to become a valuable operational system for upper air observations over the ocean.

Citation: Wang, J., et al. (2010), Water vapor variability and comparisons in the subtropical Pacific from The Observing SystemResearch and Predictability Experiment‐Pacific Asian Regional Campaign (T‐PARC) Driftsonde, Constellation ObservingSystem for Meteorology, Ionosphere, and Climate (COSMIC), and reanalyses, J. Geophys. Res., 115, D21108,doi:10.1029/2010JD014494.

1. Introduction

[2] It is undisputable that water vapor in the atmosphereserves as an effective greenhouse gas and plays animportant role in determining the sensitivity of Earth’sclimate to anthropogenic forcing [Held and Soden, 2000;

1Earth Observing Laboratory, National Center for AtmosphericResearch, Boulder, Colorado, USA.

2Department of Atmospheric Sciences, National Taiwan University,Taipei, Taiwan.

Copyright 2010 by the American Geophysical Union.0148‐0227/10/2010JD014494

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 115, D21108, doi:10.1029/2010JD014494, 2010

D21108 1 of 14

http://dx.doi.org/10.1029/2010JD014494

Pierrehumbert et al., 2006]. The effect of water vapor onEarth’s radiation budget (mainly outgoing longwave radia-tion (OLR)) is approximately logarithmic in specifichumidity and depends on vertical and horizontal distributionsand magnitudes of water vapor [Held and Soden, 2000]. TheOLR is more sensitive to water vapor changes in the freetroposphere than in the boundary layer and is also moresensitive to very low humidity than high humidity [e.g., Shineand Sinha, 1991; Spencer and Braswell, 1997].[3] Above the boundary layer, much of the atmosphere is

unsaturated and often contains very dry air of less than 10%relative humidity (RH) in the subtropics and extratropics[e.g., Spencer and Braswell, 1997; Zhang et al., 2003].The sensitivity of the OLR is greater to perturbations inwater vapor in the subtropical free troposphere than that in thetropics [Held and Soden, 2000]. There have been extensivestudies on probability density functions (PDFs) of RH in thetropical and subtropical troposphere [e.g., Spencer andBraswell, 1997; Zhang et al., 2003; Sherwood et al., 2006;Ryoo et al., 2009]. Given the importance of the subtropicaldry air to the radiation budget, a number of studies have beenfocused on understanding large‐scale control of subtropi-cal humidity using Lagrangian techniques [e.g., Salathéand Hartmann, 1997; Soden, 1998; Pierrehumbert,1998; Pierrehumbert and Roca, 1998; Galewsky et al.,2005]. The humidity simulated by the Lagrangian schemesor other empirical models needs to be evaluated against in situor satellite observations.[4] Upper air moisture observations over the ocean are

primarily obtained from satellites, especially microwave sa-tellites for all weather conditions. Satellite data can onlyprovide total column water vapor (TCWV) or layer meanhumidity. As elaborated above, Earth’s radiation budget isvery susceptible to the free troposphere humidity, whichcontributes less to TCWV. The layer mean humidity wouldsmear out the large humidity variability and skew the RHhumidity distribution. In situ soundings (both ship and islandradiosonde and dropsonde from aircrafts) from special fieldexperiments have been used to study water vapor distributionin the tropics [Brown and Zhang, 1997; Pierrehumbert,1998; Zhang et al., 2003]. However, those soundings aresubject to spatial and temporal sampling biases.[5] Global reanalysis is a comprehensive global, multi-

decadal data set generated by a constant state‐of‐the‐artnumerical data assimilation technique using various pastobservations. Satellite upper‐air observations have been theprimary input data for global reanalyses. The NCEP reanal-ysis products only assimilate satellite temperature soundingsover the ocean. Without the observational constrain to themoisture field, NCEP reanalyses perform poorly over theocean [Trenberth and Guillemot, 1998; Trenberth et al.,2005]. There is an urgent need to validate all reanalysisproducts over the ocean using independent data that are notassimilated into the reanalyses. The data over the ocean col-lected from various field campaigns can partially meet thisneed.[6] In recent years, the Global Positioning System (GPS)

radio occultation (RO) technique has shown promise inmeasuring the atmosphere with high vertical resolution andglobal coverage and under all weather conditions. TheConstellation Observing System for Meteorology, Iono-sphere, and Climate (COSMIC), composing of six satellites

in separate orbits, was successfully launched in April 2006,and provides about 2,500 RO profiles every day around theglobe [Anthes et al., 2008]. Atmospheric temperature andwater vapor profiles are derived from the COSMIC ROprofiles. In order to improve the retrieval algorithms and usethe COSMIC data to study various scientific problems, weneed to first evaluate the COSMIC‐retrieved water vaporprofiles to gain confidence on the data, especially over theocean, where upper‐air observations with high verticalresolution are missing.[7] In this article, we analyzed upper‐air observations

from dropsondes deployed from zero‐pressure balloons inthe stratosphere (NCAR’s Driftsonde system) over the PacificOcean during T‐PARC between 1 August and 30 September2008. Driftsondes were launched from Kona, Hawaii, driftedwith winds to cross the Pacific Ocean, deployed dropsondesalong the way, and provided good spatial coverage of sub-tropical Pacific Ocean. The aims of this paper are (1) to usethis unique, high‐quality and high vertical resolution data setto study water vapor distribution and variability in the region;(2) to use the reanalysis data with better spatial and temporalcoverage and sampling to test whether the findings from(1) are representative of general conditions in the region;and (3) in turn, to assess the performance of humidityprofiles from the reanalysis, COSMIC and Driftsonde byintercomparing them. The instruments and data are brieflydescribed in section 2. RH frequency distributions and lon-gitudinal variability are presented in section 3. In section 4,we discuss the problems in Driftsonde, reanalysis andCOSMIC data.

2. Instrument and Data

[8] T‐PARC was an international field project, conductedin the western Pacific from 1 August to 30 September 2008,to enhance our understanding in the mechanisms relevant toimproving prediction of weather events with high impactsand to provide data to examine typhoon genesis. During T‐PARC, ∼1900 dropsondes were dropped from four aircrafts(NRL‐P3, AF‐C130, DLR‐Falcon and DOTSTAR) and theNCAR Driftsonde system (see Figure 1 for sounding loca-tions). RH probability distributions for all soundings fromthe NRL‐P3, AF‐C130 and DOTSTAR in Figure 2 showthat the peak probability occurs at higher RHs (>70%)throughout the troposphere. This is mainly because thedropsondes from these three aircrafts were targeted fortropical cyclones and dropped in the vicinity of cyclones.The soundings from the DLR‐Falcon reveal bimodal dis-tributions of RH in the free troposphere (above 700 hPa)with one peak at RH less than 10% and the other one aroundice saturation RH. This is not surprising because the sondesfrom DLR‐Falcon were dropped in the edges of cyclones,the areas that the formation of cyclones was sensitive to, orthe extratropical cyclonic environment. Therefore, the air-craft soundings are not representative of all conditions andunsuitable for comparing with RH probability distributionspresented by previous studies. In contrast, Driftsondesoundings cover the subtropical Pacific from 140°E to155°W and sample all weather conditions (Figure 1).Therefore, this study mainly uses the soundings fromDriftsonde, including total 268 soundings dropped from

WANG ET AL.: WATER VAPOR IN SUBTROPICAL PACIFIC D21108D21108

2 of 14

16 balloons with an average drop altitude of 22.5 kmabove mean sea level.[9] The NCAR GPS dropsonde includes a pressure,

temperature, humidity sensor module, a code‐correlatingGPS receiver module for wind measurements and a 400MHz telemetry transmitter to transmit data from the sonde tothe onboard receiving system [see Hock and Franklin, 1999,Figure 1]. The dropsonde is deployed from research aircrafts

or other platforms over remote areas such as the ocean, polarregions and sparsely inhabited landmasses. It descendsthrough the atmosphere on a parachute measuring pressure,temperature, humidity and wind profiles at a 0.5 s verticalresolution (equivalent to ∼5–10 m). T‐PARC was the sec-ond deployment of Driftsonde system. Driftsonde wasdeveloped in an effort to produce a low‐cost measurementsystem capable of capturing vertical profiles of in situ

Figure 1. (top) Map with all dropsonde sounding locations from four aircrafts and Driftsonde system.(bottom) Driftsonde tracks (colored lines), dropsonde locations from Driftsonde (circles), COSMICsounding locations (triangles) and mean RH (percent) at 500 hPa from ERA‐Interim (background colorcontour with color bars on right side).


3 of 14

measurements in forecast sensitive regions, and filling crit-ical gaps in data coverage over remote locations. Driftsondesystem consists of a zero‐pressure or super‐pressure poly-ethylene balloon attached to a gondola that houses up to 50Miniature In situ Sounding Technology (MIST) dropsondes.The MIST dropsonde has the same sensors as the standarddropsonde, but it is smaller and lighter. The balloon floatsalong with the wind currents in the lower stratosphere orupper troposphere between 16 and 30 km, and can remainairborne for a minimum of six weeks (depending on thebattery life) for super‐pressure balloons. The MIST sondesare released upon command via the ground operationscenter. The zero‐pressure balloon has shorter life time thanthe super‐pressure one, and was used for T‐PARC. Sixteenballoons launched during T‐PARC stayed in the air for 3–6days. During the recent test of the driftsonde system withsuper‐pressure balloons, two fights had the lifetime of 105and 53 days. Sometimes the flight has to be terminatedbecause of political issues, flight permission, instrumentfailure and other reasons.[10] All dropsonde data have been carefully quality‐con-

trolled using several postprocessing methods includingapplying automatic sounding quality‐control software called

Atmospheric Sounding Processing Environment (ASPEN; seehttp://www.eol.ucar.edu/isf/facilities/software/aspen/aspen.html), visually examining each sounding and plotting thehistogram and time series of each parameter. Special problemsin data quality are identified and corrected if possible.[11] COSMIC includes six micro satellites launched into a

circular, 72° inclination orbit at an altitude of 512 km inApril 2006 [Anthes et al., 2008]. The COSMIC employs theGPS RO limb‐sounding technique to measure the atmo-sphere. The GPS‐RO receiver on each of COSMIC satellitesreceives the radio waves from GPS satellites. The measuredphase delay of radio waves is used to derive accurate andprecise vertical profiles of the bending angles of radio wavetrajectories. The refractivity profiles are then obtained fromthe bending angles. The refractivity depends on tempera-ture (T), pressure (P), water vapor pressure (Pw) andelectron density. The Pw profile is retrieved by using the1‐dimensional variational (1D‐Var) data assimilation tech-nique. The temperature and moisture gridded analysis fromECMWF forecasting model is used as the first guess of the1D‐Var technique; the refractivity profile is utilized toimprove the first guess field to obtain final retrieved Pwprofile. The COSMIC post processed data are used in this

Figure 2. RH probabilities (percent) for all soundings from NRL‐P3, AF‐C130, DLR‐Falcon,DOTSTAR, and Driftsonde. The RH interval is 4%. Solid and dashed black lines are means and standarddeviations, respectively. The red lines are ice saturation RHs.


4 of 14

study and include the COSMIC atmospheric profiles (referredas “wetPrf”) and ECMWF gridded analysis (referred as“ecmPrf”). The data are retrieved from the COSMIC DataAnalysis and Archive Center Version 3.0 (http://cosmic‐io.cosmic.ucar.edu/cdaac/index.html). The “wetPrf” data areavailable from surface to 40 km above mean sea level in a100 m resolution. The “ecmPrf” data are available fromsurface to about 1 hPa at 25 levels.

[12] Four global reanalysis products are used in this studyto characterize the water vapor distributions during Augustand September 2008, in the T‐PARC Driftsonde region (14–25°N, 140°E–155°W) and compare with Driftsonde andCOSMIC data. They, along with Driftsonde and COSMICdata, are summarized in detail in Table 1. One may noticethat the number of profiles during T‐PARC in the studyregion ranges from 268 (Driftsonde) to 129,320 (JRA).

Table 1. Characteristics of Six Data Sets Used in This Study

Driftsonde COSMIC ERA‐Interim JRA NCEP/DOE NCEP/NCAR

Spatial coverage targeted areas globe globe globe globe globeSpatial resolution point slant ∼1.5° × 1.5° ∼1.25° × 1.25° ∼2.5° × 2.5° ∼2.5° × 2.5°Vertical resolution 5–10 m 100 m 37 levels

(1000–1 hPa)12 levels(1000–100 hPa)

17 levels(1000–10 hPa)

8 levels(1000–300 hPa)

Temporal coverage irregular ∼2500/day 1989 to present 1979 to present 1979 to present 1948 to presentTemporal resolution irregular irregular 6 h 6 h 6 h 6 hNumber of profiles

in T‐PARC268 1625 96,624 129,320 39,528 39,528

Water vapor dataassimilated

‐ ECMWFanalysis

radiosonde,TOVS/ATOVS,SSM/I

radiosonde,TOVS/ATOVS,SSM/I

radiosonde,TOVS/ATOVS(only temperature)

radiosonde,TOVS/ATOVS(only temperature)

Figure 3. RH probabilities (percent) of Driftsonde data for the (right) eastern and (left) western Pacific.The RH interval is 4%. Solid and dashed black lines are RH means and standard deviations (percent),respectively. The dotted lines are ice saturation RHs. The red lines are mean temperature profiles (indegrees Celsius, top x axis). Mean and standard deviation of RH profiles in the WP (green line) arereplotted for the eastern Pacific.


5 of 14

Different humidity parameters are included in six data sets,RH for driftsonde, ERA‐Interim and two NCEP reanalyses,specific humidity for JRA and water vapor pressure forCOSMIC. For this study, they all are converted to RH withrespect to water or ice using Goff Gratch equations forsaturation vapor calculation over water or ice [SmithsonianInstitution, 1984].

3. RH Variability

3.1. Probability Distributions

[13] The RH probability distribution of Driftsonde datashows two distinct features below and above 700 hPa(Figure 2). The probability below 700 hPa maximizes at 70–90% RHs and has a narrow distribution. Around 700 hPa,there is a jump in locations of peak probability from ∼70%to less than 10% RH. Such discontinuous profile of RHprobability distribution was shown in the work of Sun andLindzen [1993, Figure 2]. There is a bimodal distributionin ∼700–600 hPa with one peak at ∼70% and one at RH lessthan 10%. This is consistent with the findings of Zhanget al. [2003] and Ryoo et al. [2009]. Above 600 hPa, thedistribution shows a primary peak at RH less than 10%and a secondary peak near ice saturation.[14] It is important to understand what causes the RH

bimodal distribution shown in Driftsonde data in Figure 2.Both the sampling and actual RH variability on the spatialand temporal scales can contribute to the bimodal distribu-tion. Mean ERA‐Interim RH at 500 hPa averaged for the2 month period in the background of Figure 1 (bottom) showsthe transition from RH below 20% in the eastern Pacific

(EP, east of 180°) to gradually increase of RH with longi-tude in the western Pacific (WP, west of 180°). Therefore,the data are first divided into the eastern and western Pacificregions (Figure 3). The EP and WP have similar distribu-tions below 700 hPa, but dramatic discrepancies above(Figure 3). The free troposphere is very dry in the EP with asingle peak of ∼1% RH and is drier in the EP than the WPby more than 20% on average, but has a broad distribution inthe WP with one peak near ice saturation and one varyingwith altitude. It suggests that the bimodal distribution above700 hPa seen in Figure 2 is primarily the result of mixing theEP and WP soundings together. Another thing worth noticeis that the top of the moist layer with a single peak at RH >70% is ∼700 hPa in the EP, but 600 hPa in the WP. There-fore, the following analyses are separated into the EP andWP. Such RH contrast between the EP and WP in the sub-tropics seems more profound in summer and fall than inwinter [see Peixoto and Oort, 1996, Figure 3]. Ryoo et al.[2009] show insignificant differences in subtropical Pacificupper troposphere humidity between the EP and WP in winterusing Atmospheric Infrared Sounder (AIRS) data. The pri-mary reason for the EP and WP RH distinction in the freetroposphere is the Walker circulation over the region, thesubsidence of dry air in the EP and the lifting of moist airfrom the boundary layer to the free troposphere in the WP,as shown by the pressure vertical velocity from the ERA‐Interim data (Figure 4). The spatial and temporal sampling ofDriftsonde data is not comprehensive (Figure 1). Thus, it iscrucial to examine whether the RH distributions in Figure 3are representative of general conditions (i.e., climatology)of the region. The global reanalysis data are used to test

Figure 4. Longitudinal variations of pressure vertical velocity (hPa/s) from ERA‐Interim.


6 of 14

this because of their better spatial and temporal coverageand sampling (see Table 1). However, we need to first testthe validity of the reanalysis moisture fields in the region.Figure 5 shows RH PDFs at three levels (900 hPa, 500hPa and 300 hPa) in the EP and WP from all six data setsusing all available data. It is apparent that the two NCEPreanalyses are outliers compared to the others. ERA‐Interim and JRA show similar distributions. Hence ERA‐Interim is used here. The RH PDFs at 900 hPa, 500 hPa and300 hPa are displayed in Figure 6 using all ERA‐Interim datain the region and during the T‐PARC period and the datamatched to Driftsonde locations and times along with thedriftsonde data for comparisons. There are no significantdifferences between all and matched soundings from theERA‐Interim in the shapes of PDFs. The PDF variationswith height and differences between the EP and WP arelarger than the differences between all and matched ERA‐Interim data. Qualitatively the sampling bias (differencesbetween all and matched ERA‐Interim) is smaller than thediscrepancies between matched ERA‐Interim and driftsonde.This indicates that the limited sampling of Driftsonde (total268) is enough to depict the RH PDF in the region. Figure

6 also shows that the matched‐data sample less dry events(RH < ∼40%) and more moist events at 300 hPa and 500hPa in the WP.

3.2. Zonal Variations

[15] The RH longitudinal variability from six data sets isshown more clearly in Figure 7. RH with respect to water isconverted to RH with respect to ice at temperatures below0°C (RHi), and RHi is presented in Figure 7 in order todepict ice supersaturation (ISS) more clearly. All dataexcept the two NCEP reanalyses show that the free tropo-sphere is very dry and has less variability in the EP, whereasRHi can vary from ∼20% to ∼100% in the WP. The limitedsampling in the matched ERA‐Interim data is capable ofshowing the increase of RH in the free troposphere from theEP to theWP and the longitudinal variability of RH in theWPwith a peak around 150°E (Figure 7). However, the matcheddata show drier free troposphere in the EP than all data andless smooth changes in the WP. Figure 7 suggests that thetrajectory of Driftsonde with easterly winds enables it tosample the RH longitudinal variability very well in spite of

Figure 5. RH probabilities (percent) at (top) 300 hPa, (middle) 500 hPa, and (bottom) 900 hPa in theeastern and western Pacific for six data sets. The RH interval is 4%. The vertical black dashed lines showmean ice saturation RHs. All available data are used here.


7 of 14

only 268 soundings, which is less than 0.3% of all 96624ERA‐Interim soundings.[16] Driftsonde data completely miss the high RHi

(close to ice saturation) layer at 200–100 hPa in the WPshown by ERA‐Interim, JRA and COSMIC, which isdiscussed in more detail in section 4. The RHi zonalvariations obtained from 1625 COSMIC soundings are inaccord with ERA‐Interim, especially the high RHi layer at200–100 hPa in the WP. However, caution has to betaken to interpret this as GPS‐RO’s capability to accuratelymeasure water vapor in the upper troposphere. This will beexplained in section 4. The ERA‐Interim and JRA showvery similar features although the JRA is drier than theERA‐Interim at west of ∼160°E in the free troposphere.The NCEP/DOE reanalysis shows very small RH longitu-dinal variation and fails to identify the EP/WP contrast.The NCEP/DOE significantly underestimates RH at 700–200 hPa in the WP, but overestimates it above 200 hPa inboth EP and WP. It presents unrealistically high ISS above200 hPa in the WP. The NCEP/NCAR is only availablebelow 300 hPa and shows the EP/WP contrast, but sig-nificantly underestimates RH below 700 hPa. The shortfallin NCEP reanalyses is also manifested in RH PDFs inFigure 5. We will come back to the poor performance ofNCEP reanalyses in section 4.[17] As shown in Figure 3, the free troposphere in the EP

has a single peak of ∼1% RH. Owing to the logarithmicdependence of OLR on humidity noted in the introduction,it makes a big difference whether RH is 1% or 2%. Spencer

and Braswell [1997, Figure 2] show that the sensitivity ofOLR to additive change of 3% RH could be doubled whenaveraged RH in the free troposphere is reduced from ∼6% to∼3%. Therefore, measurements of lower RHs (<20%) arepreferred to be accurate within 1% or better, have highhorizontal and vertical resolution, and be made by in situsensors rather than remote sensing. The latter two require-ments would minimize spatial and temporal averaging,which averages out extreme values. The historical radio-sonde data at lower RHs are often missing, are artificially setto a constant value, or contain significant biases. Forexample, before October 1993, the humidity data in the U.S.radiosonde network that used VIZ radiosonde were not re-ported at temperatures below −40°C, and the dew pointdepression was reported as a constant value of 30°C whenRH was below 20% [Elliott and Gaffen, 1991]. Afterward,RHs were reported at all temperatures and as measuredvalues, but contain significant moist bias [Spencer andBraswell, 1997; Wang et al., 2001]. All satellite datainclude some kind of averaging spatially, and thus overes-timate RHs at lower RH. In contrast, the driftsonde dataused in this study can meet the requirements for better RHmeasurements in dry regions as a result of the high accuracy(2% in RH) and quick response of Vaisala H‐HUMICAPthin film capacitor used in the MIST sonde, its in situmeasurement technique and its high vertical resolution. Thepercentage of RHs equal to and less than 5% at 500 hPa inthe EP is 59%, 48% and 28% for the driftsonde data, andmatched and all ERA‐Interim data, respectively. For the

Figure 6. RH probabilities (percent) at (top) 300 hPa, (middle) 500 hPa, and (bottom) 900 hPa in theeastern and western Pacific for all ERA‐Interim (solid black line) that matched at Driftsonde locationsand times (dashed black line) and Driftsonde (solid gray line). The RH interval is 4%.


8 of 14

driftsonde data, nineteen percent of cases have RHs at 1%.Such extreme dryness (RH < 5%) of the free troposphere inthe tropics has been reported by Spencer and Braswell[1997] and Sherwood et al. [2006] on the basis of micro-wave and GPS‐RO satellite data, respectively. The latterconcluded that one‐quarter of the tropical air at 8 km isbelow 4% RH. However, the AIRS data only report ∼0.5%of samplings in the tropics with RHs less than 5% [Ryoo etal., 2009, Figure 5a]. The higher population of extreme dry

cases in the driftsonde data is likely because the driftsondemeasurement is instantaneous at a point, while all othersinclude averaging either spatially or temporally.

4. Problems in Driftsonde, Reanalysis,and COSMIC Humidity Data

[18] RH distributions and variability presented in section4 from Driftsonde, reanalysis and COSMIC data reveal

Figure 7. Longitudinal variations of RHi averaged for 14–25°N for Driftsonde, matched ERA‐Interim,JRA, all ERA‐Interim, COSMIC, COSMIC first guess (“COSMIC ecm”), NCEP/DOE, and NCEP/NCAR data.


9 of 14

problems in them. In summary, the following serious pro-blems are identified:[19] 1. Driftsonde completely misses the moist layer with

RHi > 70% in the WP above ∼200 hPa.[20] 2. Both NCEP reanalyses are incapable of correctly

simulating RH magnitudes and variability in the subtropicalPacific.[21] 3. The good performance of COSMIC above 200 hPa

is questionable.[22] These three problems are investigated carefully below.[23] As shown in Figure 7, Driftsonde hygrometers fail to

capture the large RHi (>70%) layer above 200 hPa dis-played in the ERA‐Interim, JRA and COSMIC data. Thematched Driftsonde and ERA‐Interim data show that in the

WP above 200 hPa Driftsonde RH has a constant value of1%, which is the lowest limit the MIST hygrometer canmeasure, while the ERA‐Interim contains a layer of highRHi (>70%) (not shown). Twelve percent of ERA‐Interimdata points above 200 hPa in the WP have RH above icesaturation. This is inconsistent with Luo et al. [2008], whofound 100% RHi cutoff in the ECMWF analysis. This canbe explained by the change of the ISS scheme around 2006[Tompkins et al., 2007]. RHi in the old scheme was notallowed to exceed 100%, whereas the new scheme allowsISS [Tompkins et al., 2007]. Figure 8 (top) provides twocases where the ERA‐Interim shows moist layers at orabove 200 hPa. Driftsonde RHi values started to increasebelow 150 hPa and reached a local maximum at lower al-

Figure 8. RHi profiles (percent) for four matched Driftsonde (solid lines) and ERA‐Interim (dashedlines) soundings.


10 of 14

titudes. It suggests that the Driftsonde hygrometer is able torespond to humidity changes in the upper troposphere, buthas time lags and a dry bias. This is due to the slowerresponse of the Humicap used as the MIST hygrometer atcold temperatures and the fact that the sonde descends fromcolder to warmer environments. The Humicap in Driftsondeis the same as that used for Vaisala RS92 radiosonde. Pre-vious studies have shown the time lag error, dry bias andother errors in the Humicap [e.g., Wang et al., 2002; Voemelet al., 2007; Miloshevich et al., 2009]. Figure 8 (bottom)uses two examples to illustrate the advantages of high ver-tical resolution of Driftsonde data, including showing much

more detailed vertical structures of RH and better describingthe strong gradients of RHs around 700 hPa. On average,Driftsonde data agree with ERA‐Interim and JRA within±5% in RH below 500 hPa (Figure 9).[24] The two NCEP reanalyses are incapable of producing

accurate RH values and its vertical and horizontal variations(Figures 5, 7, and 10) over the ocean. At 500 hPa, thecorrelations between matched Driftsonde and the two NCEPreanalyses (NCEP/NCAR and NCEP/DOE) are 0.53 and0.26, respectively. In comparison, the correlations for ERA‐Interim and JRA are both 0.92. The disagreements betweenNCEP reanalyses and other data are also manifested by the

Figure 9. Mean and root mean square (RMS) profiles of RH differences (between four reanalysis pro-ducts and Driftsonde (percent)).

Figure 10. Spatial distributions of mean RH (percent) at (left) 300 hPa and (right) 500 hPa from fourreanalyses.


11 of 14

large mean differences and root mean square of the differ-ences in RH profiles (Figure 9). Trenberth et al. [2005] andTrenberth and Guillemot [1998] also demonstrated poorperformance of both NCEP products over the ocean. Theyfound that over the region studied here, NCEP reanalysessignificantly underestimate total column water vapor(TCWV) values and its temporal variability compared toSSM/I and ERA‐40 data. This is supported by smaller RHvalues below 500 hPa in the two NCEP data sets (Figure 9).The humidity above 500 hPa only contributes to less than3% of TCWV [Wang et al., 2007]. The deficiency of NCEPreanalyses in water vapor over the ocean is most likelybecause NCEP reanalyses did not assimilate satellite watervapor soundings and had no in situ soundings over theocean to constrain the data assimilation, except for a fewisland radiosonde observations [Kistler et al., 2001]. Figure9 also shows that at ∼450 hPa, NCEP/NCAR changes froma dry bias to a moist bias relative to Driftsonde data. Thisfeature is expressed in RH PDFs in Figure 5. This is con-sistent with the finding of >30% higher NCEP/NCARhumidity than the COSMIC data by Chou et al. [2009].Kanamitsu et al. [2002] concluded that the NCEP/DOE RHis lower than NCEP/NCAR by 10–15% at 300 hPa in thetropics and is suspected to be higher in the lower tropo-sphere since TCWV from both NCEP are about the same.We reach the same conclusions from Figures 5 and 9. It isnot clear what causes the unrealistically large ISS above200 hPa in NCEP/DOE (Figure 7).[25] It is very encouraging to see COSMIC to capture the

higher‐moisture layer at 200–100 hPa in the WP, but onehas to be cautious to simply interpret the COSMIC retrievedRH as measurements. The GPS RO measures the bendingangle of the GPS radio signal, which can be converted to therefractivity (N). N is a function of atmospheric temperatureand vapor pressure (Pw) (N = 77.6*P/T + 3.73 × 105*Pw/T

2,where P is pressure). To resolve the ambiguity of N asso-ciated with T and Pw, a 1D‐Var analysis model is used to

derive optimal T and Pw profiles. The temperature and watervapor profiles from the ECMWF forecast model interpolatedto the COSMIC sounding time and location are used as thefirst guess in the 1D‐Var algorithm. The retrieval ofhumidity in the upper troposphere from COSMIC is verydifficult because the contribution of the wet term (3.73 ×105*Pw/T

2) to total N is very small (<1%) at high altitudes.Therefore, in order to retrieve Pw accurately in the uppertroposphere, the requirement on the accuracy of N is strin-gent. The 1D‐Var uses N profiles to improve the first‐guessT and Pw profiles, but has difficulty in improving the first‐guess Pw profile in the upper troposphere. Figure 6 alsoshows the RH longitudinal variations of the ECMWF first‐guess data, and Figure 11 gives two examples. RH at 200–100 hPa in the COSMIC data is very closer to the first‐guessfield, although they can be quite different at lower eleva-tions. This suggests that the feature of the moist layer above200 hPa in the COSMIC data comes from the first guessrather than the refractivity measurement.

5. Conclusions

[26] High quality, high vertical resolution upper airobservations over open oceans have historically been scarcebecause of the limited availability of in situ data and becauseavailable satellite data is typically low resolution, and ofquestionable quality. During T‐PARC from 1 August to30 September 2008, ∼1900 soundings were collected overthe Pacific Ocean. These include dropsondes deployed fromfour aircrafts and zero‐pressure balloons in the stratosphere.The sounding data include profiles of pressure, temperature,humidity, wind speed and direction from the ocean surfaceto the troposphere for dropsondes, and from the surface to∼50 hPa for Driftsondes. The data are 0.5 s vertical reso-lution, corresponding to ∼10 m near surface and ∼30 m at50 hPa. All soundings underwent a series of vigorousquality control at NCAR. Such an unprecedented sounding

Figure 11. RHi (percent) profiles for two pairs of COSMIC retrieved (solid lines) and first‐guess(dashed lines) soundings.


12 of 14

data set is valuable for achieving scientific objectives ofT‐PARC that address various aspects of typhoon activityincluding examining typhoon genesis and better under-standing of the mechanisms that lead to improved pre-dictive skill of typhoon. However, this study has a generalgoal of exploring the scientific applications of this uniquedata set beyond T‐PARC’s specific objectives. Specifically,it focuses on making use of this unique data set to studydetailed water vapor variability over the Pacific Ocean andevaluate global reanalysis products and the COSMIC data inthe region.[27] The 268 Driftsonde soundings are analyzed in detail to

document RH probability distributions and vertical and lon-gitudinal variability over the subtropical Pacific region (14–25°N, 140°E–155°W). Unfortunately, the aircraft soundingswere targeted for specific weather events, and thus are notrepresentative of common climatology of the area. Theanalysis shows the distinct contrast between the Eastern andWestern Pacific in RH PDFs in the free troposphere. The EPis very dry with a single peak of 1% RH and is drier in the EPthan the WP by more than 20% RH on average. The fre-quency of RHs at 1% and less than or equal to 5% is 19% and59% at 500 hPa in the EP, respectively. Such frequentoccurrences of extreme dryness in the free troposphere havesignificant radiative impacts and call for better sampling ofthem in the future. In contrast, RH in the free troposphere inthe WP has a broad distribution with one peak near ice sat-uration and one varying with altitudes. Such distinctionbetween EP and WP is attributed to the Walker circulationover the region. The RH longitudinal variability is studiedand compared using Driftsonde, COSMIC and fourreanalysis data sets. The free troposphere is very dry andhas less variability in the EP, whereas RH with respect to icein the WP can vary from 20% to 100% with a consistentlymoist layer above 200 hPa. It is found that the limited sam-pling of Driftsonde (total 268) is sufficient to adequatelydescribe RH PDFs and its longitudinal variability in theregion because the trajectory of Driftsonde with easterlywinds enables it spatially sample the area well.[28] The intercomparison of RH distributions and varia-

tions using six data sets from three sources shows that theERA‐Interim and JRA perform very well. But it also revealsproblems in the data sets. Driftsonde hygrometer is inca-pable of measuring larger RHi (>70%) in the WP above200 hPa because of its slower response at colder tem-peratures. Both NCEP reanalyses fail to produce the cor-rect magnitudes and variability of RH in the subtropicalPacific as a result of not assimilating satellite water vaporsoundings and lack of in situ observations. Overall, theCOSMIC water vapor data show good performancealthough one has to be cautious on simply interpreting theCOSMIC retrieved RHs as true measurements. We proposethat the higher RHs at 200–100 hPa in the WP in theCOSMIC data are ascribed to the ECMWF first‐guess datarather than the refractivity measurements.[29] The NCAR Driftsonde system has been demonstrated

here to be a valuable, cost‐effective observing system to fillcritical gaps over oceanic and remote polar and continentalregions during periods of days and weeks. Its drifting naturewith winds and its long duration provide good spatialsampling and coverage. The high‐quality and high verticalresolution temperature, pressure, humidity and wind profiles

measured by the dropsonde enable detailed thermodynami-cal and kinematical profiles of the atmosphere, especiallyover remote and less well observed areas. This paper showsone example, an in‐depth study of RH probability dis-tributions and its vertical and horizontal variability in sub-tropical Ocean. In addition, Driftsonde data are essential tovalidate models, reanalyses and satellite observations, espe-cially over data‐sparse areas. We have shown the short-comings of the two NCEP reanalyses over the ocean, on thebasis of comparisons with Driftsonde data, and illustrated thestrong need for in situ upper‐air observations over the oceanto constrain global reanalysis. The limitation of the driftsondehygrometer in measuring high RHs in 200–100 hPa layerpointed out in section 3 calls for future development of betterdriftsonde hygrometer capable of measuring upper tropo-sphere and even lower stratosphere humidity. In addition,efforts could be made in the future to maximize the usage ofthe unique driftsonde platform by deploying other sensors tomeasure ozone and other trace gases. In summary, Driftsondesystem has the potential to become a valuable, cost‐effectiveoperational system over the ocean for upper air observations.

[30] Acknowledgments. We are grateful to all NCAR/EOL staffthat developed the Driftsonde system, and to the people who participatedin T‐PARC to deploy and operate the system and to collect the data.J. Wang would like to thank Rick Anthes and Tae‐Kwon Wee, UCAR,for useful discussions on the COSMIC retrieval algorithm. Commentsfrom Tom Horst, Jothiram Vivekanandan, and three anonymous reviewershave been very helpful. The work of P.‐H. Lin and C.‐H. Weng was sup-ported by the National Science Council, Taiwan. The NCAR is sponsoredby the National Science Foundation.

ReferencesAnthes, R. A., et al. (2008), The COSMOC/FORMOSAT‐3—Missionearly results, Bull. Am. Meteorol. Soc., 89, 313–333, doi:10.1175/BAMS-89-3-313.

Brown, R. G., and C. D. Zhang (1997), Variability of midtroposphericmoisture and its effect on cloud‐top height distribution during TOGACOARE, J. Atmos. Sci., 54, 2760–2774, doi:10.1175/1520-0469(1997)054<2760:VOMMAI>2.0.CO;2.

Chou,M.‐D., C.‐H.Weng, and P.‐H. Lin (2009), Analyses of FORMOSAT‐3/COSMIC humidity retrievals and comparisons with AIRS retrievals andNCEP/NCAR reanalyses, J. Geophys. Res., 114, D00G03, doi:10.1029/2008JD010227 [printed 115(D2), 2010].

Elliott, W. P., and D. J. Gaffen (1991), On the utility of radiosonde humidityarchives for climate studies, Bull. Am. Meteorol. Soc., 72, 1507–1520,doi:10.1175/1520-0477(1991)072<1507:OTUORH>2.0.CO;2.

Galewsky, J., A. Sobel, and I. Held (2005), Diagnosis of subtropicalhumidity dynamics using tracers of last saturation, J. Atmos. Sci., 62,3353–3367, doi:10.1175/JAS3533.1.

Held, I. M., and B. J. Soden (2000), Water vapor feedback and globalwarming,Annu. Rev. Energy Environ., 25, 441–475, doi:10.1146/annurev.energy.25.1.441.

Hock, T. F., and J. L. Franklin (1999), The NCAR GPS dropwindsonde,Bull. Am. Meteorol. Soc., 80, 407–420, doi:10.1175/1520-0477(1999)080<0407:TNGD>2.0.CO;2.

Kanamitsu, M., W. Ebisuzaki, J. Woollen, S. K. Yang, J. J. Hnilo,M. Fiorino, and G. L. Potter (2002), NCEP‐DOE AMIP‐II Reanalysis(R‐2), Bull. Am. Meteorol. Soc., 83, 1631–1643, doi:10.1175/BAMS-83-11-1631(2002)083<1631:NAR>2.3.CO;2.

Kistler, R., et al. (2001), The NCEP‐NCAR 50‐year reanalysis: Monthlymeans CD‐ROM and documentation, Bull. Am. Meteorol. Soc., 82,247–267, doi:10.1175/1520-0477(2001)082<0247:TNNYRM>2.3.CO;2.

Luo, Z. Z., D. Kley, R. H. Johnson, and H. Smit (2008), Ten years of mea-surements of tropical upper‐tropospheric water vapor by MOZAIC. partII: Assessing the ECMWF humidity analysis, J. Clim., 21, 1449–1466,doi:10.1175/2007JCLI1887.1.

Miloshevich, L. M., H. Voemel, D. N. Whiteman, and T. Leblanc (2009),Accuracy assessment and correction of Vaisala RS92 radiosonde water


13 of 14

vapor measurements, J. Geophys. Res., 114, D11305, doi:10.1029/2008JD011565.

Peixoto, J. P., and A. H. Oort (1996), The climatology of relative humidityin the atmosphere, J. Clim., 9, 3443–3463, doi:10.1175/1520-0442(1996)009<3443:TCORHI>2.0.CO;2.

Pierrehumbert, R. T. (1998), Lateral mixing as a source of subtropical watervapor, Geophys. Res. Lett., 25(2), 151–154, doi:10.1029/97GL03563.

Pierrehumbert, R. T., and R. Roca (1998), Evidence for control of Atlan-tic subtropical humidity by large‐scale advection, Geophys. Res. Lett.,25(24), 4537–4540, doi:10.1029/1998GL900203.

Pierrehumbert, R. T., H. Brogniez, and R. Roca (2006), On the relativehumidity of the atmosphere, in The Global Circulation of the Atmosphere,edited by T. Schneider, A. H. Sobel, and E. N. Lorenz, pp. 143–185,Princeton Univ. Press, Princeton, N. J.

Ryoo, J. M., T. Igusa, and D. W. Waugh (2009), PDFs of tropical tropo-spheric humidity: Measurements and theory, J. Clim., 22, 3357–3373,doi:10.1175/2008JCLI2747.1.

Salathé, E. P., and D. L. Hartmann (1997), A trajectory analysis of tropicalupper‐tropospheric moisture and convection, J. Clim., 10, 2533–2547,doi:10.1175/1520-0442(1997)010<2533:ATAOTU>2.0.CO;2.

Sherwood, S. C., E. R. Kursinski, and W. G. Read (2006), A distributionlaw for free‐tropospheric relative humidity, J. Clim., 19, 6267–6277,doi:10.1175/JCLI3978.1.

Shine, K. P., and A. Sinha (1991), Sensitivity of the Earth’s climate toheight‐dependent changes in the water‐vapor mixing‐ratio, Nature,354, 382–384, doi:10.1038/354382a0.

Smithsonian Institution (1984), Smithsonian Meteorological Tables,5th ed., 350 pp., Washington, D. C.

Soden, B. J. (1998), Tracking upper tropospheric water vapor radiances: Asatellite perspective, J. Geophys. Res., 103(D14), 17,069–17,081,doi:10.1029/98JD01151.

Spencer, R. W., and W. D. Braswell (1997), How dry is the tropical free tro-posphere? Implications for global warming theory, Bull. Am. Meteorol.Soc., 78, 1097–1106, doi:10.1175/1520-0477(1997)078<1097:HDITTF>2.0.CO;2.

Sun, D. Z., and R. S. Lindzen (1993), Distribution of tropical troposphericwater‐vapor, J. Atmos. Sci., 50, 1643–1660, doi:10.1175/1520-0469(1993)050<1643:DOTTWV>2.0.CO;2.

Tompkins, A. M., K. Gierens, and G. Radel (2007), Ice supersaturation inthe ECMWF integrated forecast system, Q. J. R. Meteorol. Soc., 133,53–63, doi:10.1002/qj.14.

Trenberth, K. E., and C. J. Guillemot (1998), Evaluation of the atmosphericmoisture and hydrological cycle in the NCEP/NCAR reanalyses, Clim.Dyn., 14, 213–231, doi:10.1007/s003820050219.

Trenberth, K. E., J. Fasullo, and L. Smith (2005), Trends and variability incolumn‐integrated atmospheric water vapor, Clim. Dyn., 24, 741–758,doi:10.1007/s00382-005-0017-4.

Voemel, H., H. Selkirk, L. Miloshevich, J. Valverde‐Canossa, J. Valdes,E. Kyro, R. Kivi, W. Stolz, G. Peng, and J. A. Diaz (2007), Radiation drybias of the Vaisala RS92 humidity sensor, J. Atmos. Oceanic Technol., 24,953–963, doi:10.1175/JTECH2019.1.

Wang, J., H. L. Cole, and D. J. Carlson (2001), Water vapor variability inthe tropical western Pacific from a 20‐year radiosonde data, Adv. Atmos.Sci., 18, 752–766.

Wang, J. H., H. L. Cole, D. J. Carlson, E. R.Miller, K. Beierle, A. Paukkunen,and T. K. Laine (2002), Corrections of humidity measurement errorsfrom the Vaisala RS80 radiosonde—Application to TOGA COARE data,J. Atmos. Oceanic Technol., 19, 981–1002, doi:10.1175/1520-0426(2002)019<0981:COHMEF>2.0.CO;2.

Wang, J. H., L. Y. Zhang, A. Dai, T. Van Hove, and J. Van Baelen (2007),A near‐global, 2‐hourly data set of atmospheric precipitable water fromground‐based GPS measurements, J. Geophys. Res., 112, D11107,doi:10.1029/2006JD007529.

Zhang, C. D., B. E. Mapes, and B. J. Soden (2003), Bimodality in trop-ical water vapour, Q. J. R. Meteorol. Soc., 129, 2847–2866, doi:10.1256/qj.02.166.

M. Bradford, H. Cole, J. Fox, T. Hock, D. Lauritsen, P.‐H. Lin, S. Loehrer,C. Martin, J. VanAndel, J. Wang, K. Young, and L. Zhang, Earth ObservingLaboratory, National Center for Atmospheric Research, P.O. Box 3000,Boulder, CO 80307, USA. ([email protected])C.‐H. Weng, Department of Atmospheric Sciences, National Taiwan

University, No. 1, Sec. 4, Roosevelt Rd., Taipei, 10617 Taiwan.


14 of 14

water vapor variability and comparisons in the subtropical...

Documents