Latest Water Vapor Results from GPS Radio Occultation (RO) + Next Gen
E. R. Kursinski
A.Kursinski1, T. Gebhardt2, C. Ao3, D. Ward4, A. Otarola5, H. Reed2, D. Erickson2, and J. McGhee1
Special thanks to Mark Ringer (Met Office)1 MOOG Advanced Missions & Science, Golden, Colorado, USA
2 University of Colorado, Boulder, Colorado, USA3 Jet Propulsion Laboratory, Pasadena, California, USA
4 University of Arizona, Tucson, Arizona, USA5 Thirty Meter Telescope, Pasadena, California, USA
GVAP Oct 1, 2013 CIRA-CSU
Outline
• Quick Background on GPS RO• GPS Water Vapor Histograms1
– Comparisons with AIRS, analyses, GCMs• Cluster analysis of GPS WV1
• Next Gen RO: ATOMMS Overview2
1Funded by NOAA & NASA2Funded by NSF and Moog Broad Reach
Kursinski et al., 2 GVAP MOOG Oct 1, 2013
JCSDAAugust 19, 2009
• Delay(t)=> bending angle(a) => refractivity(z) where a=nr– Dry conditions: => dry density(z) => P(z) => T(z) via hydrostatic eqn– Wet conditions: refractivity + T,p,q (analysis) => better T,p,q
or refractivity + T (analysis) => water vapor(z)
GPS Occultation Summary
occultation geometry
An occultation occurs when the orbital motion of a GPS SV and a Low Earth Orbiter (LEO) causes the LEO ‘sees’ the GPS rise or set across the limb
This causes the signal path between the GPS and the LEO to slice through the atmosphere
Atmosphere acts as a lens bending the signal path
1D forward relation 1D inverse relation
JCSDA
Information vs. Altitude from GPS RO
Density, pressure & temperature vs. alt.
May 4, 2010
Ionospheric refractivity & Free electron density
H2O vapor
Upper altitude depends on solar & diurnal
cycles
Atmospheric Refractivity
N a1
P
T a2
Pw
T 2 40.3106 ne
f 2
200 m vertical resolution
All-weather
Density, pressure & temperature vs. alt.
MOOG Oct 1, 2013
GPS RO Features Summary • Least biased data set available?
– Global coverage – Diurnal coverage with > 6 satellite constellation like COSMIC– Works in clear and cloudy conditions ( l ~20 cm)– Works over land and water– Unique relation between bending angle & refractivity (except super-N)
insensitive to initial guess• Resolution
– Vertical resolution ~200 m• Capable of seeing stability related effects invisible to other sounders
– Horizontal resolution• Along track horizontal resolution ~ 100-250 km• Cross track ~ 1.5 km (limited by horizontal motion of raypath)• Inherent averaging good for climate (better horiz. res. desired for NWP)
• Vertical range– Useful to ~240 K level in troposphere (~9 km alt. in tropics)– Extends down very close to surface in extratropics– If we can deal with super-refraction, lower altitude can be the
surface in the tropics
Kursinski et al., 5 GVAPMOOG Oct 1, 2013
Zonal Mean Relative Humidity GPS-MET Jun 21-Jul 4 1995
• Zonal mean relative humidity from GPS/MET July 1995
Kursinski & Hajj, 2001
Winter Summer
ITCZ subtropicssubtropics
~2,000 profiles
GPS RO Missions Mission Period Occ/day• GPS-MET 1995-1997• CHAMP 2001-2010 250• COSMIC 2006- 1500-3000• METOP-A 2010- 700• Megha-Tropiques 2011- 500?• METOP-B 2012- 700• KOMPSAT-5 2013- 500?
Kursinski et al., 7 GVAP MOOG Oct 1, 2013
Present & Future GNSS RO Missions• From www.wmo-sat.info/oscar/
Kursinski et al., 8 GVAP MOOG Oct 1, 2013
Kursinski et al., 9 GVAP
400 GNSS Occulting Satellites over 3 HOURS, World
• Recently begun study at Moog AMS
MOOG Oct 1, 2013
Low Latitude Moisture Study• Convection creates wet & dry extremes => stretches water vapor distribution• Mixing and diffusion work to homogenize the distribution toward the center• Specific humidity is conserved in the absence of sources and sinks => tracer• Relative humidity important for conversion between vapor & condensed
phases => clouds & precipitation• Reducing forecast uncertainty => reducing uncertainty about processes =>
precise, hi-res all-weather obs that capture variability & constrain processes3000 km
16 km
Zonal Mean Relative Humidity GPS-MET Jun 21-Jul 4 1995
• Zonal mean relative humidity from GPS/MET July 1995
Kursinski & Hajj, 2001
Winter Summer
ITCZ subtropicssubtropics
~2,000 profiles
SR
Free Tropospheric PW fromCOSMIC
Oct 06
Jan 07
Apr 07
Jul 07
Kursinski et al., 12 AGU Dec 17, 2010
Moisture Histograms• Low order moments like mean and variance
provide limited insight into water vapor distribution and the hydrological cycle
• Histograms of moisture on individual pressure levels yield much better indication of full range of behavior
• Plus insight into processes at work and adequacy of their representation in models
Kursinski et al., 13 GVAP MOOG Oct 1, 2013
Two Methods for Extracting Water Vapor from GPS RO Refractivity Profiles
• Direct Method: Nwet = Ntot – Ndry
– Determine dry refractivity (Ndry) from analysis temperature profile and hydrostatic equation
– Scale Nwet to get water vapor
• (1D) Variational Method– Combine GPS refractivity with temperature & water vapor
profiles and surface pressure from analysis and error covariance estimates
– Overdetermined, least squares solution
• Advantage of Direct Method: Not affected by biases in background water vapor forecast/analysis
Kursinski et al., 14 GVAP MOOG Oct 1, 2013
Negative q and Error Deconvolution
Direct method can and does produce negative q estimates
=> Produces an unphysical, negative tail in the q histograms
• This can be fixed by deconvolving the error distribution from histograms– Linearize error model: qmeasured = qtrue + eq
– Measured histogram (PDF) is then the convolution of the true PDF and the error PDF
PDFqmeas = PDFqtrue PDFe
• IF we understand the error PDF, we can then deconvolve it from the measured PDF to recover the true PDF
– Negative tail tells us shape of the error distribution
Kursinski et al., 15 GVAP MOOG Oct 1, 2013
Kursinski et al., 16 GVAP MOOG Oct 1, 2013
Automated, Error Deconvolution Low Latitude
-0.5 -0.25 0 0.25 0.5 0.75 1 1.25 1.5 1.75 20
0.05
0.1
0.15
0.2
0.25
0.3
Water vapor (g/kg)
Pro
ba
bilit
y
Histogram of water vapor levels from346mb.xlsx with25000 iterations
raw measurementsinitial truth guessupdated truth guesserror pdf w/ std dev0.14242
346 hPa
Error PDF
Deconvolved distribution
Measured distribution
• Adjust (1) (symmetric) Error PDF & (2) “true” q distribution PDF, • Convolve them to generate estimate of “measured” PDF,• Iterate adjustment until best fit to the measured PDF is achieved
Full Annual Cycle (2007)
Estimating the Accuracy of GPS-derived Water Vapor
Analogously, the error in relative humidity, U, is
where L is the latent heat and Bs = a1TP / a2es.
sq ~ 0.2 g/kg in mid & upper troposphere.
sq ~ 0.5 g/kg in lower troposphere
2/1
2
22
2
22
2
22 )2(
PB
TTR
LUB
NUB P
sT
vs
NsU
• Kursinski et al. 1995: Initial estimate of GPS water profile accuracy • Kursinski & Hajj, 2001: Error in specific humidity, q, due to errors in refractivity, N,
temperature, T, and pressure, P, from GPS
where C = a1Tmw/a2md ~ 35 g/kg
𝜎 𝑞=((𝐶+𝑞)2(𝜎𝑁
𝑁 )2
+(𝐶+2𝑞 )2(𝜎𝑇
𝑇 )2
+(𝐶+𝑞)2(𝜎 𝑃 𝑠
𝑃𝑠)
2
)1 /2
Kursinski et al., 17 GVAP Oct 1, 2013
Separating the Errors• Estimate water vapor error from negative tail of distribution• Resulting errors somewhat smaller than predictions of Kursinski &
Hajj, 2001• Presumably because analysis temperature errors are smaller
Specific
Humidity Error (g/kg)
Fractional Refractivity Error
(%)Temperature
Error (K)Reference
Pressure Error (%)
Pressure level (hPa) KH01 Error
deconv KH01 Error deconv KH01 Error
deconv KH01 Error deconv
346 0.24 0.145 0.2 0.2 1.5K 0.9K 0.3% 0.15%
547 0.31 0.25 0.5 0.6 1.5K 0.9K 0.3% 0.15%
725 0.47 0.39 0.9 1 1.5K 0.9K 0.3% 0.15%
• Dessler & Minschwaner 2007 compared advection-saturation model results & AIRS• PROBLEM: Model matches AIRS but neither looks like GPS. What’s going on?
725
Motivation
You are here
547 hPa Specific Humidity Comparisons• c
Analyses don’t like really dry air
Analyses & AIRS underestimate very high humidity air
GPS & AIRS agree well on very dry end in mid-troposph
Overestimates of mid-humidity air
~5 km altitude
Kursinski et al., 20 GVAP MOOG Oct 1, 2013
Comparison with CMIP3 & CMIP5 models• Noticeable improvement in some CMIP5 models:
NCAR and MPI in particular• Resolution contributing but definitely not the
whole answer (MIROC 5 has highest resolution)CMIP3 CMIP5
547 hPa 547 hPa
Kursinski et al., 21 GVAP MOOG Oct 1, 2013
Comparison with Advection- Saturation Model
• Model from Dessler & Minschwaner, 2007– Start with moisture in air parcel observed by AIRS– Advect parcel according to NCEP wind analyses– Limit mixing ratio to the saturation mixing encountered along trajectory
• Produces more very dry and wet air than GPS RO observes– Model’s peak in wet end is not observed
• Lack of mixing in adv-sat model likely explains model’s higher extremes than observed
547 mb
Kursinski et al., 22 GVAP MOOG Oct 1, 2013
GPS v. AIRS: 725 hPa Specific Humidity
AIRS underestimates low & high ends of distribution
IR can’t see thru clouds
AIRS doesn’t see the very dry air likely because IR vertical resolution can’t separate PBL from free troposphere
Very dry air (<5% RH) is reaching the PBL top
Kursinski et al., 23 GVAP MOOG Oct 1, 2013
725• compare
You are here
CMIP5 GCMs vs. GPS RO vs. AIRS
Too much vertical diffusion in GCMsWV feedback likely too positive in GCMs
GCMs overestimating
% of wet air
AIRS-GPSRO Histogram Comparison Summary
• AIRS misses high end of moisture distribution– Presumably due to clouds
• Cloud clearing does not eliminate fundamental IR limitation• Extremely dry air
– AIRS & GPSRO show similar % of dry air in mid-troposphere, much more than in GCMs and analyses
– GPS RO sees far more just above PBL than AIRS, • Likely associated with limited AIRS vertical resolution.
• GPS’ cloud penetration & 200 m vert. res. more important than its 100-250 km horizontal resolution
• What is the horizontal resolution of GPS RO, 250 km, 100 km ?
• These differences are important for inferring processesKursinski et al., 25 GVAP MOOG Oct 1, 2013
Error Pattern of Analyses & GCMs• Too little of the very dry air in mid & lower tropos
Þ too much vertical diffusion in modelsÞ Suggests WV feedback too strongly positive in free tropos
• Not enough high q & RH distributionsÞ “trigger happy” convection parameterizations– IF GCM cloud & rainfall are reasonable => must be
(erroneous) compensating cloud and precipitation parameterizations
– May be related to why GCMs under-predict observed increases in heavy rainfall
Kursinski et al., 26 GVAP MOOG Oct 1, 2013
GCM AssessmentGCMs improving• Resolution is contributing to model improvement
– BUT NCEP analysis res: 1o; MIROC 5 highest resolution GCM but definitely not the best GCM
• NCAR & MPI CMIP5 models agree best with GPS RO moisture distribution
Diffusion • Some diffusion necessary to reproduce what GPS RO
observes• Compressed moisture distribution indicates too much
mixing in analyses and (older) GCMs
Monthly means are inadequate for making this assessmentKursinski et al., 27 GVAP MOOG Oct 1, 2013
Causes of Limited NWP Moisture Impact?
1. Mismatch of GPS RO & NWP GCM moisture climatologies, particularly at wet and dry extremes– Moisture assimilation not utilizing observations well (particularly
mid-trop) (Analyses look like free running CMIP3 GCMs)– Sensitivity of moisture analyses to initial guess limits their utility
for assessing realism of GCM hydrological cycle IMPORTANT:– Dry part is important for suppressing convection– Assimilation of observed wet air => rain outÞ Impact cannot improve until GCMs better represent real
moisture distribution
Kursinski et al., 28 GVAP MOOG Oct 1, 2013
Causes of Limited NWP Moisture Impact?
GOOD NEWS: GCMs are getting better• AR5 models: CESM 1.0 & MPI-ESM-LR look significantly better• Good for both climate and NWP
Þ Suggests assimilation experiments with GCMs with matching climatologiesÞ Good to have same NWP & climate GCMs like Met Office
2. Relatively sparse GPS RO coverage • Water varies rapidly on short temporal spatial scales (8 day average time)
• Higher low latitude density of COSMIC2 has potential for significant increase in impact
– Low lat. equivalent ~18K global occ/day
Kursinski et al., 29 GVAP MOOG Oct 1, 2013
Relative Humidity Histogram 346 mb• Sharp fall off at 100% RH• Upper & lower edge asymmetry => T error = 0.8 K • Suggestion of small % of supersaturation• Apparently no air drier than 5% RH => 0.075 g/kg
Kursinski et al., 30 GVAP MOOG Oct 1, 2013
Cluster AnalysisUsing cluster analysis • to extract patterns in the vertical structure of the
moisture profiles• and their horizontal distribution • to help infer underlying processesThis led to identification of an ENSO related pattern and related research
Kursinski et al., 31 GVAP MOOG Oct 1, 2013
Deep dry boundary layers over North Africa
• Deep well mixed layers to 6 to 7 km altitude in July
• Note 200-300 m vertical resolution across BL tops
Kursinski et al., 32 GVAP MOOG Oct 1, 2013
Free Troposphere Water Vapor-based ENSO Index
Centroids of 2 wettest Nov-Dec-Jan (NDJ) clusters from CHAMP & COSMIC track ENSO (SST) phase and intensity• Tied to deepest, wettest convection, tracks max SST• 3rd & 4th wettest clusters do NOT track the ENSO
• Signature NOT present in AIRS or analyses
• CHAMP centroids are noisier than COSMIC due to limited # of samples
• PWFT centroids are shifted west of SST centroids
– due to trade winds?
Blue La NinaBlack neutralRed El Nino
CHAMPCOSMIC
SST
PWFT-wettest
Kursinski et al., 33 GVAP MOOG Oct 1, 2013
547 hPa Specific Humidity Comparisons• c
Analyses & AIRS underestimate very high humidity air
~5 km altitude
Kursinski et al., 34 GVAP MOOG Oct 1, 2013
Relation to Precipitation• Water vapor-precipitation relation evident in overlaying very high
PWFT over TRMM rainfall over a month
• Next year: Direct observation of large hydrometeors via polarization (Spanish PAZ experiment)
Kursinski et al., 35 GVAP MOOG Oct 1, 2013
NCARMarch 5, 2009
Driest air PDFs January 2007 minus 2008
El Nino
La Nina
Kursinski et al., 36 GVAP MOOG Oct 1, 2013
GPS vs AIRS Fractional DRH vs. AltitudeDJF 06-07 minus DJF 07-08 in %
725
600
500
400
346
650
AIRS GPS
• Similar patterns• GPS DRH is larger by x2-3
Kursinski et al., 37 GVAP MOOG Oct 1, 2013
Future cm & mm wavelength Occultation System:
Active Temperature, Ozone & Moisture Microwave
Spectrometer (ATOMMS)
Kursinski et al., 38 GVAP MOOG Oct 1, 2013
ATOMMS Overview• Actively probes H2O 22 GHz & 183 GHz lines
– Profile both speed of light and absorption of light • Profile water vapor & temperature simultaneously which
GPS RO cannot do, to much higher altitudes• Works in clear air and clouds
• Also other constituents like O3, N2O, H218O, HDO
Þ Cross between GPS RO & MLSÞ LEO Constellation of ATOMMS
ATOMMS
Kursinski et al., 39 GVAP MOOG Oct 1, 2013
Precision of Individual Water Vapor Profiles
0.01 0.1
0
20
40
60
80
lo-band: 8.0, 13.0, 17.5, 20.0, 22.21, 32.0+ hi-band: 179.0, 182.2, 183.0, 183.2, 183.3, 183.31+ high altitude pressure boundary condition
lo-band: 8.0, 13.0, 17.5, 20.0, 22.21, 32.0+ hi-band: 179.0, 182.2, 183.0, 183.2, 183.3, 183.31 without the high altitude pressure boundary condition
lo-band only: 8.0, 13.0, 17.5, 20.0, 22.21, 32.0
Fractional RMS water vapor error
alti
tud
e (
km)
183.31/179.0
183.30/179.0
10%
With turbulence
1%
Kursinski et al., 40 GVAP MOOG Oct 1, 2013
Precision of Individual Temperature Profiles
0.01 0.1 1 100
20
40
60
80
SNRv0
(183GHz) = 1800
ionosphere max day (abs)0.005 mm/s RMS velocity error
local multipathabel integral initialization
hydrostatic integral initialization
Horizontal + tropical water vapor errorWater vapor error for 35S (June)
Water vapor error for 60S (June)no hydrostatic initialization
RMS temperature error (K)
Alti
tud
e (
km)
10.1
Kursinski et al., 41 GVAP MOOG Oct 1, 2013
ATOMMS Instrument• Prototype instrument developed & making optical depth &
spectroscopy measurements on mountaintops• New instrument design to reduce mass & size for WB57
– Reflective rather than refractive optics, closer to LEO version
• Proposed to NASA IIP for aircraft to aircraft demonstration
Kursinski et al., 42 GVAP MOOG Oct 1, 2013
Water Vapor Retrievals: Clear, Cloudy & Rain• Using mountaintop observations to demonstrate ability to
retrieve water vapor spectra in clouds and rain – Enabled by calibration tone at 198 GHz– Figures show spectrum of amplitude ratios relative to calibration
tone
Clear Rain
Kursinski et al., 43 GVAP MOOG Oct 1, 2013
ATOMMS Constraints on Surface Fluxes?
• Vertical fluxes of sensible heat, FSH, & latent heat, FLH, can be represented as downgradient diffusion processes
ATOMMS can measure • Vertical gradients of q & q• Scintillations, related to turbulence & K• over regions of low topography like the oceans
FSH K HacP
z
FLH KWaLv
q
z
Kursinski et al., 44 GVAP MOOG Oct 1, 2013
Near-Surface Precision with 3, 22 & 183 GHz tones
• a
subArctic WintersubArctic Winter
Temperature Error
Temperature Error
Water Vapor Fractional Error
Water Vapor Fractional Error
TropicsTropics
Kursinski et al., 45 GVAP MOOG Oct 1, 2013
Thanks
Kursinski et al., 46 GVAP MOOG Oct 1, 2013