paper impact of assimilating cygnss data on tropical...

P A P E R

Impact of Assimilating CYGNSS Data onTropical Cyclone Analyses and Forecastsin a Regional OSSE Framework
A U T H O R SBrian McNoldyRosenstiel School of Marine andAtmospheric Science, Universityof Miami
Bachir AnnaneNOAA Atlantic Oceanographicand Meteorological Laboratory,Miami, Florida

Sharanya MajumdarRosenstiel School of Marine andAtmospheric Science, Universityof Miami

Javier DelgadoLisa BucciNOAA Atlantic Oceanographicand Meteorological Laboratory,Miami, FloridaCooperative Institute for Marineand Atmospheric Studies, Universityof Miami

Robert AtlasNOAA Atlantic Oceanographicand Meteorological Laboratory,Miami, Florida

A B S T R A C T
The impact of assimilating ocean surface wind observations from the Cyclone
Global Navigation Satellite System (CYGNSS) is examined in a high-resolutionObserving System Simulation Experiment (OSSE) framework for tropical cyclones(TCs). CYGNSS is a planned National Aeronautics and Space Administration con-stellation of microsatellites that utilizes existing GNSS satellites to retrieve surfacewind speed. In the OSSE, CYGNSSwind speed data are simulated using output froma “nature run” as truth. In a case study using the regional Hurricane WeatherResearch and Forecasting modeling system and the Gridpoint Statistical Inter-polation data assimilation scheme, analyses of TC position, structure, and intensity,together with large-scale variables, are improved due to the assimilation of theadditional surface wind data. These results indicate the potential importance ofCYGNSS ocean surface wind speed data and furthermore that the assimilation ofdirectional information would add further value to TC analyses and forecasts.Keywords: Observing System Simulation Experiment (OSSE), Cyclone GlobalNavigation Satellite System (CYGNSS), tropical cyclone

Introduction

Prior to the design and launch of anew satellite platform, a quantitativeassessment of the potential of the satel-lite to improve numerical analyses andforecasts is potentially valuable to helpagencies make informed decisions in acost-effective manner. First, prioritiesfor forecast improvement can be set;for example, data from the new plat-

form may be expected to improvepredictions of tropical cyclone (TC)track, structure, and intensity. Then,the Observing System SimulationExperiment (OSSE) framework canbe used to prepare such a quantitativeassessment. The foundation of anyOSSE is a “nature run,”which is treatedas a proxy for the “real world” provid-ing the “truth” for the simulationof observational datasets and for theverification for analyses and forecasts.An important step is to rigorouslyevaluate the physical realism of thenature run. A data assimilation schemeand numerical model are then used toevaluate the impact of assimilatingthe data in question. OSSEs offer theflexibility to test different configura-tions of existing and proposed observ-ing systems, including their error

January/Feb

characteristics, and also of the fore-cast and data assimilation systems.Traditionally, OSSEs have been con-ducted in global modeling frame-works of relatively coarse resolution(Atlas, 1997). This study presents aprototype effort using a high-resolution,regional modeling framework thatmimics the operational regional hurri-cane forecasting system at NationalOceanic and Atmospheric Administra-tion (NOAA). For further perspectiveson modern OSSEs, the interestedreader is referred to Hoffman andAtlas (2016).

This paper uses the OSSE ap-proach to demonstrate the potentialimpact of assimilating retrieved oceansurface wind speed observations fromthe Cyclone Global Navigation SatelliteSystem (CYGNSS), which was the first

ruary 2017 Volume 51 Number 1 7

Earth Venture mission selected by theNational Aeronautics and SpaceAdministration (NASA) (Ruf et al.,2016). Previously, OSSEs quantifiedthe impact of assimilating ocean sur-face vector winds from scatterometerson global model analyses and forecasts(Atlas et al., 2001). Here the focus ison CYGNSS and state-of-the-arthigh-resolution TC analysis and pre-diction. CYGNSS, a spaceborne mis-sion launched in December 2016,has a primary motivation of samplingwinds in TCs at high resolution,while largely avoiding the rain contam-ination problems of higher-frequencysatellite scatterometers and microwaveradiometers. CYGNSS consists of aconstellation of eight microsatellitesin a nominal 35° inclination orbit at500-km altitude. It combines the all-weather performance of GPS-basedbistatic reflectometry with the spatialand temporal sampling properties ofa constellation of observatories toprovide the ability to retrieve oceansurface wind speeds in all precipitatingconditions and with frequent revisittimes. This paper represents the firstattempt to quantify the potential influ-ence of CYGNSS on numerical TCanalyses and predictions, providing abaseline for future studies.

OSSE FrameworkThe regional OSSE framework,

developed jointly between the NOAAAtlantic Oceanographic and Meteoro-logical Laboratory and the Universityof Miami, is illustrated in Figure 1. Itis based on a high-resolution regionalnature run embedded within a lower-resolution global nature run (Atlaset al., 2015a, 2015b). The regionalnature run was created using Version3.2.1 of the Advanced ResearchWeather Research and Forecasting

8 Marine Technology Society Journal

model (WRF-ARW) with an outerfixed domain of 27 km grid spacingspanning the tropical Atlantic basinand three telescoping storm-followingnested grids of 9, 3, and 1 km (Nolanet al., 2013). The global nature run isthe European Centre for Medium-Range Weather Forecasts (ECMWF)T511 simulation described in Realeet al. (2007).

Simulated conventional observa-tions are generated for a variety ofplatforms from the ECMWF naturerun, including radiosondes, surfacestations, and numerous satellite-basedinstruments (e.g., GOES-Imager,GOES-Sounder, VIIRS, SEVERI,HIRS, CrIS, IASI, SSMI/S, AMSU-A,AMSU-B, MHS, ATMS, and GPS).In a fashion similar to that describedin Zhang and Pu (2010) for a Dopplerwind Lidar study, simulated CYGNSSobservations are derived from theregional WRF nature run and includerandom errors as well as realistic mea-surement uncertainty, which is a func-tion of the strength of the reflectedGPS signal at the specular point.

The full suite of synthetic observa-tions is then assimilated into the Grid-point Statistical Interpolation (GSI)3-D Variational scheme used by theNational Centers for EnvironmentalPrediction (NCEP), with 9-km gridspacing (Shao et al., 2016, and refer-ences therein). The GSI analysis isused to initialize theHurricaneWeatherResearch and Forecasting (HWRF)

reg iona l forecas t mode l (v3 .5)(Bernardet et al., 2015; Tallapragadaet al., 2014; Atlas et al., 2015c), whichis configured in this study with a fixed9-kmparent domain and a 3-kmnestedstorm-following domain. The HWRFmodel parameterizations include theGlobal Forecast System (GFS) plane-tary boundary layer scheme, the newSimplified Arakawa-Schubert cumulusscheme (only for the parent domainsince convection is explicit in the nesteddomain), the Ferrier microphysicsscheme, and the Geophysical FluidDynamics Laboratory (GFDL) schemefor shortwave and longwave radiation.The WRF nature run domain as wellas the embedded parent and nestedHWRF domains are illustrated inFigure 2.

In this study, the GSI analyses andHWRF forecasts are both verifiedagainst the regional hurricane naturerun. A similar framework was used byAtlas et al. (2015b) to investigate thepotential impact of an Optical Auto-covariance Wind Lidar (OAWL) onTC prediction.

Synthetic ObservationalData

In addition to the conventionaldata described in the previous section,CYGNSS data are simulated and uti-lized in the form of retrieved values ofocean surface wind speed (Ruf et al.,2016, and references therein). Each

FIGURE 1

Flow chart of the regional OSSE framework.

microsatellite receives useful signalsreflected off the ocean from the larger,higher-orbiting, and more expensiveGPS satellites whenever a “specularpoint” exists on the ocean and a directline-of-sight is achieved (Figure 3).Since the roughness of water has aphysical relationship to the strengthof the wind blowing across it (Atlaset al., 1996) and the reflected/scatteredsignal contains information on ocean

surface roughness, a surface windspeed can be derived (Garrison &Katzberg, 1998). The direct line-of-sight signal contains timing, location,and frequency information.

For this work, the CYGNSS Sci-ence Team created two distinct data-sets based on the same orbital dataof the eight CYGNSS satellites: anominal resolution product and anenhanced high-resolution product. In

January/Feb

the nominal resolution product, theeffective averaging area (footprint)is 25 km across, whereas in the high-resolution product, the effective foot-print is roughly 12.5 km across. Eachdata point has a quality flag and anassigned error estimate that scales in-versely with the antenna gain on theCYGNSS satellite. Hence, retrievalsderived from a weak signal will havehigher errors. Retrievals with an an-tenna gain below a certain thresholdare flagged and are not assimilated.Although the high-resolution datasetcontains roughly twice the numberof observations as compared to thenominal dataset, the observations arenoisier and, after removing the obser-vations flagged as bad, actually containfewer usable data points.

Over the subtropical and tropicallatitudes, the temporal coverage fromCYGNSS will generally be superiorto that from existing platforms thatsample ocean surface winds, with ahigher frequency of revisits at anygiven point in the CYGNSS latitudeband. An example of excellent spa-tial coverage over the western NorthAtlantic Ocean during a 6-h windowis shown in Figure 4, with nearly20,000 data points within the plotteddomain (at the nominal resolution)and with complete coverage of theTC during this period.

Assimilation-ForecastExperiments

To evaluate the potential impactof assimilating different configu-rations of CYGNSS data, severalHWRF analysis-forecast cycles are runwithin the OSSE system. First, as abenchmark, a “control run” is preparedusingmany of the conventional data thatare routinely assimilated now, includ-ing radiosondes, atmospheric motion

FIGURE 2

The WRF nature run outer domain is outlined by the thick black line, and the HWRF forecastmodel’s parent and nested domains are outlined by the blue lines (the nested domain moveswith the TC). The thin gray line traces the TC center through its evolution in the nature run.(Color version of figures are available online at: http://www.ingentaconnect.com/content/mts/mtsj/2017/00000051/00000001.)

FIGURE 3

Geometry of GPS-based, quasispecular surface scattering. The GPS direct signal provides thelocation, timing, and frequency references, whereas the forward scattered signal contains informationon ocean surface properties. Components and distances in the schematic drawing are not to scale.Background photograph of the ocean surface in Hurricane Isabel (2003) is courtesy of Will Drennan.

ruary 2017 Volume 51 Number 1 9

vectors, infrared and microwave radi-ances, but not scatterometers. Next,the question of the impact of assimilat-ing CYGNSS data in the nominal res-olution and high-resolution datasets,in which observations are discarded ifthe antenna gain is too low and realisticerrors are assigned to the remainingdata points, is addressed. Finally, thequestion of the maximum potentialbenefit that could be derived fromCYGNSS data is addressed by assimi-lating “perfect”CYGNSS observationsof wind speed or of wind speed anddirection, this latter case to assess theadded benefit of directional infor-mation. This dataset is generated bysimply interpolating the nature runto each latitude-longitude location inthe high-resolution CYGNSS datasetand assigning zero error. Althoughthe planned CYGNSS data productwill not include wind direction, thereare two reasons to study this configura-

10 Marine Technology Society Journa

tion: (1) there is some directional in-formation in the reflected signal,which might be extracted with morecapable hardware or software (e.g.,Komjathy et al., 2004), and (2) a vari-ational analysis method discussed inConcluding Remarks could be appliedto the wind speed retrievals to createdynamically realistic wind vectors.

Thus, the following experimentswere conducted:1. CONTROL: Conventional data

only, listed in OSSE FrameworkNo CYGNSS data.

2. REAL_SPD: CONTROL plusquality-controlled nominal resolu-tion CYGNSS data, with realisticerror assignments.

3 . REAL_SPD_HI : S imi l a r toREAL_SPD, bu t u s ing thequality-controlled high-resolutionCYGNSS dataset.

4. PERFECT_SPD: CONTROLplus all available high-resolution

l

CYGNSS data points, where thewind speed is interpolated fromthe nature run and assumed tohave zero observational error.

5 . PERFECT_VEC : S imi la r toPERFECT_SPD, but wind speedand direction are interpolated.Each of these five listed experi-

ments is initiated with a “cold start”at 0000 UTC on 1 August in whichglobal analyses are used as initial andboundary conditions, and then cyclingis performed every 6 h through to0000 UTC on 5 August, for a totalof 16 analyses. Note that the datesand times in this study only corre-spond to a TC in the nature run andnot an actual TC. A 5-day HWRFforecast is integrated from each analy-sis. Each experiment is then verifiedagainst the nature run. The cold startand the first four cycles are discardedto avoid the artificial effects of vortexspin-up and model adjustment (asdescribed by Atlas et al., 2015a),which leaves 12 reliable cycles for cal-culating the error statistics.

ResultsThe impact of assimilating the

different configurations of CYGNSSdata listed in the previous sectionis evaluated on both the basin scaleand vortex scale. Overall, the addi-tion of CYGNSS observations im-proves upon CONTROL, bringingthe analyses closer to the nature runon average.

For the basin scale verification,values are averaged over the HWRFparent domain and the identical areain the nature run. Although CYGNSSonly introduces surface wind dataover water, the improvements toCONTROL appear to extend beyondthe surface wind field. Smaller, short-term improvements are also found in

FIGURE 4

An example of excellent TC coverage by CYGNSS wind speed retrievals over a 6-h window cen-tered on the 0000 UTC 8 August synoptic time in the nature run. These wind speed data are se-lected from the nominal resolution data product with flagged observations removed. The dashedgray line traces the center of the TC throughout the duration of the nature run, whereas the solidblack segment highlights the TC’s position during this 6-h period.

the domain-averaged winds up to200 hPa as well as the correspondinggeopotential height, pressure, andtemperature (not shown).

Figure 5 shows an example ofthe improvements in the basin scale10 m wind field (top) and 200 hPawind field (bottom). The errors arepresented as skill scores relative tothe CONTROL run. The noisyhigh-resolution CYGNSS wind exper-iment (REAL_SPD_HI) had the leastskill of the experiments, and as expected,the lowest errors were found for thewind vectors interpolated directlyfrom the nature run at the CYGNSScoordinates (PERFECT_VEC), withan improvement of approximately

32% (5%) over CONTROL at 0 h(24 h) at 10 m, and then at 200 hPa,the improvement was approximately7% (1%) at 0 h (24 h). In general, littleimpact is seen in the forecasts beyond24 h, and statistically significant differ-ences do not extend beyond 12 h.

For the verification of TC track andintensity, the impact of assimilatingCYGNSS data is generally positivethough small. Slight improvements tothe track are found in the two experi-ments using “perfect” wind vectorsand speeds at the analysis time and inforecasts out to approximately 18 h(Figure 6a). It should be noted thatvortex relocation is not used in thisstudy. It is also important to note

January/Febru

that the average 12- to 48-h trackforecast errors in CONTROL areconsiderably smaller than the corre-sponding average track forecast errorsof the 2014 version of the operationalHWRF for real TCs, leaving littleroom for improvement from theassimilation of extra data. The intensi-ty errors, as represented by the error ofthe maximum 10-m wind speed, areimproved at the analysis time for alldatasets except for the high-resolutiondataset. For PERFECT_VEC (REAL_SPD), the average improvement isaround 35% (25%) (Figure 6b). Thecorresponding intensity forecasts outto 24 h are slightly improved inPERFECT_VEC and REAL_SPDby the addition of CYGNSS data byaround 15% at 12 h and decreasingto 10% at 24 h. In general, the im-provements in intensity and trackerrors are not significant at the80% level, with the exception of thePERFECT_VEC intensity analy-sis. However, many of the increased er-rors in intensity and track from theREAL_SPD_HI experiment are statis-tically significant. It should be notedthat there is more room for improve-ment in the intensity analyses andvery short range forecasts than for fore-casts beyond 1 day. The correspondingresults for minimum central pressureare similar to those of Figure 6b (notshown). Due to the small sample sizecommon in OSSE studies of TCsand to the relatively small contributionthat CYGNSS data make to the overalldata volume, we would not expect tofind statistically significant differencesamong all of the experiments andacross all forecast lead times. It is likelyfor these reasons that the “perfect” ex-periments do not always show lowererrors than the “real” experiments.

An examination of maps of the sur-face wind analyses provides additional

FIGURE 5

Average skill score of 5-day HWRF forecasts of (a) 10 m wind speed and (b) 200 hPa wind speedover the HWRF parent domain, averaged over the 12 forecast cases. Skill is calculated relative toCONTROL, and squares mark where the values are statistically significant at the 80% level.

ary 2017 Volume 51 Number 1 11

insights and reveals the inherentstrengths and limitations of the dataassimilation system. In a typical exam-ple shown in Figure 7, at 1200 UTC 3August 2005, the nature run features asymmetric hurricane with a peak windof 89 knots (Figure 7a), whereas theCONTROL analysis, at the 10thdata assimilation cycle at this time,shows a much weaker and more asym-metric vortex (Figure 7b). A compari-son of the quality-controlled datareveals the locations of the data thatare flagged as bad in the realistic con-figuration and renders the distributionof assimilated data more asymmetricover the TC in REAL_SPD comparedto PERFECT_VEC (Figures 7c and


7d). In contrast, after assimilation ofthe perfect wind speed and directionaldata, the vortex in the PERFECT_VEC analysis has a lower central pres-sure and is more symmetric than inCONTROL and overall closer to thenature run (Figure 7e). Finally, poorerCYGNSS data coverage in the REAL_SPD assimilation results in a lopsidedvortex, although it is still an improve-ment upon CONTROL (Figure 7f ).Since the covariance structure in the3-D Variational GSI is quasi-isotropicand not flow-dependent, the correc-tions to the surface wind analysis arestrongest locally where the CYGNSSdata exist, to the north of the TC,with no equivalent correction on the

l

southern flank, which is devoid ofdata; this is general characteristic of3D-VAR and is not specific to thisstudy, surface wind speed observa-tions, or CYGNSS.

Concluding RemarksThe potential for surface wind data

sampled from the constellation ofCYGNSS satellites to improve nu-merical analyses and forecasts of TCshas been examined using a novel,high-resolution regional OSSE frame-work. Both “perfect” and realistic con-figurations of wind data at CYGNSSlocations over the ocean surface were ex-amined, at nominal and high resolution.

Most promisingly, the assimilationof CYGNSS data almost always im-proved the HWRF/GSI analyses ofTC track, intensity, and structure.The corresponding forecasts wereimproved out to 1 day on average,when wind speed and direction sam-pled at CYGNSS locations were as-sumed to be perfect. In reality, theobservations will of course be im-perfect, and wind direction data areunlikely to be available initially. How-ever, the REAL_SPD experiments sug-gested that the assimilation of realisticwind speed data do have the potentialto improve the vortex scale analysesand short-range forecasts. Additionally,the basin scale analyses and short-termforecasts of wind speed through thetroposphere were improved, with thelargest improvements in the surfacewind field analysis. Impacts becomesmaller and less statistically significantat levels higher in the atmosphere andat longer forecast lead times.

The added directional informationwas found to produce better analysesand forecasts than if wind speed onlywere assimilated. This suggests thatadditional efforts to incorporate wind

FIGURE 6

Errors of TC (a) track and (b) maximum 10 m wind speed averaged over the twelve 5-day HWRFforecasts. Note that negative wind speed errors imply that the TC is stronger in the nature run.

direction would be important, viaexploiting the oversampling of theocean surface and/or using a methodsuch as the Variational AnalysisMethod (Atlas et al., 2011, and refer-ences therein) as a preprocessing stepto the assimilation. Finally, althoughthe enhanced high-resolutionCYGNSSdataset containedmore data points thanthe nominal product, the retrievals werevery noisy, and many of them had tobe discarded, resulting in degradedanalyses and forecasts.

In summarizing these results, it isimportant to recognize that the samplesize from a single TC is small, so theerror statistics are not always robust. Asignificantly larger number and variety

of cases is required to draw statisticallysignificant conclusions, and the find-ings in this paper are presented primar-ily to suggest that CYGNSS shouldbe a beneficial addition to the suiteof TC monitoring platforms alreadyin existence. Even if a few additionalnature runs with TCs in them wereavailable, the verification sample sizewould still be relatively small, a com-mon limitation of TC OSSE studies.

Although the regional OSSE con-figuration used here is novel for TCsand closely mimics NOAA’s opera-tional system, it possesses limitations.One well-known problem in allmodels of TCs is that intense vorticesin the analysis suffer from an unreal-

January/Febru

istic spin-down as the model attemptsto adjust and balance the wind andmass fields (Hendricks et al., 2013;Gopalakrishnan et al., 2012). Typi-cally, the stronger a vortex is in theanalysis, the worse the spin-downproblem is at the beginning of themodel forecast. Additionally, a signifi-cant shortcoming of the GSI 3-D var-iational assimilation scheme is that it isextremely sensitive to the locations ofavailable observations. Even for a sym-metric hurricane, observations only onone side of the vortex will degrade theanalysis and cause the analyzed vortexto be asymmetric. The influence fromthe surface wind observations may nottranslate through the depth of the tro-posphere in the subsequent analysis.To remedy these issues, a Hybrid3D-Variational/Ensemble KalmanFilter assimilation scheme (Wang et al.,2013) is being implemented in theOSSE system. In parallel, a study un-derway seeks to determine the optimalassimilation frequency, which for rap-idly changing weather systems such asTCs and for data available at continu-ous times such as CYGNSS may beshorter than the 6 h used in this studyand in NOAA’s operational system.Another OSSE challenge is to providecontrol simulations whose errors asevaluated against the nature run arecomparable inmagnitude and structureto errors in operational forecasts. Moti-vated by the REAL_SPD_HI andPERFECT_VEC results, the OSSEframework presented here could alsobe used to evaluate future CYGNSS-like platforms with more sensitive re-ceivers capable of producing reliablehigher-resolution retrievals or of pro-viding wind direction information.Ongoing efforts in these areas will be re-ported in the near future.

The OSSE framework offers theflexibility to examine the impact of

FIGURE 7

Top row: 10 m wind speed and surface pressure analyses from (a) NATURE and (b) CONTROL.Middle row: CYGNSS data coverage maps spanning the 6-h period centered on 3 August 20051200 UTC for the (c) PERFECT_VEC and (d) REAL_SPD experiments. The red dashed boxes high-light the area shown in the panels on the top and bottom rows. Bottom row: 10 m wind speed andsurface pressure analyses from (e) PERFECT_VEC and (f) REAL_SPD.

ary 2017 Volume 51 Number 1 13

assimilating different configurations ofexisting and future datasets. For exam-ple, it is straightforward to build onthis study and examine the impact oflaunching another eight CYGNSS sat-ellites, to place the satellites in differentorbital configurations, or other trade-offs. The benefits of activating ahigh-resolution sampling mode onCYGNSS, localized over TCs, canalso be examined. Furthermore, thesynergies between CYGNSS, specialhigh temporal resolution atmosphericmotion vectors, and wind data fromfuture spaceborne Lidars (Bakeret al., 2014) can be examined. Thus,realistic and validated OSSE systemsoffer a promising pathway forward inoptimizing the use of new and futureinstruments to improve forecasts ofhigh-impact weather events.

AcknowledgmentsThis study was supported by NASA

Award NNL13AQ00C. The authorsthank Christopher Ruf at the Univer-sity of Michigan and the CYGNSS Sci-ence Team for the simulated CYGNSSdatasets, the NOAA Office of Weatherand Air Quality for funding the devel-opment of the regional OSSE frame-work, the NOAA Hurricane ForecastImprovement Project for computingresources, the Developmental TestbedCenter for the GSI and HWRF codeand support, and David Nolan at theUniversity of Miami for providing theWRF nature run dataset.

Corresponding Author:Brian McNoldyRosenstiel School of Marine andAtmospheric Science, Universityof Miami4600 Rickenbacker CausewayMiami, FL 33149Email: [email protected]


ReferencesAtlas, R. 1997. Atmospheric observations and

experiments to assess their usefulness in data

assimilation. JMeteorol Soc Jpn. 75(1B):111-30.

Atlas, R., Bucci, L., Annane, B., Hoffman, R.,

& Murillo, S. 2015a. Observing System Sim-

ulation Experiments to assess the potential

impact of new observing systems on hurricane

forecasting. Marine Technol Soc J. 49(6):140-8.

https://doi.org/10.4031/MTSJ.49.6.3.

Atlas, R., Hoffman, R.N., Ardizzone, J.,

Leidner, S.M., Jusem, J.C., Smith, D.K., &

Gombos, D. 2011. A cross-calibrated, multi-

platform ocean surface wind velocity product

and meteorological and oceanographic appli-

cations. B Am Meteorol Soc. 92:157-74.

https://doi.org/10.1175/2010BAMS2946.1.

Atlas, R., Hoffman, R.N., Bloom, S.C.,

Jusem, J.C., & Ardizzone, J. 1996. A multi-

year global surface wind velocity dataset using

SSM/I wind observations. B Am Meteorol

Soc. 77:869-82. https://doi.org/10.1175/

1520-0477(1996)077<0869:AMGSWV>2.0.

CO;2.

Atlas, R., Hoffman, R.N., Leidner, S.M.,

Sienkiewicz, J., Yu, T-W, Bloom, S.C., …

Jusem, J.C. 2001. The effects of marine winds

from scatterometer data on weather analysis and

forecasting. B Am Meteorol Soc. 82:1965-90.

https://doi.org/10.1175/1520-0477(2001)

082<1965:TEOMWF>2.3.CO;2.

Atlas, R., Hoffman, R.N., Ma, Z., Emmitt,

G.D., Wood, S.A., Jr., Greco, S.,…Murillo, S.

2015b. Observing System Simulation Ex-

periments (OSSEs) to evaluate the potential

impact of an Optical Autocovariance Wind

Lidar (OAWL) on numerical weather predic-

tion. J Atmos Ocean Tech. 32:1593-613.

https://doi.org/10.1175/JTECH-D-15-0038.1.

Atlas, R., Tallapragada, V., &Gopalakrishnan, S.

2015c. Advances in tropical cyclone intensity

forecasts. Mar Technol Soc J. 49:149-60.

https://doi.org/10.4031/MTSJ.49.6.2.

Baker,W.E., Atlas, R., Cardinali, C., Clement,

A., Emmitt, G.D., Gentry, B.M., … Yoe,

J.G. 2014. Lidar-measured wind profiles: The

missing link in the global observing system.

l

B Am Meteorol Soc. 95:543-64. https://doi.

org/10.1175/BAMS-D-12-00164.1.

Bernardet, L., Tallapragada, V., Bao, S.,

Trahan, S., Kwon, Y., Liu, Q., … Gall, R.

2015. Community support and transition

of research to operations for the Hurricane

Weather Research and Forecasting model.


org/10.1175/BAMS-D-13-00093.1.

Garrison, J.L., & Katzberg, S.J. 1998. Effect

of sea roughness on bistatically scattered range

coded signals from the Global Positioning

System. Geophys Res Lett. 25:2257-60.

https://doi.org/10.1029/98GL51615.

Gopalakrishnan, S.G., Goldenberg, S.,

Quirino, T., Zhang, X., Marks, F., Jr., Yeh,

K.-S., … Tallapragada, V. 2012. Toward

improving high-resolution numerical hurri-

cane forecasting: Influence of model horizon-

tal grid resolution, initialization, and physics.

Weather Forecast. 27:647-66. https://doi.org/

10.1175/WAF-D-11-00055.1.

Hendricks, E.A., Peng, M.S., & Li, T. 2013.

Evaluation of multiple dynamic initialization

schemes for tropical cyclone prediction. Mon

Weather Rev. 141:4028-48. https://doi.org/

10.1175/MWR-D-12-00329.1.

Hoffman, R.N., & Atlas, R. 2016. Future

Observing System Simulation Experiments.


org/10.1175/BAMS-D-15-00200.1.

Komjathy, A., Armats, M., Masters, D., &

Axelrad, P. 2004. Retrieval of ocean surface

wind speed and direction using reflected GPS

signals. J Atmos Ocean Tech. 21:515-26.

https://doi.org/10.1175/1520-0426(2004)

021<0515:ROOSWS>2.0.CO;2.

Nolan, D.S., Atlas, R., Bhatia, K.T., Bucci,

L.R. 2013. Development and validation of a

hurricane nature run using the Joint OSSE

nature run and the WRF model. J Adv Model

Earth Syst. 5: 24pp. https://doi.org/10.1002/

jame.20031.

Reale, O., Terry, J., Masutani, M., Andersson,

E., Riishojgaard, L.P., & Jusem, J.C. 2007.

Preliminary evaluation of the European

Centre for Medium-range Weather Forecasts’

(ECMWF) nature run over the tropical Atlan-

tic and African monsoon region. Geophys

Res Lett. 34: 6pp. https://doi.org/10.1029/

2007GL031640.

Ruf, C., Atlas, R., Chang, P., Clarizia, M.P.,

Garrison, J., Gleason, S., … Zavorotny, V.

2016. New ocean winds satellite mission to

probe hurricanes and tropical convection.


org/10.1175/BAMS-D-14-00218.1.

Shao, H., Derber, J., Huang, X., Hu, M.,

Newman, K., Stark, D., … Brown, B. 2016.

Bridging research to operations transitions:

Status and plans of Community GSI. B Am

Meteorol Soc. 97:1427-40. https://doi.org/

10.1175/BAMS-D-13-00245.1.

Tallapragada, V., Bernardet, L., Biswas, M.K.,

Gopalakrishnan, S., Kwon, Y., Liu, Q., …

Zhang, X. 2014. Hurricane Weather Research

and Forecasting (HWRF) Model: 2014 Scien-

tific Documentation. NCAR Development

Tested Bed Center Report, 105pp.

Wang, X., Parrish, D., Kleist, D., &Whitaker,

J. 2013. GSI 3DVar-Based Ensemble–

Variational hybrid data assimilation for NCEP

Global Forecast System: Single-resolution

experiments. MonWeather Rev. 141:4098-117.

https://doi.org/10.1175/MWR-D-12-00141.1.

Zhang, L., & Pu, Z. 2010. An Observing

System Simulation Experiment (OSSE) to

assess the impact ofDopplerWind Lidar (DWL)

measurements on the numerical simulation of

a tropical cyclone. Adv Meteorol. 2010: 14pp.

https://doi.org/10.1155/2010/743863.

January/February 2017 Volume 51 Number 1 15

paper impact of assimilating cygnss data on tropical...

Documents