Iodine-filter-based mobile Doppler lidar to make continuous and...

Iodine-filter-based mobile Doppler lidar to makecontinuous and full-azimuth-scanned wind

measurements: data acquisition andanalysis system, data retrievalmethods, and error analysis

Zhangjun Wang,1,3,* Zhishen Liu,1 Liping Liu,2 Songhua Wu,1

Bingyi Liu,1 Zhigang Li,1 and Xinzhao Chu3

1Ocean Remote Sensing Institute, Ocean University of China, 5 Yushan Road, Qingdao 266003, China2State Key Laboratory of Severe Weather, Chinese Academy of Meteorological Sciences, China Meteorology

Administration, 46 Zhongguancun Nanda Street, Beijing 100081, China3Cooperative Institute for Research in Environmental Sciences & Department of Aerospace Engineering Sciences,

University of Colorado at Boulder, 216 UCB, Boulder, Colorado 80309, USA

*Corresponding author: [email protected]

Received 23 March 2010; revised 27 October 2010; accepted 7 November 2010;posted 9 November 2010 (Doc. ID 125867); published 17 December 2010

An incoherent Doppler wind lidar based on iodine edge filters has been developed at the Ocean Univer-sity of China for remote measurements of atmospheric wind fields. The lidar is compact enough to fit in aminivan for mobile deployment. With its sophisticated and user-friendly data acquisition and analysissystem (DAAS), this lidar has made a variety of line-of-sight (LOS) wind measurements in different op-erational modes. Through carefully developed data retrieval procedures, various wind products are pro-vided by the lidar, including wind profile, LOS wind velocities in plan position indicator (PPI) and rangeheight indicator (RHI) modes, and sea surface wind. Data are processed and displayed in real time, andcontinuous wind measurements have been demonstrated for as many as 16 days. Full-azimuth-scannedwindmeasurements in PPImode and full-elevation-scanned windmeasurements in RHImode have beenachieved with this lidar. The detection range of LOSwind velocity PPI and RHI reaches 8–10 km at nightand 6–8 km during daytime with range resolution of 10 m and temporal resolution of 3 min. In thispaper, we introduce the DAAS architecture and describe the data retrieval methods for various operationmodes. We present themeasurement procedures and results of LOSwind velocities in PPI and RHI scansalong with wind profiles obtained by Doppler beam swing. The sea surface wind measured for the sailingcompetition during the 2008 Beijing Olympics is also presented. The precision and accuracy of wind mea-surements are estimated through analysis of the random errors associated with photon noise and thesystematic errors introduced by the assumptions made in data retrieval. The three assumptions of hor-izontal homogeneity of atmosphere, close-to-zero vertical wind, and uniform sensitivity aremade in orderto experimentally determine the zero wind ratio and the measurement sensitivity, which are importantfactors in LOS wind retrieval. Deviations may occur under certain meteorological conditions, leading tobias in these situations. Based on the error analyses and measurement results, we point out the applica-tion ranges of this Doppler lidar and propose several paths for future improvement. © 2010 OpticalSociety of AmericaOCIS codes: 280.3640, 170.3340, 300.6320.

0003-6935/10/366960-19$15.00/0© 2010 Optical Society of America

6960 APPLIED OPTICS / Vol. 49, No. 36 / 20 December 2010

1. Introduction

Wind is one of the most important variables in atmo-spheric dynamics, and it is crucial to weather fore-casting using numerical models. It is also key tostudying atmospheric coupling and transport. Windmeasurements are important to many other applica-tions like aviation control, sea surface wind monitor-ing, space shuttle launch and landing, etc. Dopplerwind lidars have demonstrated their efficacy in windmeasurements where spatially- and range-resolvedmeasurements of wind velocity are needed.

Both coherent-detection and direct-detectionDoppler lidarshavebeendeveloped forwindmeasure-ments.TheNOAAEnvironmentalTechnologyLabde-veloped a scanning coherent Doppler lidar based on apulsed 10:6 μm CO2 laser in the 1980s [1]. This lidarwasoperatedwitha circular scannerat a rate of3:3°=sand at a constant elevation angle of 0:5° for numerousfield campaigns of wind measurements [2]. NOAA’sHigh-ResolutionDopplerLidar (HRDL) [3]wasdevel-oped for measuring the mean and turbulent velocitystructure of the atmospheric boundary layer. This li-dar is normally mounted in a seatainer and has beenoperated in ground-based and shipborne field experi-ments. Coherent Doppler lidars have been employedto measure the boundary-layer mixing height, seabreeze, downslope windstorms, and flow in complexterrain in thenear-horizontal scanmode over the past25 years [4].

Compared to coherent Doppler lidars that rely onaerosol backscatter signals, and thus are limited tolower atmosphere where aerosols are abundant, inco-herent direct-detection Doppler lidars that can ex-ploit both aerosol and molecular backscatter signalsaswell as atomic resonance fluorescence cover amuchlarger altitude range, to as high as 110 km withresonance-fluorescence Doppler lidars [5], or fromnear ground to∼60 kmwith Rayleigh/Mie Doppler li-dars [6–20]. However, the near-horizontal scan is achallenging task for incoherent lidars. Instead of em-ploying optical heterodyne detection technology to in-fer wind-induced Doppler shift in coherent lidars,direct-detection Doppler lidars convert the Dopplershift into a change in received light power or powerratio using various optical frequency discriminators.These discriminators include Fabry–Perot interfe-rometer, Fizeau etalon, iodine vapor cell filter, Na va-por cell filter, etc., implemented in lidar receiverchains. Metal atoms such as Na, Fe, and K in the me-sosphere and lower thermosphere provide naturalfrequency discriminators with their absorption lines[5]. In 1989, the Service d’Aeronomie ofNational Cen-ter for Scientific Research (CNRS) in France demon-strated a ground-based Rayleigh Doppler lidar usinga frequency-doubled Nd:YAG laser and double-edgefilter made of a dual Fabry–Perot interferometer formeasurements of wind profiles between 25 km and60 km [6]. Later, theGLOW (Goddard LidarObserva-tory for Winds) mobile direct-detection Doppler lidarwas developed by researchers at the NASA/GoddardSpace Flight Center in the U.S.A. It measures wind

from aerosol scattering at 1064 nm and frommolecu-lar scattering at 355 nm, each using a double-edgeFabry–Perot interferometer. The 355 nm systemmeasures wind profiles up to 35 km altitude [7]. In1997, a high-resolution incoherent Doppler lidarusing “full fringe imaging” was constructed at theSpace Physics Research Laboratory at the Universityof Michigan [8]. The ADM-Aeolus Atmospheric LaserDoppler Instrument (ALADIN) will employ an aero-sol channel using a Fizeau interferometer and amolecular channel using a Fabry–Perot etalon [9].A multibeam Fizeau interferometer combined witha linear detector as a spectral analyzer has been pro-posed byMcKay [10]. A direct-detectionDoppler lidarutilizing a multibeam Fizeau interferometer was de-veloped in the Shanghai Institute of Optics and FineMechanics in 2005 for planetary boundary-layerwindfieldmeasurement [11]. At the same time, theUniver-sity of Geneva [12] was involved in the further devel-opment and exploitation of the direct-detection lidar(DDL) double-edge technique.

Resonance-fluorescence Doppler lidar is anothertype of direct-detection Doppler lidar [5], which canmonitor temperature and wind in the mesosphereand lower thermosphere (75–110 km) where the at-mospheric density and aerosol concentration are fartoo low for molecular or aerosol signals to be usedfor windmeasurements. The beauty of this lidar is itsnature-provided frequency discriminator in the at-mosphere—the atomic absorption lines of Fe, Na,and K atoms that are Doppler shifted and broadenedso that they can be used to infer atmosphere tem-perature and wind; thus, the lidar receiver is broad-band for maximum return signals. Recently, Huanget al. [13] proposed a Na double-edge magneto-opticfilter (Na-DEMOF) to extend Na Doppler lidarmeasurements via Rayleigh-Mie scattering into thelower atmosphere. Field campaigns demonstratedwind and temperature profiling in the altitude rangeof 10–45 km with 1 km resolution and 60 minintegration [14].

In 1997, Friedman et al. [15], building on the workof Tepley et al. [16], demonstrated wind measure-ments in the stratosphere and lower mesosphereusing a Rayleigh Doppler lidar based on an iodineedge filter at Arecibo Observatory. Wind profileswere obtained from 18 to 45 km with 1:5 km rangeresolution neglecting the influence of Mie scattering.Around the same time, Liu et al. [17] proposed usingan iodine edge filter for frequency analysis of thebackscattered light from both atmospheric moleculesand aerosols for lower atmosphere wind measure-ments. The iodine-filter-based Doppler lidar hastwo advantages in this application. First, such a mo-lecular filter has absolute frequency calibration andis repeatable from day to day and from location tolocation, in contrast to an etalon system. Second, thisfilter has an absorption line at 532 nm, reachable byfrequency-doubled Nd:YAG laser. The Nd:YAG laserhas reached mature status with high power, singlemode, and single frequency. Diode-laser pumping

20 December 2010 / Vol. 49, No. 36 / APPLIED OPTICS 6961

has further made it robust and reliable, with a highrepetition rate. Therefore, iodine-filter-based direct-detection Doppler lidars are expected to have largepower-aperture product and reliable performance.

Based on the proposal by Liu et al. [17], a mobileDoppler wind lidar was developed at the Ocean Uni-versity of China for boundary-layer and tropospherewind measurements [18,21]. This iodine-filter-basedlidar is capable of scanning measurements of three-dimensional wind field in real time. The lidar systemis compact and integrated into a minivan. Numerousobservational campaigns have been made with thislidar for sea surface, boundary-layer, and tropo-sphere wind measurements [22,23]. It is noteworthythat continuous measurements of sea surface windwere made for 16 days during the 2008 BeijingOlympic Games in support of the sailing competitionin Qingdao, China.

In this paper, we briefly introduce the lidar instru-mentation (Section 2) and then present the dataacquisition and analysis system (DAAS) (Section 3)developed for the mobile Doppler wind lidar. An at-tractive feature of this lidar and its DAAS is its cap-ability of handling a variety ofmeasurements, such aswind profile, line-of-sight (LOS) wind velocity in planposition indicator (PPI) and range height indicator(RHI) modes, sea surface wind (SSW), horizontal vis-ibility, and aerosol backscattering ratio in different

operation modes, while processing data and display-ing retrieved results in real time. This paper will in-troduce the data retrieval methods and discuss theassumptions made in each case (Subsection 4.A), andthen analyze the measurement errors introduced byphoton noise and several assumptions (Subsection4.B). We also present the measurement proceduresand results for wind profiles (Subsection 5.A), LOSwind velocities in PPI and RHI modes (Subsections5.B and 5.C), along with SSWs measured for the2008 Beijing Olympics (Subsection 5.D). Finally, inSection 6 we conclude the paper by pointing out theapplication ranges of this Doppler lidar and proposea few ideas for future improvement.

2. Doppler Wind Lidar Instrumentation

Figure 1 is a schematic diagramof themobile incoher-ent Doppler wind lidar. The entire system comprisesthreemajor function blocks: the lidar transmitter, thelidar receiver, and the data acquisition and analysissystem. An add-on two-axis scanner with 32 cm dia-meter aperture is unique to this lidar as it providesthe full-azimuth and full-elevation scanning capabil-ity. The lidar transmitter provides high-repetition-rate and frequency-stabilized laser pulses at 532 nmby a diode-pumped, frequency-doubled, pulsedNd:YAG laser, injection seeded by a continuous wave(CW) seed laser and stabilized to an iodine absorption

Fig. 1. (Color online) Schematic diagram of the incoherent mobile and scanning Doppler wind lidar.


line [24,25]. The laser repetition rate is 2:8 kHz, andthe single pulse energy is ∼2 mJ with pulse width of30 ns. The seed laser is a CWNd:YAG laser and it hasdual outputs: the 1064 nm laser is injected into thepulsed Nd:YAG laser while the frequency-doubled la-ser at 532 nm is locked to the 50% absorption level ofthe I2 line 1109 to stabilize the seed laser frequency.The frequency stability of the seed and pulsedNd:YAG lasers are 200 kHz and 1 MHz, respectively.The spectral linewidth of the laser pulses is within35 MHz. The outgoing laser beam is expanded to70 μrad divergence by a 20× beamexpander, and thendirected into the skyvia a steeringmirror coaxialwiththe scanner. The two-axis scanner provides coverageover an entire hemisphere (0° − 360° azimuth and0° − 180° elevation). This scanner can accommodatecoaxial transmission and reception with continuousscanning speed of 1°–10°=s and directional accuracyof 0:1°. It enables the scanning measurements withhigh spatial and temporal resolutions.

The lidar receiver is based on the iodine edge-filtertechnique [17], and the iodine absorption line 1109 ischosen as the frequency discriminator. A 30:5 cm di-ameter Schmidt–Cassegrain telescope with a field ofview of 120 μrad is employed to collect backscatteredphotons. The return signals collected by the telescopeare coupled through a fiber and pass through an in-terference filter. This filter is centered at 532 nmwith a passband of 0:11 nm and a peak transmissionof 76%. The signals are then split into two channels:one is detected directly by a photomultiplier tube(PMT) as a reference channel, and another goes

through an iodine filter before being detected by an-other PMT in the measurement channel. Acquisitionof both channels of data is handled by the data acqui-sition and analysis system (DAAS). To cover the hugeand variable dynamic ranges of backscatter signalsfrom the troposphere, a multichannel scaler (MCS)device, manufactured by Licel, is operated in bothphoton counting and analog modes to receive thePMT signals from both channels. The DAAS thenprocesses the data in real time with 10 m resolutionfor the altitude range from 0 m to 20 km to retrievewind products. The DAAS also controls the two-axisscanner and displays the retrieved winds in realtime.

There are three key components for continuousand full-azimuth-scanned wind measurements: thetransmitter with a high-power and high-repetition-rate laser, the receiver with a well-calibrated iodineedge filter, and the user-friendly data acquisition andanalysis system (DAAS) that includes a two-axisscanner. The focus of this paper is the DAAS andthe data retrieval principles and error analysesbehind the DAAS.

3. Architecture of Data Acquisition andAnalysis System

As laid out in Fig. 2, the overall computer-controlsystem of the Doppler wind lidar is comprised oftwo major programs: the laser frequency control pro-gram and the data acquisition and analysis system(DAAS). The frequency-locking module, developedin the National Instrument LabVIEW program, is

Fig. 2. (Color online) Architecture of the overall computer control system for the mobile Doppler wind lidar.


used to stabilize the seed laser frequency to the de-sired points on the I2 absorption line number 1109. Itcan also be adapted to lock the pulsed laser frequencydirectly to the I2 line. Details about this program canbe found in Refs. [24,25] so they are omitted from thispaper. The DAAS consists of five function modules: atwo-axis scanner control program, a data acquisition(DAQ) subsystem, a data storage subsystem, a dataretrieval subsystem, and a display module. As part ofthe “Lidar System Control Software Suite,” the scan-ner control program controls both azimuth and ele-vation scans, and monitors the scanner’s status. Afunction generator is used to provide the time syn-chronization between the transmitter laser and thedata acquisition subsystem. This DAQ subsystemworks with the Licel MCS device to detect the returnphotons in both photon counting and analog modes,and then accumulates the signals for a set number oflaser shots before it writes the raw data profiles intothe data storage area. In addition to storing the rawlidar data in both ASCII and binary formats, thedata storage subsystem also saves the lidar systemparameters provided by the control software suiteand the retrieval products provided by the data re-trieval program. The data retrieval is performed inreal time, and as such, provides users with great flex-ibility. It can derive all possible data products includ-ing retrieval of backscatter signals to further derivehorizontal visibility, retrieval of the aerosol backscat-ter ratio, and retrieval of the LOS wind velocity fromthe measured Doppler shift. Depending on operatingmodes, the LOS wind can be further turned into var-ious products like wind-altitude profiles, LOS windvelocity PPI, LOS wind velocity RHI, and sea surfacewind, etc. All these retrieval products can be illu-strated in the display module depending on users’choices.

The entire DAAS code was written in a combina-tion of C and Cþþ under Microsoft Visual Cþþ6.0 and OpenGL. Figure 3 shows the DAAS user in-terface with raw backscatter photon profiles for themeasurement (lower profile) and reference (upperprofile) channels displayed. The interface includesthe wind profile controller pair, PPI controller pair,RHI controller pair, scanner controller pair, back-scattered signal controller pair, and SSW controllerpair for operation in different scanning modes. Themonitor unit of the DAAS serves as a display modulefor all products and operation parameters. Here allthe peripherals are controlled and system param-eters are kept in ASCII and Binary files. This inter-face was designed to hide as much of the complexityas possible while giving the user a clean but stillinformative display. The lidar system has been de-ployed in many field campaigns [22,23] and theDAAS allows hands-off operation for many days. Itcan be handled easily and efficiently.

The DAQ subsystem is built around the Licel MCSEthernet controller. Its Ethernet-based modulespave the way for truly remotely controlled experi-ments. These use a TCP/IP-based protocol with syn-

tax similar to conventional GPIB-based instruments.The Licel MCS transient recorder systems have par-allel analog and photon-counting detection chains.The combination of the two signals gives the highlinearity of the analog signal for strong signalsand the high sensitivity of the photon counting forweak optical signals. The integration of both detec-tion mechanisms into a single device avoids groundloops and other problems that make the combinationotherwise cumbersome [26].

A flowchart of the DAQ program is plotted inFig. 4. When starting a data acquisition session, theuser can choose between analog and photon-countingsignals, to be displayed in linear or logarithmic coor-dinates. Users can also zoom in or out to see thedetails or entire profiles. DAAS, built around amulti-channel environment, can collect and display thebackscatter signals in real time. The format for datasaving was designed based on the characteristics oflidar data. The data include the raw signals, radialvelocity, elevation, azimuth, time, laser repetitionrate, and other key parameters.

4. Data Retrieval Methods and Error Analysis

As the Doppler lidar measures the LOS wind veloc-ity, the lidar must be pointed zenith and then off-zenith in order to obtain the wind vector, which areprimarily the horizontal wind components. In thisway, the horizontal wind produces a LOS projectionto the lidar beam, and with appropriate scanningschemes, the three-dimensional wind vector can beinferred. A necessary assumption is the horizontalhomogeneity of the atmosphere (wind and tempera-ture fields as well as aerosol distributions) over thescanned volume and during the time of integration.Vertical homogeneity is not required [27]. Such as-sumption of spatial uniformity has been a commonpractice for other Doppler lidars [28].

A. Data Retrieval Methods

The LOS wind velocity VLOS is the sum of wind com-ponents produced by the horizontal and vertical

Fig. 3. (Color online) DAAS user interface with raw lidar dataprofiles and various control modules displayed.


winds along the lidar beam direction. Thus, the totalLOS wind velocity is given by

VLOS ¼ 0þ VLOS;H þ VLOS;V ; ð1Þ

where 0 represents zero LOS wind, VLOS;H andVLOS;V are the horizontal and vertical wind compo-nents to the LOS direction. Positive LOS wind veloc-ity is defined as radially outward from the lidar. Wedefine RW;i as the wind ratio of signal photon countsof the measurement channel to the reference channelin the ith (azimuth) LOS direction, which containsthe information of Doppler shift. Using a linear ap-proximation that the wind ratio after normalizationto the zero-wind ratio is linearly proportional to theLOS wind velocity, we can express the overall windratio RW;i by

RW;i ¼ R0 −ΔRH;i −ΔRV;i; ð2Þ

where R0 is the zero-wind ratio corresponding to the“0” written in Eq. (1), i.e., the ratio of the measure-ment channel to the reference channel when the LOSwind velocity is zero, and ΔRH;i and ΔRV ;i are thewind ratio changes caused by the horizontal andvertical wind components in the ith LOS direction,respectively. This linear approximation is a goodfirst-order approximation to the reality, as demon-strated in Fig. 3 of [18] where a good linearity exists

between the normalized wind ratio and the LOSwind velocity in the range between �50 m=s. Suchlinearity was also assumed by other groups [15].

Under such linear approximation, we have therelationship

ΔVLOS ¼ dVLOS

dRWΔRW ; ð3Þ

where ΔVLOS ¼ VLOS − 0 and ΔRW ¼ RW − R0.Therefore, the LOS wind velocity can be calculatedfrom the measured wind ratios using

VLOS;i ¼RW;i − R0

R0S; ð4Þ

where RW;i=R0 is the normalized wind ratio as de-fined in [18], and S is the wind measurement sensi-tivity. The sensitivity S is defined as the fractionalchange in the wind ratio per unit change in LOSwindvelocity

S ¼ 1R0

dRW

dVLOS¼ 3:76�

1R0

dRW

dν ; ð5Þ

where ν is the laser frequency and 3:76 MHz=ðm=sÞis the Doppler shift of backscattered light caused by1 m=s LOS wind velocity, which is given by2=λ ¼ 2=532 nm. Here λ ¼ 532 nm is the outgoinglaser wavelength. Therefore, after S is derived from

Fig. 4. Flowchart of the DAQ subsystem.


the experimentally measured dRWdν and R0, we can cal-

culate the LOS wind velocity as follows:

VLOS;i ¼RW;i − R0

3:76�dRW=dν : ð6Þ

Equation (6) is the basic relationship used in thederivation of the LOS wind velocity. Besides the mea-sured wind ratio RW;i, the fidelity of this data retrie-val method relies on how well the two parameters R0and dRW=dν can be determined experimentally. To doso, two assumptions are further made in addition tothe spatial homogeneity assumption mentioned ear-lier. First, the vertical wind assumption—the verti-cal wind averaged over sufficient periods is close tozero. Second, the uniform sensitivity assumption—the slope dRW=dν remains constant from the zenithto the off-zenith directions within a short mea-surement period and over a small scanned volume.Any deviation of the assumptions from the actual at-mosphere conditions will lead to systematic errors inthe retrieved LOS wind. The error analysis providedin Subsection 4.B will impose limitations on theapplicable range of this Doppler lidar with currentretrieval methods.

In practice, the sensitivity S is measured in the ze-nith direction with the Nd:YAG laser tuned to threefrequencies sequentially. As three unknown vari-ables (wind, temperature, and aerosol backscatterratio) are involved in forming wind ratios, the three-frequency measurements allow one to derive S accu-rately. A least-square fit to the three wind ratiosobtained in zenith direction at three predeterminedfrequencies, corresponding to absorption of the seedlaser by the absorption line of 30%, 50%, and 70% onthe I2 line 1109, gives dRW

dν as the slope, and the mid-point wind ratio in zenith direction provides R0. TheS is then calculated from Eq. (5). Based on the uni-form sensitivity assumption mentioned above, thesensitivity S is identical in all directions; thus, the Smeasured in zenith direction can be applied to all off-zenith directions.

The retrieval method of LOS wind velocity de-scribed above is applicable to all measurementmodes for this iodine-filter-based Doppler wind lidar.Because different modes deliver different final windproducts, these measurement modes have differentrequirements on how precisely the three parameters(RW;i, R0, and dRW=dν) have to be determined experi-mentally. For example, the data retrievals for hori-zontal wind in wind profile mode and for thehorizontal LOS wind in PPI mode cancel out theR0 factor, and thus are immune to potential issuescaused by the vertical wind assumption. Therefore,it is necessary to describe the retrieval methods ofthese major wind products as below.

1. Line-of-Sight Wind Velocity in Plan PositionIndicator and Range Height Indicator Modes

The PPI is a well-known radar display that can showthe distribution of LOS components of horizontalwind on a plane with a constant elevation. PPI scan-ning [29] is carried out with the apex of the scanningcone sitting at the lidar scanner, i.e., at a fixed eleva-tion angle, the azimuth angle of the lidar beam isscanned through the entire 360 degrees. RHI scan-ning is used when the elevation of the lidar beam isvaried to cover the majority of the upper hemisphereat a fixed azimuth direction. To illustrate the dataretrieval procedure, a general flowchart includingthe procedure of LOS wind measurements in PPIand RHI modes is plotted in Fig. 5.

In the PPI mode, for a given range, the LOS windratio RW;i plotted as a function of azimuth angleshows a sinusoidal behavior (Fig. 6) in the ideal caseof homogeneous atmosphere. This is because that theLOS wind velocity in the ith direction is given by

VLOS;i ¼ VLOS;i;V þ VLOS;i;H

¼ −VV sinφþ VH cosφ cos�2π360

ðθi − θ0Þ�; ð7Þ

where VLOS;i;V and VLOS;i;H are the LOS wind compo-nents contributed by the vertical wind VV and thehorizontal wind VH , respectively, to the ith LOS di-rection, φ is the elevation angle, θi is the azimuth an-gle in the ith direction (in degrees), clockwise fromnorth, and θ0 denotes the wind direction (in degrees).The linear relationship between RW and VLOS, men-tioned earlier, leads to the wind ratio written as

RW;i ¼ RV þΔR cos�2π360

ðθi − θ0Þ�; ð8Þ

whereRW;i is the wind ratio measured at azimuth an-gle θi, RV is the vertical-wind-induced wind ratio cor-responding to the fitted DC offset, and ΔR is thewind ratio amplitude corresponding to the LOS windof horizontal contributions.

To derive the LOSwind velocity from themeasuredwind ratio, we take the following steps. First, wemea-sure the sensitivity slope dRW=dν in zenith directionwith the lidar running at three frequencies sequen-tially as described above. Second, we conduct off-zenith PPI scans, and then fit Eq. (8) to the measuredoff-zenith wind ratios RW;i versus azimuth at eachheight to deriveRV andΔR. An example of the fittingis shown in Fig. 6. Finally, the LOS wind velocity ofhorizontal contributions VLOS;i;H at each azimuthangle is computed from RW;i, RV , and dRW=dν usingEq. (9), which is derived from Eq. (3):

VLOS;i;H ¼ RW;i − RV

3:76�dRW=dν ¼ΔR cos

�2π360 ðθi − θ0Þ

�

3:76�dRW=dν : ð9Þ


These LOS winds are then interpolated to a polarcoordinate.

Equation (7) indicates that zero horizontal windleads to zero amplitude of the sinusoidal curve, whichprovides an absolute wind calibration to horizontalwinds. Furthermore, it is clear from Eq. (9) that in

the PPI mode, the derived LOS winds of horizontalcontributions are determined by themeasured ampli-tude ΔR of the sinusoidal curve, independent ofthe zero-wind ratio R0. Thus, the LOS wind retrievalin PPI mode is not subject to the vertical windassumption.

In the RHI mode, the wind ratio measured at ze-nith direction when the laser frequency is locked tothe midpoint of the I2 filter edge is taken as the zero-wind ratio R0. After we measure the sensitivity S inzenith direction as described above, VLOS at each ele-vation angle of a RHI scan is calculated from Eq. (4)and then interpolated to a polar coordinate.

2. Wind Profile

The Doppler beam swing (DBS) technique [27] is ap-plied to measure wind-altitude profiles, and rests onthe assumption that the wind fields are horizontallyhomogeneous. By measuring the LOS wind velocitiesin four directions (east, west, south, north) and mea-suring the sensitivity in the zenith direction, we canderive the wind vector components (u, v, w) at thetarget altitudes. The flowchart plotted in Fig. 5 illus-trates the procedure of wind profile measurement.

Fig. 5. General flowchart of all the products.

Fig. 6. (Color online) Example of wind ratio versus azimuthmeasured in PPI scan.


At a chosen elevation φ, the wind ratios obtained infive directions are given by

8>>>>><>>>>>:

RW;Z ¼ R0 þΔRV

RW;E ¼ R0 þΔRH;E þΔRV ;E

RW;W ¼ R0 þΔRH;W þΔRV;W

RW;S ¼ R0 þΔRH;S þΔRV ;S

RW;N ¼ R0 þΔRH;N þΔRV ;N

; ð10Þ

where the second subscripts, Z, E,W, S, andN repre-sent the zenith, east, west, south, and north direc-tions, while H and V denote the horizontal andvertical directions. Taking two opposite directions,east and west, for example, the vertical wind contri-butions ΔRV ;E and ΔRV ;W are equal, while the hor-izontal wind contributions ΔRH;E and ΔRH;W havethe same absolute values but opposite signs. Thus,the vertical components and the factor R0 are can-celled out when taking the difference betweenRW;E and RW;W as shown in

ðRW;E − RW;WÞ=2 ¼ ðΔRH;E −ΔRH;WÞ=2 ¼ ΔRH;E:

ð11Þ

The LOS wind velocity of horizontal contributions inthe east-to-west direction can then be computed as

VLOS;EW;H ¼ ΔRH;E

3:76�dRW=dν ¼ðRW;E − RW;WÞ=23:76�dRW=dν : ð12Þ

The zonal wind profile is given by

u ¼ VH;EW ¼ VLOS;EW;H= cosφ: ð13Þ

Similarly, we obtain the meridional wind profile as

v ¼ VH;SN ¼ VLOS;SN;H= cosφ: ð14Þ

Therefore, the horizontal wind speed and directioncan be derived by Eq. (15) according to the trigono-metric relationship

VH ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiu2 þ v2

p¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiV2

H;EW þ V2H;SN

q;

θH ¼ arctanðu=vÞ ¼ arctanðVH;EW=VH;SNÞ;ð15Þ

where the signs ofVH;EW andVH;SN are used to deter-mine the wind direction (θH) in the correct quadrant.Here we take the definition of wind direction used inthe World Meteorological Organization (WMO). Thatis, θH ¼ 0° corresponds to the wind blowing from thenorth while 90° is for the wind blowing from the east.

Equation (12) indicates that zero horizontal windresults in zero ΔRH;E and ΔRH;W , leading to RW;E ¼RW;W . This fact provides an absolute wind calibrationto the horizontal wind retrieval; thus, the horizontalwind profiles are unaffected by the vertical windassumption.

3. Sea Surface Wind (SSW)

A very attractive product of the Doppler lidar is thetwo- and three-dimensional (2D and 3D) wind field inclear air, which is essential to high-resolution weath-er forecasting. SSW is one of the examples. The pro-cedure of SSW measurement can be found in Fig. 5.

To measure SSW, the lidar is scanned in low-elevation PPI mode. First we obtain R0 and S by ex-perimental measurements using the three-VLOSmethod described in [30,31]. That is, the LOS windvelocities are measured in three directions spreadevenly across a nearly horizontal sector of 120°, i.e.,the adjacent azimuths are separated by 60°. Accord-ing to Eq. (8) and under the vertical wind assumption,

RW;0° ¼ R0 þΔR cosðΔθ0Þ;

RW;�60° ¼ R0 þΔR cosðΔθ0 � 60°Þ:

Thus, R0 can be derived from

R0 ¼ RW;þ60° þ RW;−60° − RW;0°

by recalling the trigonometric relationship that

cosðΔθ0 þ 60°Þ þ cosðΔθ0 − 60°Þ − cosðΔθ0Þ ¼ 0:

For the middle LOS direction, the laser is tuned tothree preset frequencies, enabling the measurementof sensitivity S as described earlier. Assuming thesensitivity is the same for observation points withthe same radius from the lidar, we can derive allLOS wind velocities VLOS in this low-elevation planeusing Eq. (4) within the scanned range. Then we de-rive the horizontal wind for each observation pointfrom measured VLOS using the approach illustratedin Fig. 7.

In the near-horizontal scan, the LOS wind veloci-ties at azimuth θ, θ þΔθ, and θ −Δθ are given by

Fig. 7. (Color online) Schematic diagram of the sea surface windretrieval method.


8<:

VLOS;θ−Δθ ¼ VH cosðα −ΔθÞVLOS;θ ¼ VH cos αVLOS;θþΔθ ¼ VH cosðαþΔθÞ

; ð16Þ

where α is the angle between wind direction and la-ser beam, VH is the horizontal wind speed, Δθ is theazimuth interval. Under the assumption of localhomogeneity, that is, the wind vectors at the sameradius within small local area are uniform [30,31],the 2D wind field can be retrieved by Eq. (17) accord-ing to the trigonometric relationship. In other words,the angle α is calculated from two adjacent LOS windvelocities VLOS;θ−Δθ and VLOS;θþΔθ using Eq. (17a),and then the horizontal wind is computed byEq. (17b):

α ¼ arctan�VLOS;θ−Δθ − VLOS;θþΔθVLOS;θ−Δθ þ VLOS;θþΔθ

· cotΔθ�; ð17aÞ

VH ¼ VLOS;θcos α : ð17bÞ

The wind vectors derived for observation pointswere interpolated to the latitude and longitude gridpoints and displayed according to the criteria set bythe Qingdao Meteorological Bureau. We first decom-pose the wind speed and wind direction of each obser-vation point into the u and v weights. The u and vweights of each grid point are inversely proportionalto the distance between the grid point and the obser-vation points. The weight factor is larger for shorterdistances, and vice versa. Assuming that ui and vi arethe u and v values at an observation point, di is thedistance between a grid point and the observationpoint, andM is the number of observation points thatare used to calculate wind at one grid point, the u andv at this grid point can be derived as

�u ¼ P

Mi¼1ðui=diÞ=

PMi¼1ð1=diÞ

v ¼ PMi¼1ðvi=diÞ=

PMi¼1ð1=diÞ

: ð18Þ

Finally, we calculate the wind speed and the wind di-rection of each grid point based on the trigonometricrelationship given by Eq. (15).

B. Error Analysis

AsVLOS is calculated fromseveralmeasuredvariablesusing Eq. (6), three terms, RW , R0, and dRW=dν, con-tribute to themeasurementerrors ofVLOS.We firstde-rive the general expression of the uncertaintyΔVLOSto give the propagation of errors from RW , R0, anddRW=dν to VLOS, and then discuss the estimated er-rors resulting from various factors.

To derive the VLOS measurement errors, we takethe derivatives of Eq. (6)

ΔVLOS ¼ ∂VLOS

∂RWΔRW þ ∂VLOS

∂R0ΔR0

þ ∂VLOS

∂ðdRW=dνÞΔðdRW=dνÞ; ð19Þ

where ΔRW , ΔR0, and ΔðdRW=dνÞ represent the er-rors in the corresponding measured parameters, re-spectively. As the three error terms are not correlatedand the partial derivatives on the right side ofEq. (19) can be easily derived from Eq. (6), the totalroot mean square (rms) LOS wind error is given bythe error propagation

ðΔVLOSÞrms ¼1

3:76�ðdRW=dνÞ×

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðΔRWÞ2 þ ðΔR0Þ2 þ ½3:76VLOSΔðdRW=dνÞ�2

q;

ð20ÞThe sources of errors can be classified into two ca-

tegories. One is the uncertainties in measured RW ,R0, and dRW=dν, mainly resulting from the photonnoise of the return signals. The errors caused bythese uncertainties are termed the “precision of windmeasurements,” and are random errors. The othersource is the biases, mainly caused by the deviationsof our assumptions from the actual meteorologicalconditions in our case. The resulting errors aretermed the “accuracy of measurements,” and are sys-tematic errors. Here we discuss the precision and ac-curacy, respectively, of the lidar wind measurements.

With respect to the measurement precision, thephoton-noise-induced random errors are estimatedbelow forRW,R0, and dRW=dν. Thewind ratio is givenby RW ¼ NM=NR, where NM and NR are the signalphoton counts of the measurement and referencechannels, respectively. Following the derivation in[5] for photon noise and recognizing that the photoncounts obey a Poisson distribution, the rms error ofthe wind ratio RW is given by

�ΔRW

RW

�rms

¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ 1

RW

qffiffiffiffiffiffiffiNR

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ

2BM þ 2BR

R2W

NR

�1þ 1

RW

�vuuuut ; ð21Þ

whereBM andBR are thebackgroundphoton counts ofthe measurement and reference channels, respec-tively. As expected, the measurement errors decreasewith increasing signal counts and decreasing back-ground counts. The random error ofR0 can be derivedin the same way as RW, and the two error terms aresimilar: ðΔR0Þrms ≈ ðΔRWÞrms. dRW=dν is determinedthrough a linear fitting to the three-frequency mea-surements, as described in Subsection 4.A. The fittingsignificantly reduces the uncertainty of dRW=dν, sothe random error ½ΔðdRW=dνÞ�rms is negligible in com-parison to the other two terms. Therefore, the LOSwind measurement precision is estimated to be


ðΔVLOSÞrms ¼ffiffiffi2

p· ðΔRWÞrms

3:76�ðdRW=dνÞ ≈ffiffiffi2

p

S·�ΔRW

RW

�rms

:

ð22Þ

An example of the LOS wind velocity is illustrated inthe left panel of Fig. 8. The random errors computedaccording toEq. (22), illustrated in themiddle panel ofFig. 8, are less than 1 m=s below 5:5 km, increasewithaltitude as the signal levels decrease, and reach about2:8 m=s at 8 km for an integration of 100 s. It isclear that the measurement precision is mainlydetermined by the signal-to-noise ratio (SNR) and thesensitivity S, as indicated by Eq. (22).

For the data retrieval methods described inSubsection 4.A, R0 and dRW=dν are two parametersthat must be determined experimentally under threeassumptions (the horizontal homogeneity of atmo-sphere, the zero vertical wind, and the uniform sensi-tivity assumptions) mentioned above. Any deviationsof the assumptions from the actual atmospheric con-ditions will result in bias in the measured R0, anddRW=dν, thus affecting the accuracy of LOS wind re-trieval.We examine the roles of these two terms in thedata retrieval. Equation (6) can be rewritten as

VLOS;i ¼RW;i

3:76�dRW=dν −R0

3:76�dRW=dν :

Thus, it is clear that dRW=dν provides the propor-tional coefficient betweenVLOS andRW , whileR0 pro-vides a “marker” for the zero wind calibration. Threeunknowns (LOS wind, temperature, and aerosolbackscatter ratio) are involved in the formation ofthe wind ratio; however, temperature and aerosolbackscatter ratio no longer are included in our dataretrieval equation. Therefore, R0 and dRW=dν arethe equivalent variants of these two unknown param-eters. Using the three assumptions we made, R0 anddRW=dν can now bemeasured in real time by pointing

the lidar beam to zenith direction and operating atthree preset frequencies sequentially as discussedin Subsection 4.A. This data retrieval approach leadsto real-time data retrieval without any model input.Such real-time andmodel-independent retrieval is animportant capability in meteorological applications,and thus is the primarymotivation for our data retrie-val development. However, the validities of the threeassumptions must be examined carefully in order toassess the applicable ranges of our approach.

The horizontal homogeneity of atmosphere is agood assumption for short integration periods andsmall scanned volume. As pointed out by Zhu et al.[32], the sensitivity is actually not uniform fromzenith to off-zenith directions because the LOS windvelocity affects dRW=dν. The systematic error of LOSwind due to the LOS-wind-induced changes ofdRW=dν alone can be derived from Eq. (20) as

ΔVLOS ¼ VLOSΔðdRW=dνÞðdRW=dνÞ ; ð23Þ

where ΔðdRW=dνÞ is the deviation of the off-zenithsensitivity from the zenith sensitivity. It is clear thatthe wind bias is proportional to the percentage of thedRW=dν deviation and the LOS wind itself. As illu-strated in [32], typical ΔðdRW=dνÞ is negative, i.e.,the sensitivity atVLOS ¼ 0 is larger than the sensitiv-ity VLOS ≠ 0, so the retrieved VLOS is biased towardssmaller magnitudes. The sensitivity change withthe LOS wind depends on the backscatter ratio Rb ¼βaer=βmol, where βaer and βmol are the backscatter coef-ficients of aerosols and molecules, respectively. WhenRb is small (∼0:1 or less), the relative deviationΔðdRW=dνÞ=ðdRW=dνÞ is also small (∼3%) even for20 m=s of LOS wind velocity. However, if Rb is largerthan 0.5, then ΔðdRW=dνÞ=ðdRW=dνÞ could be large(>10%) if the LOS wind is over 10 m=s [32]. Whenthe LOS wind is within �5 m=s, ΔðdRW=dνÞ=ðdRW=dνÞ is no larger than 3%. The estimated bias for the

Fig. 8. (Color online) LOS wind velocity measured at azimuth of 90° (east) by the Doppler lidar at 22∶09 LT on 4 March 2008, and thewind uncertainty under different vertical winds. The integration time Δt ¼ 100 s and the spatial resolution Δz ¼ 10 m.


example of LOS wind in Fig. 8 is illustrated as thedashed curve in the right panel of the figure. TheRb profile used in our error estimate is modeled tostart with an initial value of 5 at ground, exponen-tially decayed to 0.1 around 2:5 km, and remainingat 0.1 throughout the higher altitudes. Through a si-milar simulation as carried out by Zhu et al. [32], therelative deviation ΔðdRW=dνÞ=ðdRW=dνÞ was esti-mated and the wind bias was computed accordingto Eq. (23). The wind bias caused by the deviationΔðdRW=dνÞwas less than 1 m=s, comparable or smal-ler than the uncertainty caused by the photon noiseshown in the middle panel of Fig. 8. Therefore, theuniform sensitivity assumption is applicable to thealtitude range above the atmospheric boundarylayers where Rb is nearly negligible if clouds are ab-sent. It is also applicable to the range within theboundary layers if the LOS wind is less than 10 m=s.

TheR0 determination depends on the vertical windassumption.Under quiet atmosphere conditions (e.g.,nighttime), averaging over several minutes results ina vertical wind level close to zero, and our measure-ments in the zenith direction provide a good estimateof R0. However, under convective conditions withinthe atmospheric boundary layers, the vertical windcan deviate significantly from zero for considerabledurations. Thus, the R0 determination can be biasedby the local verticalwind, introducing a systematic er-ror to the LOSwind calibration. According to Eq. (20),the systematic error ofLOSwind causedbyR0 alone isgiven by

ΔVLOS ¼ ΔR0

3:76�ðdRW=dνÞ ¼1S·ΔR0

R0; ð24Þ

where ΔR0 is the deviation of measured R0 from thereal zero-wind ratio. This error could be significant ifthe local vertical wind is much larger than �1 m=sduring themeasurement period.Wedemonstrate thissituation in the right panel of Fig. 8. If ΔR0 is mea-sured in zenith direction when the vertical windVV ¼ �1 m=s, the resulted LOS wind error is givenby the dash-dotted curve in Fig. 8, ranging from∼1 m=s or less above the altititude of 2 km to about2:5 m=s in the first 1 km of altitude. However, ifthe vertical wind is�4 m=s, then the error introducedby ΔR0 will dominate the overall LOS wind error(shown as the solid curve), especially at the lowerheights. R0 and ΔR0 are also dependent on Rb. Oursimulation shows that larger Rb leads to largerΔR0, thus to larger wind bias, for the same verticalwind deviation. In contrast to the situation of the sen-sitivity deviation, ΔR0 could be either positive or ne-gative, depending on the vertical wind deviation, andthis also applies to the wind bias. The analysis wehave presented here indicates that our data retrievalapproach using R0 and dRW=dν is not suitable for theconvective conditions within atmospheric boundarylayers.Basedonourdiscussion ofR0 anddRW=dν, thisreal-time and model-independent data retrieval ap-proach is primarily applicable to altitudes above

the atmospheric boundary layers under quiet andclear atmosphere conditions. Within the boundarylayers, this approach is only applicable when theLOS wind is small and the atmosphere is lessconvective.

It is worth pointing out that the horizontalwind measurements in the wind profile modeand the horizontal LOS winds in the PPI mode arefree of the vertical wind issue. As analyzed inSubsections 4.A.1 and 4.A.2, the vertical wind termsand zero wind terms are cancelled out, and the homo-geneous atmosphere assumption provides an abso-lute zero wind calibration to horizontal winds.Both horizontal winds, however, are still affectedby the sensitivity measurements. The errors in lasercentral frequency and linewidth also affect the mea-surement accuracy and precision. As described in[24,25], the Doppler lidar incorporates a well-controlled and well-calibrated laser system and thuswind errors (much lower than 1 m=s) are well withinthe measurement precision and accuracy discussedabove. Finally, since both R0 and dRW=dν arefunctions of the aerosol backscatter ratio and tem-perature, our data retrieval methods rely on thehorizontal homogeneity assumption of the atmo-sphere within the scanned volume and measurementperiods.

5. Results

The incoherent Doppler wind lidar has been de-ployed in numerous campaigns, and a large amountof data has been collected. Here we highlight somekey results to demonstrate the capability of the lidarand the data retrieval products.

A. Wind Profiles Measured by Doppler Beam SwingScans

Themeasurement procedure of DBS for wind profileshas been illustrated in the general flowchart of Fig. 5.We start with the sensitivitymeasurements at zenithdirection with the laser tuned to three frequencies asdescribed earlier. Then the Doppler lidar beam ispointed off-zenith at an elevation of 60° and swungat four azimuth angles of east (90°), west (270°), south(180°), and north (0°) sequentially to measure theLOS wind velocities in these four directions. The ze-nith and off-zenith measurements form a completedata set for deriving wind profiles. The data are pro-cessed in near-real time and the retrieved wind pro-files are displayed simultaneously. We then canchoose either to move the lidar beam back to zenithto repeat the procedure, or to continue the off-zenithmeasurements, assuming the sensitivity unchanged,within an hour. The integration time for each fre-quency in zenith is 100 s, so it takes 5 min for the sen-sitivity measurements. We also integrate 100 s ineach direction for the four azimuths. Adding the timeto swing the beam, wind profiles are obtained with atemporal resolution of 15 min (or 10 minwithout thesensitivity measurement). Wind profile data areretrieved with a vertical resolution of 30 m.


Two examples of obtained wind profiles are plottedin Fig. 9 for daytimemeasurements, and in Fig. 10 fornighttime measurements. The lidar data (solidcurves) are compared to the sonde data (dashedcurves) of a balloon launched at the Qingdao Meteor-ological Station. The horizontal separation betweenthis station and the lidar site at the OceanUniversityof China is 900 m. The error bars in Figs. 9 and 10 re-present the wind measurement uncertainties causedby photon noise. The differences in wind speed andwind direction between the lidar and balloon dataare, respectively, plotted as ΔVH and ΔθH in Figs. 9

and 10, in order to quantitatively assess the agree-ment between the two sensors. Figure 9 shows thehorizontal wind speed on the left and the winddirection on the right, measured from 09∶13 to09∶28 local time (LT) on 19 March 2008 in Qingdao,China. The wind was approximately northerlythrough the measurement altitudes, but slightlynortheasterly below 3 km, and then shifting to north-westerly above. The measurement uncertainty was∼1 m=s below 3 km and increased to ∼4 m=s at6 km and ∼8 m=s at 8 km. The wind profile (Fig. 10)measured by theDBS scan at 20∶00 LT on19October

Fig. 9. (Color online) Horizontal wind profile measured by the Doppler lidar from 09∶13 to 09∶28 LT local time on 19 March 2008 iscompared to the balloon sonde data taken at the same time. The spatial resolution is 30 m in order to compare with sonde. ΔVH and ΔθHrepresent the differences between the lidar and balloon data for the wind speed and direction, respectively.

Fig. 10. (Color online) Horizontal wind profile measured by the Doppler lidar from 19∶52 to 20∶07 LT on 19 October 2007 is compared tothe balloon sonde data taken at the same time. The spatial resolution is 30 m for both the lidar and the balloon data. ΔVH and ΔθHrepresent the differences between the lidar and balloon data for the wind speed and direction, respectively.


2007 indicated that thewindwasnorthwesterlyat theground, and shifted to nearly westerly at altitudesabove 5 km. The wind speed increased from 10 m=sat the surface to ∼40 m=s near 10 km. The wind un-certainty was about 1 m=s below 3 km, and increasedto ∼8 m=s at 10 km.

The horizontal speed and direction of the wind ob-tained by the lidar generally agree well with thosemeasured by the balloon at Qingdao. The rms differ-ences in wind speed between the lidar and balloondata are computed to be 2:6 m=s and 2:3 m=s forthe daytime and nighttime measurements, whilethe corresponding rms differences in wind directionare 10:7 deg and 10:2 deg, respectively. These re-sults demonstrate that we can measure wind profilesaccurately in clear atmosphere with the Dopplerwind lidar. Such vertical profiles of wind speedand direction are valuable for applications includingnumerical weather forecasting models, rocket launchvehicle safety and load alleviation.

B. Line-of-Sight Wind Velocities Measured by PlanPosition Indicator Scans

As illustrated in Fig. 5, the measurement procedureof PPI scans also starts with the sensitivity measure-ments at zenith direction by tuning the laser to threepresent frequencies. Then the lidar beam is pointedoff-zenith to an elevation angle of 60° and scannedthrough the full azimuth of 0° − 360° with a constantangular speed of 2°=s. Thus, it takes 3 min for onefull-azimuth scan. Such scanning can continue foras many cycles as specified by the user before return-ing to zenith to repeat the sensitivity measurements.During each full-azimuth scan, the DAAS reads thescanner’s azimuth angle in real time, and integratesthe lidar data over an interval of 2°, before saving onelidar profile to computer. After a full-azimuth scan iscompleted, the DAAS processes the data in real timeto retrieve LOS wind velocities for 180 directions ofLOS, as described in Subsection 4.A.1, and displaysthe results as a color contour in polar coordinates.Such LOS wind velocities are obtained and savedfor each full-azimuth scan. As specified by the user,between two zenith sensitivity measurements, theresults of new scans can be averaged with the pre-vious results to produce a smoother color contourfor better visual effects.

Examples shown in Fig. 11 demonstrate the LOSwind velocities measured by the PPI scans at a fixedelevation angle of 60°. The measurements were madeat night, from 20∶41 to 21∶41 LT on 30 December2007 inQingdao, China. The radius in the plots repre-sents altitude, with a maximum displayed altitude of8 km, and white color denoting zero wind velocity.Duration of one full-azimuth scan in PPI mode is3 min; the results are shown inFig. 11(a)witha rangeresolution of 10 m. For the 180 directions, each LOSintegrates for 1 s. Because of the high SNR, the mea-surement precision within 2:2 km appears to be high,even for this very short integration time. To visualizethe wind fields better, we average 7 and 20 full-

azimuth scans and display them in Fig. 11(b) and11(c), respectively. The zero wind line went fromnortheast to southwest below 3 km. The wind wasnorthwesterly below 3 km, and shifted to westerlyat 8 km. The wind speed increased sharply atabout 7:5 km.

Fig. 11. (Color online) LOS wind velocity measured by PPI scansfrom 20∶41 to 21∶41 LT in the night on 30 December 2007. Colorbar scales are from −33 to þ33 m=s. White color denotes zero windvelocity. North is up, and east is to the right. Positive LOS windvelocity is defined as radially outward from the lidar. The radiusin the plots represents the height, and the maximum displayedheight is 8 km. Each circle is separated by 1:14 km. The spatialresolution Δz ¼ 10 m, while the temporal resolutions are (a)Δt ¼ 3 min, (b) Δt ¼ 21 min, and (c) Δt ¼ 60 min.


Figure 12 illustrates a comparison among threewind profiles obtained in different scan modes withdifferent retrieval methods on 30 December 2007.The solid curve shows thewindprofilewith error bars,measured in DBS scan from 20∶00 to 20∶15 LT, withan elevation of 60° and retrieved by the method de-scribed in Subsection 4.A.2. The integration time is100 s for each direction. The wind profile shown bythe dot-dashed curve was also retrieved by the DBSmethod described in 4.A.2, but the data of the four azi-muth directions (east, west, south, and north) weretaken from the PPI scan (shown in Fig. 11). The inte-gration time of each direction is 20 s. Thewind profileshown by the dashed curve was derived from the PPIscan data shown in Fig. 11 using the velocity azimuthdisplay (VAD) algorithm described in [27]. The inte-gration time of each LOS direction is 20 s. As ex-pected, the three wind profiles agree with eachother quite well below 6 km where the VAD methodhas high SNR. The horizontal wind direction θH , asdefined in Eq. (15), is shown in Fig. 12 and confirmsthe interpretation of Fig. 11. The comparison showsthat the normal DBS scan provides the bestmeasure-ments of wind profiles as it has larger detection rangeand higher SNR with higher temporal resolution.

The LOS wind velocities measured in PPI scansduring daylight conditions are shown in Fig. 13.Again, one full-azimuth scan takes 3 min and Fig. 13is anaverage of 8 full-azimuth scanswith the data col-lected from 15∶36 to 15∶59 LT on 30 December 2007in Qingdao, China at a fixed elevation of 60°. The PPIrange scales extend from 0 to 8 km. The LOS velocityis plotted as color-shaded values between −33 and33 m=s with positive defined as moving away fromthe lidar. Note the large apparent shear region tothe ENE. The detection range of LOS wind velocityPPI reaches 8–10 km at night and 6–8 km during

the daytime. It is difficult to obtain LOSwind velocityPPI beyond these ranges due to the low SNR.

C. Line-of-Sight Wind Velocity Measured by Range HeightIndicator Scans

The RHI scan is used to obtain altitude informationabout the LOS wind velocity in the direction of thepredominant horizontal wind. Themeasurement pro-cedure of RHI scan is similar to that of the PPI scan,except inRHImode the azimuth angle is fixed and theelevation angle is scanned. In addition, the R0 is ob-tained from the zenith beam. Examples of the LOSwind velocities measured by the RHI scans from00∶33 to 00∶56 LT on 30 December 2007 in Qingdao,

Fig. 12. (Color online)Wind profilesmeasured by the Doppler lidar in different scanmodes with different data retrieval methods between20∶00 and 21∶41 LT on 30 December 2007.

Fig. 13. (Color online) LOS wind velocity measured by PPI scanfrom 15∶36 to 15∶59 LT during daytime on 30 December 2007.The integration time is 8 s in each direction and the spatial reso-lution is 10 m. The radius in the plots represents the height, andthemaximum displayed height is 8 km. Each circle is separated by1:14 km.


China are displayed in Fig. 14. TheRHIwas producedat the azimuth of 0° and 180° with elevation scannedfrom 20° to 170° in a speed of 2°=s, stepped in 2° in-crements. Maximum measurement height was 8 kmdue to limited SNR. One full RHI scan takes about100 s, and such a single-scan result is shown inFig. 14(a). To visualize the wind fields better, Fig.14(b) exhibits an average over 13 RHI scans.

The zero wind line shown as white color inFig. 14 is along the zenith direction. A low velocitylayer around 2:5 km to the south is clearly evident inthe RHI display. The velocity below 2 km is about15 m=s, decreases to 7–8 m=s between 2 km and2:5 km, and increases to 13–14 m=s between 2:5 kmand 3:5 km.

The PPI and RHI wind measurement capabilitiesdemonstrated above with the incoherent Doppler

lidar may have numerous applications. One of themis to monitor aircraft wing tip vortices generated byair flow over the wing [33]. Heavy aircrafts can gen-erate very high-speed rotating flows that can remainfor several minutes and can be very dangerous toother aircraft landing in similar areas within shortintervals. These wake vortices may be detected byDoppler wind lidar in PPI or RHImodes when a high-er scanning speed is used.

D. Sea Surface Wind Measured in 2008 Beijing Olympics

During a serial campaign in 2006 and 2007 [22], theincoherent Doppler lidar showed its capability in thenear-horizontal wind measurements, due to its highpulse repetition rate, high scanning speed, and thesophisticated DAAS. Such near-horizontal windmeasurements are very important to observe the in-homogeneous wind distribution in small scale. Basedon these successful experiences, this mobile Dopplerlidar was used to measure sea surface wind, provid-ing meteorological service for the sailing competitionduring the 2008 Beijing Olympics from 8–22 August2008 and for the 2008 Paralympics sailing com-petition from 8–13 September 2008 in Qingdao,China [23].

The measurement procedure of SSW is illustratedin the general flowchart of Fig. 5. The sector scanduring the 2008 Olympics covered azimuth from86° to 246° at an elevation of 1°. First, we measuredR0 at the azimuth of 106° and 226°, and measuredthe sensitivity S in the direction of 166° with the li-dar running at three frequencies sequentially as de-scribed earlier. Then the horizontal wind speed anddirection were derived from R0, S, and the measuredwind ratio of each azimuth using the method de-scribed in Subsection 4.A.3. A full scan over the azi-muth range of 160° was achieved in 160 s with acontinuous azimuth scan speed of 1° per second.We averaged three or four adjacent LOS wind veloc-ities, which resulted in 10 min averages, to achieve a

Fig. 14. (Color online) LOS wind velocity measured by RHI scansfrom 00∶33 to 00∶56 LT on 30 December 2007 in Qingdao, China.Color bar scales are from −33 toþ33 m=s. North is to the right andpositive LOS wind velocity is defined as radially outward from thelidar. The radius in the plots represents the height, and the max-imum displayed height is 8 km. Each circle is separated by 1:6 km.The spatial resolution is 10 m, while the temporal resolutions are(a) 2 min and (b) 24 min.

Fig. 15. (Color online) LOS wind velocity measured in sector PPIscan from 08∶26 to 08∶28 LT on 17August 2008. The radius of scanis about 3 km, and the spatial resolution is 100 by 100 m. The x axisand y axis represent longitude (deg) and latitude (deg), respectively.


better SNR. More than 1000 LOS wind fields wereobtained in a wide variety of conditions includingboth day and night operations. The detection rangevaried from 3 km to 8 km, depending on the atmo-spheric conditions. Figures 15 and 16 give examplesof the measured LOS wind velocity PPI and SSW, re-spectively, from 08∶26 to 08∶28 LT on 17 August2008. The colors indicate the wind speed and the ar-rows indicate the wind direction.

The high-resolution sea surface wind measure-ments are crucial for sailing regatta, but there isno instrument available so far to measure it exceptscanning Doppler wind lidar. The wind field mea-surement results were uploaded through a wirelessnetwork to the FTP server of Qingdao MeteorologicalBureau for weather forecasts, with a period of30 min for normal observations and a higher resolu-tion of 10 min for critical applications. The data wereused as the input parameters to the wind forecastmodels.

Based on the above measurement results and erroranalyses, we compare this incoherent Doppler lidarwith current coherentDoppler lidars. Coherent lidarsemploy heterodyne detection of aerosol scattering sig-nals. Since the local oscillator and frequency offset areaccurately known, coherent lidars are in principlebias-free in wind measurements. The heterodyne de-tection technique is very efficient, leading to veryhightemporal resolution (several to tens of seconds) ofwindmeasurements in the near-ground environmentwhere the aerosols are abundant. However, it sufferslowSNRathigher altitudeswhere the aerosol concen-tration is limited. Thus, the detection altitudes of co-herent lidars are quite limited (∼1 km) in relativelyclear atmosphere when the concentration of aerosolsis low. In contrast, the incoherent Doppler lidar pre-sented in this paper relies on the iodine edge-filtertechnique and detects both molecular and aerosolbackscatter signals. Hence, its detection range ismuch larger, even in clear atmospheric conditions.

The iodine edge filter used in the lidar receiver is in-expensive and well calibrated [18]. However, the de-tection principle results in a large portion of thereturn signal being blocked, leading to relatively lowtemporal resolution (∼several minutes) when com-pared to coherent lidars in the aerosol abundant re-gions. Certainly, the temporal resolution can besignificantly improved if higher laser power or largerreceiver aperture is employed. Like other filter-basedincoherent Doppler lidars, this Doppler lidar alsofaces issues of absolute wind calibration. High accu-racy, however, can be obtained when extra frequencyinformation and aerosol backscatter ratio are sought,as we will suggest in Section 6.

6. Conclusions and Outlook

A mobile Doppler wind lidar based on iodine filtershas been developed at the Ocean University of Chinawith full-azimuth-scan and full-elevation-scan cap-abilities as well as versatile measurement capabil-ities. An important feature of the lidar is itssophisticated and user-friendly Data Acquisitionand Analysis System (DAAS), which allows windmeasurements in various operation modes and dataretrieval in real-time. The DAAS places the power ofthe Doppler wind lidar firmly in the hands of itsusers. This lidar has been deployed to measure windprofile, LOS wind velocity in PPI and RHI scans, andsea surface wind, etc., in numerous field campaigns.The detection range of LOS wind velocity in PPI andRHI modes can reach 8–10 km at night and 6–8 kmduring daytime with resolutions of 10 m and 3 min.The detection range of SSW varies from 3 km to8 km, depending on the atmospheric conditions.

The incoherent mobile Doppler wind lidar offersopportunities for meteorological, environmental,and atmospheric research. It can also be utilized todetect and track aircraft wake vortices in PPI andRHI modes, detect wind shears around airports,and in wind field mode, provide wind field measure-ments during space shuttle landing. The wave struc-tures shown in the RHI scans (Fig. 14) indicate thepotential of such a lidar to detect atmospheric waves.In addition, this lidar has shown good potential formaking 3D volume scans of the wind field. Combina-tions of various scan patterns, such as combiningRHI with PPI scans, or RHI scans at several azimuthangles, would provide more information useful forlocal weather forecasts.

We have developed a set of data retrieval methodsto process the lidar data of various operation modesfor the LOS wind velocities and various final windproducts. The basic approach is to retrieve the LOSwind velocity from the measured wind ratio utilizingthe zerowind ratio and the sensitivity that are experi-mentally determined by pointing the lidar beam tozenith and operating at three preset frequencies se-quentially. Theadvantage of this approach is to obtainhigh-resolution and real-time data retrieval withoutmodel input of temperature and aerosol backscatterratio.However, the accuracy of this approach depends

Fig. 16. (Color online) SSWmeasured from 08∶26 to 08∶28 LT on17 August 2008. The radius of the scan is about 3 km, and thespatial resolution is 100 by 100 m. The x axis and y axis representlongitude (deg) and latitude (deg), respectively.


on the validity of three assumptions (horizontalhomogeneity of atmosphere, zero vertical wind, anduniform sensitivity) we made in the data retrieval,which leads us to conclude that this approach is lim-ited to certain applicable ranges. Error analyses showthat this approach is primarily applicable to altitudesabove the atmospheric boundary layers under quietatmosphere conditions (i.e., the averaged verticalwind is close to zero and the aerosol backscatter ratiois small).Within the boundary layers, this approach isonly applicable if the LOSwind is small and the atmo-sphere is less convective. The horizontal wind mea-surements in the wind profile mode and thehorizontal LOS winds in the PPI mode are free ofthe vertical wind issue because the vertical windterms and zero wind terms are cancelled out. Thesedata retrieval methods rely on the assumption of hor-izontal homogeneity of atmosphere that is a good ap-proximationwithin a small scanned volumeand shortmeasurement periods.

Error analyses are useful in assessing the accuracyand precision of the lidar wind measurements. Theyalso shed light on the potential improvements of thisiodine-filter-based Doppler lidar. For example, if thethree-frequency sensitivity measurements are per-formed for off-zenith LOS directions in the wind pro-file mode, then the sensitivity can be determinedaccurately even if local inhomogeneity occurs. Thisthree-frequency technique will lead to the accuratedetermination of wind profiles. This accuracy im-provement sacrifices the precision or temporal resolu-tion to a certain degree if the laser power and receiveraperture remain the same. Furthermore, since bothR0 and dRW=dν strongly depend on the aerosol back-scatter ratio Rb, it is beneficial to tune the lidar fre-quency to the bottom of the iodine absorption lineoccasionally to measure Rb in real time. Such a cap-ability has been demonstrated by this lidar [34]. Withproper algorithms, the information of Rb will help re-duce the errors in themeasurement accuracy. Finally,the correlation lidar technique that tracks distinctfeatures of aerosols or clouds to deducewindvelocitiesmay help to provide wind information in the crossLOS directions for the sea surface wind retrieval.

We sincerely acknowledge Professor Huang Li forhis inspiration and advice in the development ofDAAS. This work was supported by the NationalNatural Science Foundation of China (NSFC) grants40427001 and 60578038, the China MeteorologicalAdministration project Mobile Doppler Lidar, andthe State Key Laboratory of SevereWeather, ChineseAcademy of Meteorological Sciences project Three-dimensional wind retrieval method based on singleDoppler wind lidar. Zhangjun Wang sincerely ac-knowledges the generous support from the ChinaScholarship Council for his study at the Universityof Colorado, Boulder. Xinzhao Chu and ZhangjunWang were in part supported by the National ScienceFoundation (NSF) grants ATM-0545353 and ATM-0723229.

References

1. M. J. Post and R. E. Cupp, “Optimizing a pulsed Dopplerlidar,” Appl. Opt. 29, 4145–4158 (1990).

2. R.M.Banta, L. S. Daray, J. D. Fast, J.O. Pinto, C.D.Whiteman,W. J. Shaw, and B. W. Orr, “Nocturnal low-level jet in a moun-tain basin complex. part I: Evolution and effects on local flows,”J. Appl. Meteorol. 43, 1348–1365 (2004).

3. C. J. Grund, R. M. Banta, J. L. George, J. N. Howell, M. J. Post,R. A. Richter, and A.M.Weichmann, “High-resolution Dopplerlidar for boundary layer and could research,” J. Atmos. Ocean.Technol. 18, 376–393 (2001).

4. S. C. Tucker, W. A. Brewer, R. M. Banta, C. J. Senff, S. P.Sandberg, D. C. Law, A. N. Weickmann, and R. M. Hardesty,“Doppler lidar estimation of mixing height using turbulence,shear, and aerosol profiles,” J. Atmos. Ocean. Technol. 26,673–688 (2009).

5. X. Z. Chu and G. C. Papen, “Resonance fluorescence lidar formeasurements of the middle and upper atmosphere,” in LaserRemote Sensing, T. Fujii and T. Fukuchi ed. (CRC Press, 2005)ISBN: 0-8247-4256-7, 179–432.

6. M. L. Chanin, A. Garnier, A. Hauchecorne, and J.Porteneuve, “A Doppler lidar for measuring winds in the mid-dle atmosphere,” Geophys. Res. Lett. 16, 1273–1276 (1989).

7. B. M. Gentry and H. Chen, “Tropospheric wind measurementsobtained with the Goddard Lidar Observatory for Winds(GLOW): validation and performance,” presented at the Inter-national Symposium on Optical Science and Technology, SanDiego, Calif., USA, 30–31 July 2001.

8. M. J. McGill, W. R. Skinner, and T. D. Irgang, “Validation ofwind profiles measured using incoherent Doppler lidar,” Appl.Opt. 36, 1928–1939 (1997).

9. Y. Durand, A. Culoma, R. Meynart, D. Morancais, andF. Fabre, “Pre-development of a direct detection Doppler windlidar for ADM/AELOUS mission,” Proc. SPIE 5234, 354–363(2004).

10. J. A. McKay, “Assessment of a multibeam fizeau wedge interfe-rometerforDopplerwindlidar,”Appl.Opt.41,1760–1767(2002).

11. J. Q. Liu, J. Zhou, and W. B. Chen, “Boundary Doppler lidarbased onmultibeam Fizeau interferometer,” Proc. SPIE 5653,273–280 (2005).

12. C. Flesia and C. L. Korb, “Theory of the double-edge moleculartechnique for Doppler lidar wind measurement,” Appl. Opt.38, 432–440 (1999).

13. W. T. Huang, X. Z. Chu, B. P. Williams, S. D. Harrell, J. Wiig,and C. Y. She, “Na double-edge magneto-optic filter for Na li-dar profiling of wind and temperature in the lower atmo-sphere,” Opt. Lett. 34 199–201 (2009).

14. W. T. Huang, X. Z. Chu, J. Wiig, B. Tan, C. Yamashita, T. Yuan,J. Yue, S. D. Harrell, C. Y. She, B. P. Williams, J. S. Friedman,and R. M. Hardesty, “Field demonstration of simultaneouswind and temperature measurements from 5 to 50 km witha Na double-edgemagneto-optic filter in a multi-frequencyDoppler lidar,” Opt. Lett. 34, 1552–1554 (2009).

15. J. S. Friedman, C. A. Tepley, P. A. Castleberg, and H. Roe,“Middle-atmospheric Doppler lidar using an iodine-vaporedge filter,” Opt. Lett. 22, 1648–1650 (1997).

16. C. A. Tepley, “Neutral winds of the middle atmosphere ob-served at Arecibo using a Doppler Rayleigh lidar,” J. Geophys.Res. 99, 25,781–25,790 (1994).

17. Z. S. Liu, W. B. Chen, T. L. Zhang, J. W. Hair, and C. Y. She, “Anincoherent Doppler Lidar for ground-based atmospheric windprofiling,” Appl. Phys. B 64, 561–566 (1997).

18. Z. S. Liu, D. Wu, J. T. Liu, K. L. Zhang, W. B. Chen, X. Q. Song,J. W. Hair, and C.-Y. She, “Low-altitude atmospheric windmeasurement from the combined Mie and Rayleigh backscat-tering by Doppler lidar with an iodine filter,” Appl. Opt. 417079–7086 (2002).


19. C. Y. She, J. Yue, Z. A. Yan, J.W.Hair, J. J. Guo, S. H.Wu, and Z.S. Liu, “Direct-detection Doppler wind measurements with aCabannes–Mie lidar: A. Comparison between iodine vapor fil-ter and Fabry-Perot interferometer methods,” Appl. Opt. 46,4434–4443 (2007).

20. C. Y. She, J. Yue, Z. A. Yan, J. W. Hair, J. J. Guo, S. H. Wu, andZ. S. Liu, “Direct-detection Doppler wind measurementswith a Cabannes–Mie lidar: B. Impact of aerosol variationon iodine vapor filter methods,” Appl. Opt. 46, 4444–4454(2007).

21. Z. S. Liu, B. Y. Liu, Z. G. Li, Z. A. Yan, S. H. Wu, and Z. B. Sun,“Wind measurements with incoherent Doppler lidar based oniodine filters at night and day,” Appl. Phys. B. 88, 327–335(2007).

22. Z. S. Liu, B. Y. Liu, S. H. Wu, Z. G. Li, and Z. J. Wang, “A highspatial and temporal resolution mobile incoherent Dopplerlidar for sea surface wind measurements,” Opt. Lett. 33,1485–1487 (2008).

23. Z. S. Liu, Z. J. Wang, S. H. Wu, B. Y. Liu, Z. G. Li, X. Zhang, D.C. Bi, Y. B. Chen, R. Z. Li, and Y. Q. Yang, “Fine-measuringtechnique and application for sea surface wind by mobileDoppler wind lidar,” Opt. Eng. 48, 066002 (2009).

24. Z. S. Liu, S. H. Wu, and B. Y. Liu, “Seed injection and fre-quency locked Nd:YAG laser for direct-detection wind lidar,”Opt. Laser Technol. 39, 541–545 (2007).

25. S. H. Wu, Z. S. Liu, and B. Y. Liu, “Dual-wavelength laserfrequency locking for the direct-detect wind lidar,” J. Mod.Opt. 53, 333–341 (2006).

26. Licel Ethernet Controller Installation and Reference Manual(Licel GmbH, 8 December 2005).

27. C. Werner, “Doppler wind lidar,” in Lidar: Range-resolvedOptical Remote Sensing of the Atmosphere, C. Weitkamp,ed. (Springer, 2005) ISBN 0-387-40075-3, 325–354.

28. B. M. Gentry, H. Chen, and S. X. Li, “Windmeasurements with355 nm molecular Doppler lidar,” Opt. Lett. 25, 1231–1233(2000).

29. K. A. Browning and R. Wexler, “The determination ofkinematic properties of a wind field using Doppler radar,” J.Appl. Meteorol.7, 105–113 (1968).

30. Z. Y. Tao, “The VAP method to retrieve the wind vectors fieldbased on single-Doppler velocity field,” Acta meteorol. Sinica(In Chinese) 50, 81–90 (1992).

31. C. Paulson, “The mathematical representation of wind speedand temperature profiles in the unstable atmospheric surfacelayer,” J. Appl. Meteorol. 9, 857–861 (1970).

32. J.S.Zhu,Y.B.Chen,Z.A.Yan,S.H.Wu,andZ.S.Liu, “Relation-ship between the aerosol scattering ratio and temperature ofatmosphere and the sensitivity of a Doppler wind lidar withiodine filter,” Chinese Opt. Lett. 6, 449–453 (2008).

33. F. Köpp, S. Rahm, and I. Samlikho, “Characterization ofaircaft wave vortices by 2 μm pulsed Doppler lidar,” J. Atmos.Ocean. Technol. 21, 194–206 (2004).

34. Z. G. Li, Z. S. Liu, Z. A. Yan, and J. J. Guo, “Research on char-acters of the marine atmospheric boundary layer structureand aerosol profiles by high spectral resolution lidar,” Opt.Eng. 47, 086001 (2008).


Iodine-filter-based mobile Doppler lidar to make continuous and...

Documents

Transcript of Iodine-filter-based mobile Doppler lidar to make continuous and...