850JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGYVOLUME !0A Critical Review of the Database Acquired for the L~ng-Term Surveillance of the Middle Atmosphere by the French P~yleigh LidarsP. KECKHUT, A. HAUCHECORNE, AND M. L. CHANINService d;4&'onomie du CNRS, Verrieres le Buisson, France(Manuscript received 19 October 1992, in final form 19 April 1993) ABSTRACT The database obtained by Rayl-igh lidars over the south of France is now used for monitoring the middleatmosphere structure and to validate satellite data. For these reasons it is crucial to ensure the quality of thedata. The purpose of this paper is to review all possible sources of errors that could induce random or systematicbias in the temperature measurements, The characteristics of the lidars, the procedures used, as well as the datareduction software arc then reviewed. Comparisons made between the lidar and other available techniques andbetween lidars of different characteristics lead to the conclusion that an accuracy of I K can be attained between$0 and about 70 km depending on the lidar power. The method itself is not affected by drift with time andprovides absolute temperature data without any need of calibration and therefore is one oftbe best instrumentsfor long-term monitoring..1. Introduction Vertical soundings of the atmosphere by Rayleighlidar make it possible to determine an absolute measurement of middle stratosphere (30 kin) and uppermesosphere (80-90 km) temperatures. The high precision of these measurements, the ease of implementation, and the possibility of adapting the integrationperiod and vertical resolution to the temporal and spatial variations of the atmosphere, have resulted in thestudy of a wide range of geophysical phenomena including: gravity waves (Wilson et at. 1990, 1991a,b),tides (Gille et al. 1991 ), stratospheric warmings andplanetary waves (Hauchecorne and Chanin 1982,1983), mesospheric inversions (Hauchecorne et al.1987 ), the oscillation of the 27-day solar cycle (Keckhut and Chanin 1992), climatology (Chanin et al.1985, 1990), the influence of the I l-year solar cycle(Chanin et al. 1987; Keckhut and Chanin 1989), andlong-term trends of anthropogenic origin (Hauchecome et al. 1991 ). Due to the precision of the measurements taken with this instrument (no drift or adjustment) they are often used as references when comparing other measuring techniques (Finger et at. 1993). The possibility of ob taining long-term Wends, in spite of the short-term variance in the temperature, is directly related to the duration and precision of the database, as well as to Corresponding author address: Dr. P. Keckhut, Cenlre Nationalde la Recherche Scientifiqu-, Service D'Aeronomie, BP N 3, Verrieresle Buisson, France, 91371.our knowledge of the natural variations in the atmosphere. As the slratosphere is expected to cool downapproximately by 1 K per decade under the combinedinfluence of O3 depletion and CO2 increase, the absolute precision of these measurements is a fundamental parameter. The measuring instrument must be sensitive enough to discern such variation, and also beable to guarantee precision throughout the entire period. All possible sources of error (random and systematic), and their evolution in time must thereforebe identified. Until now, observations of temperature variationsin the middle atmosphere were carried out by rocketsonde, by the study of falling spheres, and most recentlyby satellite. The data treatment used in analyzing routine rocketsonde measurements performed by a number of American and Soviet bases for over 20 years hasevolved, as have the instruments themselves. The different data intercomparisons and the resulting adjustments have allowed for a decrease in the sources oferrors, but have also introduced discontinuities intothe measurement database. This has made the studyof long-term temperature change very difficult. Atmospheric trends, on a global scale, can only be determined with the help of satellite measurements. These measurements are, however, essentially radio metric and therefore need calibration by ground-based measurements. Rayleigh lidar could be the ideal in strument for taking these measurements, but it is first necessary to consider all possible sources of error and their nature. In this review, the different sources of error that may arise when taking temperature meac 1993 American Meteorological SocietyDECEMnER 1993 KECKHUT ET AL. 851surements by Rayleigh lidar will be studied in detail.The emphasis will be on the effects that these errorshave on absolute measurements and their influence onthe long-term database. It was a great help, in order tocarry out this study, to be able to use several neighboring but different lidar stations.2. Method used To measure the density and then the temperatureof the middle atmosphere, the Rayleigh backscatteringof a monochromatic light pulse by air molccutcs is used.The possibilities of this method were demonstrated forthe first time using mechanically modulated searchlights by Eltcrman (1951, 1953, 1954). Preliminarystudies using laser pulses as monochromatic light wereobtained by Kent and Wright ( 1970); however, theseauthors did not use these measurements to obtain significant geophysical results. It was not until December1978 that Chanin and Hauchccornc (1981) obtainedthe first results based on density and temperature profiles from the lidar set up at the Haute-Provence Observatory (OHP; 44-N,6-E) with the help of a systemdesigned to measure alkali atoms. The method used to measure tcmpcraturcs by Rayleigh tidar has been described in detail in several publications (Hauchecorne and Chanin 1980; Chanin andHauchccornc 1984). As our object here is to attemptan in~depth study into the causes of error and theirevolution over the years, we will only briefly summarizethis method. A short monochromatic light pulse is emitted vertically into the atmosphere by a laser. Data on the vertical structures of the atmospheric layers that arecrossed by the beam are obtained by a time analysisof the number of backscattered photons collected bythe tclcscopc. The number of photons backscatteredby a layer Az at altitude z can be described in the fol~lowing manner: C~NoRq&A(z)T2(zo, z) Topt X [amnm(Z) + t~a(z)]m/XzN(z) = (z- z0)2 , (1)with:Rq:A(z):T(z0, z):crm:Proportionality constant.Overall efficiency of the optics.Number of photons emitted.Quantum efficiency of the photomultiplier.Area of the optical collector.Geometric factor (depending on the over lapping between the telescope field ofview and the laser beam).Atmospheric transmission between altitude Zo and z at the emitted wavelength.Altitude of the lidar station.Molecular cross section of Rayleigh back scattering.n,,,(z):g.(z):a~(r):n.(z, r):Az: Mass atmospheric concentration at altitude Z. Coefficient of Mie backscattering equiva lent to Zraa(r)na(z, r). Mie backscattering cross section for parti cles of radius r. Mass concentration of aerosol particles of radius r at the altitude z. Thickness of the elementary backscattering layer (vertical spatial resolution).m: Number of laser shots.Under the following conditions, it is possible to deducethe molecular density n(z) of the atmosphere by thenumber of photons received N(z):1) The scattering due to aerosols is negligible compared to molecular scattering, ~(z) <~ a,~nm(Z); (2)2) The atmospheric transmission is constant or knownthroughout the entire zone, T(zo, 2)2 = T2 = constant; (3)3) The telescope field of view is large enough to includethe entire volume of the scattered beam, A(z) = 1. (4)In that case, the expression ( 1 ) previously mentionedcould be transformed into the following form: C2(z - Zo)2N(z) n(z) = (5) ~zThe coefficient C2, which represents a normalizingconstant, does not depend on the altitude, but its valuemay not be determined in an absolute manner as itdepends on the power of the laser pulse, the quantumefficiency of the photomultiplier, and atmospheric andoptical transmission. These parameters, particularly theatmospheric transmission, can vary from one shot toanother. To obtain an absolute measurement of thedensity, the C2 constant may be determined by identifying the density measured with a model here, theCIRA 86 [COSPAR (Committee on Space Research)International Reference Atmosphere] at 40 km, orwith the help of a radiosonde measurement at 30 kmgiven by the nearest meteorological station. If one considers that the atmosphere is in hydrostaticequilibrium, a pressure profile can be determined fromthe density profile and an initial value of the pressureP~ at the top of the profile (Jtop): 2tOp P(z) = ~ [n(z)g(z)Az] + P,u(Ztop). (6) gThe top of the density profile is determined when thesignal-to-noise ratio becomes less than 3. Taking theair mass as a constant value, the different parameters,852 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 10air pressure P(z), density n(z), and temperature T(z),can be linked together with the perfect gas law: MP(z)r(z) = 'Rn(z) Ztop M ~ E [n(zi)g(zi)Az] + PM(ztop) = z,=z (7) Rn(z)This equation shows that the temperature measurements do not depend on the normalizing constant C2[ which links together the density n (z) and the numberof photons received N(z)]. Due'to the exponential decrease of the pressure, the influence of the initializationpressure rapidly become.s negligible, and an absolutetemperature measurement can be obtained. This is whymore attention is devoted to the temperature than tothe density parameter in long-term atmospheric studiesperformed by Rayleigh lidar.3. Accuracy of the instrument For any given lidar sounding, the accuracy in determining density is directly related to photon noise.This uncertainty in the number of photons detected isgiven by the Poisson law of statistics where the standarderror is expressed as the square root of this number.Theoretical and experimental values of the variance(Appendix) for the night of 18 January 1990, are givenin Table 1. The relationship between these two quantities shows that the variance is less than 10%. Thisresult is encouraging if one considers that the photoncounting is not absolutely accurate, due to the fact thata certain number of photons too close in time are notdetected, or that photons may cause a double electricalpulse.TABLE 1. Typical theoretical and experimental statistical standarderrors (rms) obtained for one night at the CEL (see text). Theoretical Experimental standard standardAltitude range (km) error (%) error (%) Ratio32-40 0.0070 0.0069 1.0240-48 0.0294 0.0282 1.0448-56 0.1141 0.1053 1.0856-64 0.4209 0.3853 1.09- 64-72 1.690 1.573 1.0772-80 5.210 5.106 1.0280-88 13.80 12.53 1.1088-96 21.91 18.88 1.1696-104 24.24 22.04 1.10104-112 26.23 23.38 1.12112-120 27.49 24.06 1.14120-128 27.87 24.44 1.14128-136 27.83 24.79 1.12136-144 27.17 25.17 1.08144-152 29.44 25.50 1.15 On top of the signal backscattered by the moleculesN(z), there is a parasite signal coming from the skybackground BsB, and from the dark current BpM of thephotomultiplier: S(z) = N(z) + BpM q- BsB. (8)T, he uncertainty of the density measurement is thengiven by the following equation: An(z) _ AS(z) _ ( N(z) + BpM + BSB)1/2 (9) n(z) S(z) N(z)At middle altitudes ( <60-70 km), where the Rayleighbackscattering signal is predominant, and noise comesexclusively from the photon counting, the uncertaintyis only related to the number of photons, but at higheraltitudes the precision of the measurement degradesrapidly as a function of altitude: An(z) _ 1 _ C2(z - z0)2 n(z) [N(z)]t/2 [F/(Z)]1/2 (10)The principal improvement enabling the reduction ofthis source of noise consists of increasing the numberof photons collected by increasing the power of thelaser (No), the surface of the collector (S~.), the detection efficiency (Rq), and the optic transmission (Topt).At higher altitudes, noise coming from the photomultiplier and the background light must also be takeninto consideration. These elements should be reducedas much as possible. For the first source of noise, thisis done by cooling the photomultiplier's photocathodeby means of a water-cooled Peltier effect, thereby limiting the dark current to a value of about 25 pulses persecond. The background light can be reduced by takingmeasurements at nighttime and filtering through spaceand time. The use of the lowest possible reception fieldof view allows for a geometrical separation of the signalfrom most of the noise. This solution, however, is limited very quickly because accurately fitting the telescopefield of view with the scattering volume is a possiblesource of error. The background noise is also reducedby spectral filtering around the wavelength of the received signal. Use of a filter f with a more selectivebandpass/xX/.is often coupled with a less efficient optictransmission T~.. The best solution requires a compromise; if the wings of the filter are well blocked, thenthe choice of the optimum filter may be made takinginto account the effect of the filter's two characteristicson statistical error: an(z) (zxxj.)~/2 -C3-- (ll) n(z)In our case, we have used an interference filter of 10 Aand a field of approximately 10-4-5 X 10-4 rad, whichreduces the noise to less than 500 pulses per second. Daytime measurements have also been carded outusing a Perot-Fabry interferometer, but as these mea1993 KECKHUT ET AL. 853surements are still too irregular to be entered into along-term database, we will not refer to them. The wavelength corresponding to the third harmonicof the Nd:Yag laser (355 nm) could also be considcrcdfor use. The available laser energy is weaker at thiswavelength, and the optics transmission less favorablethan for visible wavelengths, but this is compensatedby a larger cross section of Rayleigh baekscattering andquantum efficiency of the detector (Fig. 1 ). Nevertheless, weaker performance levels are to be expected fromthis wavelength (a ratio of 2 in temperature error) thanfrom a 532-nm wavelength because of a less efficientspectral and spatial filtering of the signal. The transmission of interferential filters for UV wavelengths isnot as good and the alignment of the emission andreccption telescopes is more difficult in the UV andmay end up lobc less accurate.4. Description of the instruments The first measurements were obtained by the lidarinstalled at the Haute-Provence Observatory (OHP;44-N,6-E). This station has provided surveillance ofthe middle atmosphcrc, with a good temporal consistency (a00roximately 100 profiles pet- year), since June1981. A second station, located at the Centre d'Essaides Landes at Biscarosse (CEL; 44-N,1 -W) has helpedto enrich this database since March I986. A third instrument, installed on the French Navy ship the HenriPoincatx~, demonstrates the highest-quality performances and has operated sincc May 1989. Finally, a0.6 0.8 1.0WAVELENGTH ql ~,1 Fin. I. Top: Raylcigh backscattering cross section (dash line),quantum cOficicncy of the photomultiplier (full line), and atmospherictransmission for a clear atmosphere (Cote et al. 1965 ) along the pathlight From the ground to 30 km and From 30 km to the ground(dotted line). Bottom: Rayleigh lidar efficiency as a function ofwuvclcngth.mobile ground station, quite similar in conception tothe ship version is operational since October 1991.These four instruments were designed according to thesame principle and use the same algorithm for datacalculation. The experience, however, of the OHP station 10-year operation, along with the technologicalevolution of the component elements, and most notably the lasers, have benefitted not only the more recently built stations (CEL, H. Poincar~, and the mobilelidar), but also the OH? station, which has been upgraded several times in the last decade. The characteristics of each of the four stations as well as their evolution in time are given in Table 2 and schematic diagrams are given in Fig. 2. The four lidars all use the second harmonic of anNd:Yag laser, which emits a light pulse of around 10ns at the wavelength o- 532.2 nm. This emission isoutside all absorption and resonance bands for atmospheric components above 30 km. Taking into consideration the atmospheric and optical transmissions, thedetection efficiency, the cross section of the Rayleighbackscattering, and the power of the available lasers,this choice allows for the maximal number of photonsto be received. The initial divergence of these lasers ( 10-3-10-4 rad)is reduced by a factor of 10-15 with an afocal opticalsystem. This emitter is placed either in the center ofthe reception mirror (CEL), next to it (OHP), or inthe center of a group of receiving mirrors (H. Poincar~and mobile lidar). The photons backscattered by theair molecules and particles are collected at the focusof these converging mirrors and transmitted to a detection box by means of mirrors (CEL), or by opticalfibers (OHP, II. Poincar~, mobile lidar), in these detection boxes, the light of the backscattered signal iseventually separated into different channels, then filtered from the background light and detected by thephotomultiplier in a "counting mode." The "power" of the lidar is generally defined as thereception area multiplied by the average power of thelaser. This term largely determines the accuracy andrange of the system. A value of approximately 43 m2 Wwas attained for this parameter by the H. Poincard lidar,thereby obtaining temperature profiles with a range ofover 100 km. This is greater than those obtained bythe OHP and CEL lidars, which have a power of approximately 7 m2 W. This increase was achieved usingeight 50-cm-diameter telescopes, constituting a 1.6-m2collecting surface, and six 150-mJ lasers, which alternately operate in groups of 3 at a 60-Hz frequency. At least two receiving channels are necessary to coverthe altitude range from 30 to 100 km. One for thelower atmospheric layers (25-60 km), and the otherfor higher altitudes (40-100 km). Two different methods were used to obtain these two detection channels.The first one consisted of separating the signal in thereception box with a separating mirror (CEL). Recentconfigurations (OHP, H. Poincard, mobile lidar) use ?~o ~o - xo- ~ - xx x ~x x .~ ~ '~x x ~-- ro1993 KECKHUT ET AL. 855doubled Nd:YA~ laser 5~2 nm jUPPER CHANNEL(A)LOWER CHANNEL( J~ ) UPP[R CHANNEL FIG. 2, Schematic diagram of the CEI, lidar (type A) and of theother tidars (type B: OHP, H. Poincard, mobile lidar; which havercspectivcly 2, 8, and 4 received mirrors denoted RM).several reception telescopes with different fields of viewfor the two channels. The time resolution, which is theoretically limitedby the repetition rate of the laser pulses (15-60 Hzdepending on the station), is in fact determined by theinitial integration time of the shots: 3 nm (summationof 3000 to 10 000 consecutive shots) or more reccntly1 nm. The maximal vertical resolution is first definedby the duration of the electronic gate of the photoncounting unit. Counting time is 2 ~s for the first threestations, providing a 300-m resolution. The mobilestation can provide a 75-m resolution; however, upuntil now it has been operated with the same resolutionas the other lidars. It is always possible to degrade thisinitial resolution by grouping together several channels. As the accuracy obtained is the result of a compromise between the integration period and the thicknessof the layers sounded, the chosen temporal and spatialresolutions are adapted to the observation of the different geophysical phenomena under study. In the caseof long-term trend studies, the temporal resolution isdegraded for the entire sounding over one night. Thevertical resolution used is 3 km, which partly eliminatesvariances due to gravity waves. The mean accuraciesobtained at the present time for a 3-km resolution anda 3-4-h integration period are: less than 1 K from 30to 70 km, 3 K at 80 km, and 10 K at 90 km. The samevalues are obtained with the H. Poincard lidar at analtitude level 10 km higher. It is very difficult to theoretically determine the exactperformances of a lidar, as it depends both on the atmospheric transparency, which is never perfect, andthe estimates (often optimistic) of the optical transmission and quantum efficiency of the detectors. Table3 indicates the number of photons received at 60 kmby three of the stations, estimated both by theoreticaland experimental means, by comparing the results overthree nights during which meteorological conditionswere satisfactory. The OHP station, situated at 600-maltitude, seems to be placed at the most favorable location, whereas the H. Poincar~ being on the ocean isthe least favorable one, and results from the CEL station, as it is located on the Atlantic coast, fall betweenthe two. The ratio between the theoretical and experimental value implies that a factor of at least 3 mustbe used in the theoretical calculations (Ci) when sizingthe power of the instrument in order to come as closeas possible to reality.5. Sources of error The study of long-term trends in the thermic ~tructure of the middle atmosphere is currently the subjectof increasing interest, especially as these data are notvery numerous. It is therefore necessary to be awareof the possible sources of error (random and systematic), and their evolution in time in order to evaluatetheir contribution to an apparent trend that would bepurely instrumental. We will analyze the different possible sources of error one by one in the following para~graphs.a. Validity of hypotheses The assumption of a constant value for the meanatmospheric molecular mass is justified by the constantTABLE 3. Number of received photons (in photoelectrons per pulse per microsecond) calculated and measured for a clear atmosphere at 60 km for lidars of OHP, CEL, and H. Poincar~.Number of photon countsLidar Ratio theory/station Theory Measurement Date measureOHP 0.31 0.11 18 January 1991 2.8CEL 0.27 0.08 19 February 1991 3.4H. Poincard 0.89 0.19 22 November 1989 4.7856 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 10mixing ratio of the major gases in the middle atmosphere (N2, 02, and Ar) and the negligible presenceof water vapor. This has been proven up to 80 km.Beyond this altitude, dissociation of 02 must be takeninto account. This percentage of error is relatively low,but a 2% correction must be applied at 100 km and a7% correction at 110 km (CIRA 1990). Temperature profiles are obtained from d~nsity profiles, based on the assumption that the atmosphere isin hydrostatic equilibrium. This is not true locally inturbulent layers, nor in the mesosphere where largefluctuations in the density may occur due to the breaking of gravity waves. But, taking into consideration thetime and space resolutions used by the lidar, we canassume that this does not affect the mean atmosphericdensity. Assuming that the atmosphere is a perfect gas,Jenkins et al. ( 1987 ) calculated that the percentage oferror in the temperature profile due to the presence ofgravity waves could reach a value of 5%; meaning approximately 1 K at more than 60 km for a wave witha 20-K amplitude and an integration during 30 min.When we take measurements corresponding to a 3-kmresolution and integrated during 3-4 h this error becomes negligible. The atmospheric transmission at the wavelength thatwe use, (532 nm) is due to the Mie scattering, the Rayleigh scattering, and absorption by ozone. The attenuation due to the Mie scattering by aerosols, clouds,haze, and fog is the most difficult to estimate. Nevertheless, as the measurements are limited to a zonewhere this scattering does not exist, we can consider itas constant. Molecular scattering and absorption byozone are only taken into account. The atmospherictransmission at altitudes between 30 and 100 km isgreater than 0.995 (Cole et al. 1965 ). The attenuationis therefore very low and may be determined by anatmospheric model. The resulting error is very smalland much less important than from other sources orfrom the photon noise. To initialize the. pressure profile, we must assumethat the values of this parameter at the top of the profileI00 .....................................................zc0,0130 35 40 45 50 55 60 65 70 75 80 85 90 95 100HEIGHTFIG. 3. Temperature uncertainty due to the pressure initialization(full line) compared to typical statistical noise (dash line).(i.e., for the last l0 km) are, on the average equal tothose of the standard atmosphere model (CIRA 1990)for the same altitude layer. The error due to this normalization may be estimated at 15% at the mesopausealtitude (Hauchecorne et al. 1991 ). The calculation ofuncertainty (Hauchecorne and Chanin 1980) showsthat this error becomes rapidly negligible (due to theexponential decrease in the atmospheric pressure) asopposed to the noise statistic, which increases with thealtitude (Fig. 3). As soon as we drop ,in altitude, thenormalizing factor P~(ztop) becomes rapidly negligiblecompared to the pressure at a given altitude. As anexample, the error due to the normalization of thepressure profile at 110-km altitude (which may reacha maximum of 20 or 30 K) decreases by a factor of 10to 100 at the altitudes of 95 and 80 km, respectively.Therefore the temperature obtained, even with thisnormalization, can be considered as absolute and theestimated error consists mainly of the photon noise(section 4). A systematic trend may exist because of the improvements made throughout the years tO the range ofthe instrument. Currently around 60-70 kin, the errordue to normalization may largely be considered asnegligible. Normalization was made at 80 km duringthe first years of measurements. Then errors at 60 and70 km were smaller with a factor of 30 and 3, respectively, than the initializing error due to the uncertaintyof the model in this height range. If we compare theclimatological data obtained from 1984 to 1989 by lidar(Ztop = 100 km) to the CIRA 86 model (Hauchecorneet al. 1991 ), we find differences of+10 K around 80km. This implies that during the first years of operation,systematic variations of 3 and 0.3 K could have beeninduced, respectively, at 60 and 70 km, that is, for alinear increase in range, an apparent trend of 0.3 to0.03 K yr-~. To avoid this source of error, the profilesconsidered in the database are systematically limitedupward to 20 km below the altitude of initialization. Maximal variability at around 70 km generated bythe occurrence of frequent mesospheric inversions hasappeared in the climatology obtained by lidar. Thisresult implies that lidars limited to a 70-km range wouldhave a greater uncertainty as regards normalization bymodel, which may not become negligible at 50 km. The altitude measurement is deduced from the wellknown velocity of light by the delay between the timewhen the light pulse is emitted and the time when thebackscattered photons are received by the photomultiplier. The data acquisition system is initiated by thesignal going out from an optical fiber that has collectedpart of the diffuse light of the emitted laser pulse. Thecorrect synchronization between the laser pulse andthe electronic sampling system could be a source oferror that is dittScult to detect with only one instrumentand could induce a systematic bias in the altitude determination. We have used an electronic system thatcould give a calibrated light pulse of 6 us through anDF('EMBFR 1993 KECKHUT ET AL. 85'7LED, to test the validity of the altitude determinationwithin a resolution of 0.2 us (30 m).b. Dynamic of the signal One of the difficulties encountered when measuringthe atmospheric density on a large vertical scale is thedynamic of the signal. From 30 to 90 km, an exponential decrease in density of more than a power offour is added to a variation of a factor of 10 in thesolid anglc( 1/z2), No detector exists on the marketwith the capacity of taking measurements within sucha range. To solve this problem, two separate detectionchannels with different sensitivities are used in orderto decrease the extent of the mcasurcd range. With thelasers and telescopes available today, the number ofphotons backscattered by the atmospheric layers above50 km is very. low, and only represents an average ofseveral photons per several thousands of shots. Meastm:mcnts of signals backscattcrcd at higher altitudesis possible only with a photomultiplier used in a counting mode, For lower altitudes, ho,a~ver, the numberof backscattered photons would be too high. A fractionof only several percent of the returning signal is therelbre received by a second reception channel that is specially implemented lbr the lower layers of the profile, When the actual number of photons received by thephotomultiplier becomes too high, the phenomenonof saturation takes place and the photons received arenot all accounted for. This is due to the limitation ofthe bandpass ( 100 MHz) of the electronics. This produces a nonlinear response that is well represented byan exponential law with approximately 5% accuracy: NreceivedIf we carefully analyze the correlation between the tworeception channels (Fig. 4), a more complicated lawis necessary to represent the photomultiplier's response: ( Nrecei"ed 2)Ncount ~ Nreeeivect exp Nmax KN,.=~i~a - (13)When the number of photons received is low, they arenot all counted by the electronic counting mechanism.This is due to the fact that the pulses generated are notall of the same amplitude and some are below the detection range. This correction affects less than 1-2 K,and an error ofless than 0.1 K can be anticipated. Foran equal mean power laser, an increase in the tkequencyof the shots decreases the saturation of the detector.Lasers with a 60- and 50-Hz repetition rate were theretbre selected for the two most recent instruments.c. Noise ertraction Despite the use of an inferential filter with a narrowbandpass, and a cooled photomultiplier, a parasite sig351 0,0 0.2 0.4 0.6 0.8 1.0 !.2 1.4 1.6 1.8 2.0 2.2 2.4L0%~R-HE!GHT CHANNEL FIG. 4. Number of photons (photoelectrons per pulse per microsecond) received with the main channel as a function of the numberof photons received with the other channel used for the lower altitudes.This figure gives the degree of saturation of the upper-height channelon 27 November 1989 on the lidar 1t. Poincar& the lower-heightchannel is used as the reference.nat is superposed to the signal backscattered by themolecules N(z). This parasite signal is partly due tothe sky background light BsB (25 cps), and partly dueto the photomultiplier's dark current Bp~ (500 cps). As this noise stems from a random process, it ischaracterized by a signal that is statistically constantin time. If the integration period is long enough, a constant signal superposes the actual scattered signal. Ataltitudes where the signal can be considered negligible,the observation of a signal constant with altitude canbe anticipated: 120<z< 150km, S(z)=Bp~a+BsB. (14)However, at this level of altitude, a time-dependentsignal may often be observed. If the photomultiplier isnot protected, the large scattering from the lower atmospheric layers causes an induced current on thephotomultiplier cathode, which disturbs the measurements throughout the entire altitude range. To resolvethis problem, it is necessary to protect the detector'sphotocathode. Three solutions are possible: 1 ) the use of a geometric shutter (by adjusting thealignment and increasing the distance between emitterand receiver), 2) the use of a mechanical shutter ("rotating pallet" ), 3) the use of an electronic shutter (which controlsthe amplification of the photomultiplier).The first solution has been rejected because it representsa source of error. In the very first experiments the second solution was used, but more recently electronicshutters were preferred to mechanical ones because oftheir greater flexibility, and especially due to the possibility of adjusting the shutter opening time according858 JOURNAL oF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 10to the signal received and the region that is sounded.This method consists of applying an inverted voltageto the photomultiplier's second dynode in order to stopphotoelectron acceleration. A certain amount of timeis necessary, however, before the photomultiplier returns to stable operating conditions ( 100 ~ts or 15 km).Recent improvements in shutter functions have beenmade, and by simultaneously applying a weaker inverted voltage, but on several dinodes, this transitionperiod is reduced to a value of around 10 its, whichcorresponds to a distance of several kilometers. A slightslope bs(z) persists, however, in the signal over 120km: S(z) = N(z) + BpM q- BL q- bs(z). (15)A statistical analysis (Fig. 5) characterizes this signaland shows that it can be normalized by a model usinga parabolic function of time (or altitude): bs(z) = Az + Bz2. (16)But its determination must be made at an altitudewhere the scattered signal can be considered as negligible. As the range of altitude must be as large as possible to determine accurately this function, the analysisis made starting from the channel corresponding to analtitude 20 km beyond the top of the profile ( 100-120km) to the last measured channel (150 km). If thesignal giving th.e noise level is incorrectly estimated,an error may appear in the temperature profile andcause disturbances of up to 20%. This error becomesquickly negligible though at lower altitudes ( 1-2 K 10km below the top of the profile) compared to the errordue to photon noise ( 10 K at 90 kin).d. Geometric factor In order for the atmospheric density to be deducedfrom the number of photons backscattered by mole*10 -48.07.55.0 100 105 110 11.5 11~0 11~5 1~0 135 140 145 10 155 160 ALTITUDE (K~) FIG. 5. Statistical background noise (photoelectrons per pulse permicrosecond) as a function of altitude obtained for an average ofone week of measurement in May 1991 at the CEL. IDefocalization effect '~ Scattering volume tCz)+~iCz) l, j I / Mirror r~ceiver D FIG. 6. Schematic diagram of the geometrical obstruction due toparallax and focal aperture obstruction effect in the focus plane ofthe received telescope.cules, a perfect fit between the telescope field of viewand the illuminated volume in the whole height range(30-100 km) must be guaranteed. The diaphragm (oroptic fiber) of the telescope focus determines the fieldof view at infinity. But in reality, the divergence of thelaser is not the only parameter to be considered, asthere are two instrumental causes of enlargement ofthe focal image of the backscattered volume: the parallax effect and the focal aperture obstruction (nonfocusing effect). A noncoaxial configuration of the emitter and receiver axes leads to a parallax effect. This is made apparent by a transversal movement e(z) of the focal spotat the focus of the telescope, depending on the altitudeof the sounded layer. This is demonstrated in a simplified manner in the following equation: FA e(z) - , (17)with F the focal distance of the telescope, A the distancebetween emission and reception axes, and z the altitudestudied. Decreasing the distance A between the axes limitsthis effect. The ideal solution is a coaxial geometry. Nevertheless, as the layers sounded are not situatedat infinity, their image is not formed strictly at the samedistance of the focal plane, depending on whether thephotons are received from 30- or 100-km altitude. Because of the focal aperture obstruction phenomena,the width of the beam in the plane of the diaphragmplaced in the telescope focal plane is larger than thereal image, and more so if we take into considerationthe low-altitude atmospheric layers (Fig. 6). The following simplified equation demonstrates the imporDECEMBER 1993 KECKHUT ET AL. 859tance of this effect on the diameter of the image l(z)of the scattering volume: AI(z) = D (18) l(z) za'with D the diameter of the collecting surface and a thedivergence of the laser. It is possible to theoretically calculate (Halldorsonand Laderholc 1978) the geometric factor and the neeessary optical components in order to enable this factorto be considered constant throughout the entire rangeof measure (30-100 km). Because of parallax and focal aperture obstructioneffects, as well as other defects that, to a smaller extent,affect the enlargement of the image at the telescope'sfocus (turbulence, optic quality, thermic and mechanical tolerances, etc.), and because of uncertainty inalignment, the diameter of the diaphragm has beenchosen oversized compared to the theoretical diameterfor an infinite range. Oversizing of a factor of 2.5, 2,and 1.5 was chosen for the OHP (noncoaxial), H.Poincard (quasi.coaxial), and CEL (coaxial) stations,respectively. In the first experiments where temperature profileswere obtained, the field diaphragm was greatly oversized in order to guarantee alignment between theemission and reception axes. But, in order to improvethe instrument's performance and to reach higher atmospheric layers, the noise due to the sky backgroundwas partly eliminated by reducing the field of view ofthe receiving telescope. However, this improvementmakes more difficult the good fit between the divergence of the laser beam and the telescope field of viewand the guarantee to keep it so, over long periods oftime; therefore, it could become a source of error. Theinverse dependency of the alignment errors as a function of altitude makes the lower altitudes more sensitiveto this effect, and we can anticipate temperature measurements around 30-35 km to be slightly too high. To ensure this alignment, the most commonly usedmethod is to optimize the signal coming from altitudesof 40-50 km on an oscilloscope. However, the atmospheric, transmission, laser energy, and detection efficiency vary from one shot to another, thereby causingthis operation to become subjective; and it is extremelydifficult to detect variations of less than several percentby eyc. Furthermore if the laser degrades, its divergencemay increase, causing an even greater focal apertureobstruction effect. The quality of alignment can bechecked objectively by using two receiving telescopeswith different fields of view. The channel receiving signals from the lower layers (25-70 km) is less affected:by noise coming from background light and can havea targcr field of view and therefore does not need suchrigorous alignment. The channel used for higher altitudes, however, must have a smaller field of view, andits alignment with the laser beam is more difficult. Theratio of the number of photons in the common operating zone (40-70 kin) of the two channels should beconstant for corresponding altitudes. If the alignmentis incorrect, this ratio is characterized by an inversefunction of the altitude, as demonstrated in the simplified equations of movement of the focal image andits change in size. This new configuration, which isinstalled on the OHP, H. Poincar~, and mobile lidarsallows us to verify the alignment and to measure thegeometric factor (Fig. 7). With this new possibility, avery small field of view, close to the theoretical values,can be used for the "high-altitude" channel, therebyimproving the range and guaranteeing that this errorwilt not exist for low-altitude measurements (30-35km) in the future. Concerning the previous measurements, this error is difficult to quantify on an a posteriori basis.e. The contribution of aerosols Generally speaking, the term coming from the Miescattering due to aerosols t~a(Z) is added to the Rayleighterm ~,,(z) in the lidar equation. The number of photons received is multiplied by a factor R, known as thescattering ratio, and defined as follows: &(z) R = 1 +- (19) ~m(z)In the presence of aerosols, the determination of thedensity of the atmosphere by the measurement ofbackscattered photons is overestimated and is a function of R. The existing error An(z) as a function of Rcan be expressed as follows: AT(z) = An(z) ~ = R - I. (20) T(z) n(z)During periods of major volcanic eruptions, Mie scattering due to particles in suspension can contribute tothe signal detected in backscattering up to altitudes of5O464442 40~38~:~ 36 34 32 30 28 26 100.0 8(J.0 7(~.0 51.0 51~.0 44.0 4~.0 ~T~TUD~. (CM)FIG. 7. Ratio between both channels as a function of the inverse of altitude in a case of a misalinment.860 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 10 38 km, as was the case for several months after the El Chichon eruption in 1982 (Lefr~re et al. 1981 ). This represents a major limiting factor for the downward extension of the measurements. Recently developed solutions consist of using vibrational (Keckhut et al. 1990) or rotational ( Hauchecorne et al. 1992) Raman backscattering of the nitrogen molecule to extend the temperature measurements in the lower part of the at mosphere, even in the presence of clouds and aerosols. These two methods each have their adyantages, but measurements obtained using them are not yet nu merous enough, nor do they extend over a long enough period of time to be able to be used in long-term studies. When trying to accurately determine the altitude atwhich the presence of aerosols is negligible, we runinto the problem of determining R with a precision ofless than 0.3% (Russell et al. 1989). One solution is tocompare simultaneous measurements of Rayleigh andRaman backscattering (Fig. 8). But even though thissolution is not based on any hypothesis on the natureand characteristics of the aerosols, it does not provide- the necessary accuracy. In most cases (from 1984 to1990), at 20 km the scattering ratio R remained lessthan 1.1; the error induced is then around 2 K and theRayleigh profiles were limited safely to 30 km. Afterthe El Chich6n eruption (1982-83), it was necessaryto set the lower limit at 40 and then 35 km. At thisdate (in January 1992) the Pinatubo clouds have beenlocalized below 32 km at our sites. In the mesosphere, the problem of detecting aerosols is more difficult, as the atmospheric variability in this3o282624222O141086420 1.0 , , , . 1,1 1.2 1.3 1. 4BACKSCATTERING RATIOt.5FIG. 8. Backscattering ratio at 532 nm using simultaneous Raman and Rayleigh signal on 9 August 1989. .... I ?0r , ~ ' ~ . , (.. , ~ ~-"7"r'-~, , ~ ~--~- I , \ - ~ - ~ : 60 ~ : / ; ~. ; .. // ~ / . %. qn / '~ ~ ~ :U~ D '~ ~- ~ 4,< 40- ,~ 3 Mai 1983 ~ 30~ ~ ~ I I I I ,~ 0.8 0.9 1.1 1.2 Signal 532 nm / 355 nmFiG. 9. Ratio of the backscattered signals obtained on 3 May 1983at 532 and 355 nm. The error bars represent _1 standard deviation.region is greater and very few instruments are capableof obtaining sufficiently accurate measurements atthese altitudes. However, numerous lidar observationsover the past decades have indicated the possibility ofdust particles being present in the upper atmosphere.These conclusions have been made from the observation of an excess of backscattered light between 60and 90 km when compared with a pure Rayleigh atmosphere as given by the U.S. 1966 atmospheric model(Fiocco and Smullin 1963; Bain and Sandford 1966;Fiocco and Grams 1966; Sandford 1967; Mac Cormicket al. 1967). Latitudes where such effects were observedvary from middle to high latitudes. A few times, thesemeasurements coincided with the observation of noctilucent clouds in nearby areas and could be attributedto such phenomena known to occur frequently at highaltitudes during the summer. Hansen et al. (1989) recently identified strong signals from a 1-km-thick noctilucent cloud at 83.2-km altitude with a scattering ratioof 450 (at 589 nm). These signals were identified simultaneously by Rayleigh lidar and sodium lidar froma high-latitude site (Andoya, Norway; 69-N, 16-E). AtDECEMBER 1993 KECKHUT ET AL. 861latitudes of 45 -N, however, this type of phenomenonis rare (Thomas et al. t984). However, at all latitudesa meteoric source is suggested to explain such a contribution because of the processes of ablation and con- o~densation, or the formation of ion aggregates. For ~measurements taken from middle latitudes, or high zlatitudes out of the summertime, large differences be- ~otween the atmospheric density and the models suggest ~that its interpretation in terms of aerosols could be a~simply explained by ~e use of an inappropriate reference model. The large variability, such as that due ~to a mesospheric inversion, is regularly observed bylidar from our latitudes and could lead one to believethat aerosols are present. Soundings from two wavelengths were carried out in 1983 by using the secondand third harmonic of an Nd:Yag laser (532/355 nm).This made it possible to verity the negligible influenceof particles from 35 to 70 km (Fig. 9). But the energyavailable at 355 nm was too low to provide a definiteanswer tbr the upper altitudes. As at that time, therange of the lidar systems was not reaching as far as otoday and this result was enough to validate measure- ~ mcnts. Duc to the extended range of the new lidars, it a~is necessary to have a second look at this problem, oz Assuming the presence ofnonspheric aerosols, lidar ~ soundings based on the simultaneous recording of the ~ two polarization components of the backscattered light i over 13 nights spread evenly throughout the year did ~ not show any significant structures. Only one case, on 27 November t989, showed two zones where the po larization ratio indicated a significant difference of 10% at 70-85 km (Fig. 10), No definite conclusion may be drawn from this isolated result, but it implies that this type of study should be undertaken anew, as the ab sence ofbackscattering layers in the upper mesosphere has never been verified in an absolute manner and could be a source of error at these altitudes.1.020.980.960.940,90/,,,0'~ Lo ;~ & & & 10 40 is 9'0 0'0 ~U~UD~ (~:~)IA1009'O80?0.6050.4 20 ~'~ ~'0 ~'0 10 ;~ ~'0 0'0 io & 40 4 0'0 do 9'0 ;0 1oo ^LT~TUI)~ (~) FIG. 10. (a) Mean polarization ratio as a function of height for 13nights. The error bars represent _1 standard deviation. (b) Same as(a) for the particular night of 27 November 1989.f Summary of the impact of different sources of error This analysis shows that anticipated instrumentalerrors are inferior to photon noise in most of the heightrange. And the random nature of these sources of noisemakes us confident as to the determination of longterm trends. The different sources of error are localizedin two distinct height ranges: 1 ) The extraction of background noise, the initiabization of the pressure profile, and the photon noisemay disturb temperature measurements at the top ofthe profile, which corresponds to the upper mesosphere.These sources of errors decrease very rapidly, however,in the middle mesosphere. 2) Sources of error in the lower part of the temperature profile are due to the presence of aerosols, thenonlinear correction of the photomultiplier, and thealignment of the emission and reception axes. As theaccuracy of the measurements at this altitude is excellent, it is the most critical zone, as those errors cannotbe neglected compared with photon noise. It is difficultto precisely quantify the effects of these disturbanceson measurements, nevertheless the preceding analysisdemonstrates that we can hope for accuracy of about1 K, but its value will be better given by comparisonwith other instruments. For future uses, a certain number of improvementshave been implemented. A diagram of the configurations of such a lidar as it can be conceived today isrepresented in Fig. 2b. Due to the central location ofthe emitter, the use of a multimirror collecting surfaceincreases the "power" of the lidar and limits the parallax effect. With the use of optical fiber, the receivingtelescopes and the reception box are separated mechanically. With two separate reception channels ofdifferent sensitivities and independent fields of view,it is possible to artificially reduce the dynamic of thesignal to be measured, and to carry out measurements862 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 10of the geometric factor as well as the nonlinear correc- 0tion in the counting process.In the following section, we propose to demonstrate, ~ -20by comparing lidar measurements against those ob- ~ 40rained by other methods, if any instrument bias exists, -in particular below 50 km, where the accuracy of inca- ~ -6osurement is very good. ~ -806. Comparison with other instruments Detection of sources of error by comparison withdifferent instruments is not easy because there are fewinstruments that can take temperature profiles as accurately as a lidar. The first reported comparisonsshowed differences of several kelvins (Hauchecorne andChanin 1980). This was considered satisfactory at thetime, as these measurements were, for the most part,neither taken simultaneously nor from the same site.The number of measurements and instruments todayis still very low, and searching for errors of the orderof 1 K or less remains difficult.~ -100 -120 28 _~.~V \~ -- J - - - ~1~ I : I t I I : ', I I I I I I : : i : I I : I I : ~, : ', : I ', : : 32 37 41 46 50 55 59 64 68 73 77 HEIGHT (KM) FIG. 12. Vertical temperature profiles obtained on 14 December1989 on board the ship H. Poincar~ with lidar given with _+ 1 standarddeviation (full lines) and different sondes on board the same SuperAreas rocket (dashed lines).tained by rocket sondes by Schmidlin ( 1981 ) have beenestimated to be larger than 1.5 K below 60 km. a. RocketsondeSoundings by rocket were, for a long period of time,. the only means of obtaining vertical temperature profiles with a good vertical resolution. The first comparisons made in 1978-79 and in 1981 (Hauchecorne andChanin 1980; Chanin and Hauchecorne 1984) werequite satisfactory considering the distance between thetwo sites where measurements were taken (560 and120 km, respectively). Comparisons carried out morerecently from a single site largely confirm this indication, especially under 60 km (Fig. 11 ). But it has beendiscovered that the validation of lidar soundings byrocket measurements is impossible as the differencesobserved between two sondes on board the same rocketwere as great as those observed between the sondes andthe lidar (Fig. 12). The repeatability of the results ob10090~ 80aO 2O xso I~O 260 2~0 2rio 2rio 2;o 2~o 2~0 2~o 280 TEMPERATURE (Kelvin) FIG. I 1. Vertical temperature profiles obtained on 29 May 1986with lidar (errors are represented by horizontal bars) and a rocketsonde (full line) at the CEL.b. Falling spheres From January to March 1990, during the DYANA(Dynamic Adapted Network for the Atmosphere)campaign at the CEL station, a certain number of vertical temperature profiles were obtained simultaneouslybetween 30 and 90 km by falling spheres (Schmidlinet al. 1991 ) and Rayleigh lidar. When the simultaneousprofiles obtained from the two techniques are compared, the same structures are observed. It must beremembered that the spheres, unlike the lidar, allowan instantaneous measurement that can be slightly displaced compared to the vertical lidar sounding. Statistical comparisons of nine simultaneous temperatureprofiles demonstrated and quantified certain imperfections in the falling-sphere technique. Notably, a systematic bias at 65-75 km, where the speed of the spherepasses from a supersonic speed to a subsonic one, anda strong dispersion occur below 45 km when, due toits low speed, the sphere is affected by vertical winds(Lubken et al. 1992). Between the altitudes of 45 and65 km, however, the two instruments register the samevalues with differences of less than 2 K with a 3-Kstatistical dispersion (95%).c. Satellite measurements It has been possible to make comparisons with temperatures obtained from radiance measurements takenaboard satellites. Certain instruments are of little interest in validating lidar measurements as they providea too low vertical resolution. This is due to the largeweighting function obtained with a nadir pointing instrument. Nevertheless, comparisons with the Stratospheric Sounding Unit (SSU) 27 channel showed satisfactory results considering the variability of thestratospheric winter temperature, and the fact thatP)FCEMBER 1993 KECKHUT ET AL. 863mcasurcments are not rigorously taken simultaneously(Chanin and Hauchecorne 1984). A recent comparisonwith the SSU channel 47X (quite at the upper limit ofthe instrument), shows, however, a systematic differ~ence of approximately 10 K (Aikin et al. 1991 ). The technique of determining temperatures by limbradiance is independent of any exterior calibration, andprovides much better vertical resolutions (~3 km)[Limb Infrared Monitor of the Stratosphere (LIMS)and Solar Mesosphere Explorer (SME)]. Comparisonbetween statistical averages obtained by the OHP lidarand the LIMS measurements from the Nimbus-7 satellite (Remsbe~g 1986 ) showed differences of less than3.5 K between 37 and 64 kin. These differences are ingood agrccment, considering the expected errors of thetwo instruments, the interannual atmospheric variability, and the respective zonal and local nature of theLiMS and lidar measurements. The statistical comparison between lidar measurements and measurements taken on board SME (Clancyand Rusch 1989) during the period of 1982-85 wassatisfactory ( 1 K) during the months of April and July.Noticeable differences at the end of 1983 and duringthe summer of 1984, however, remain unexplained.d. Radiosonde balloon Unfortunately, comparisons between lidar measurements and radiosondes remain limited to low altitudes ( 30-35 km). This is duc to the fact that balloonsoundings rarely culminate above 35 km, whereas lidarsoundings are limited in the lower atmosphere due tothe presence of aerosols (around 30 km). Nevertheless,this altitude zone is a critical one because a number ofsources of crror may be present in the temperaturemeasurements by lidar (alignment of the lidar measurements: parallelism of the emission-reception axes,presence of aerosols, correction of linearity in thecounting process). These comparisons are, therefore,of great interest. Temperatures measured by balloon70060O500400300200100 0-1003 5 7 9 11 13 15 17 19 21 23 25 27 29 31 HEIGHT (Kt~I) F~G. 13. Difference of altitude measurement of a radiosonde between the gcopotential height obtained with the pressure sonde locatedon board and geometrical altitude deduced from a radar trajectog-.raphy.-35-45-50 28,35 29,55 30,75 31,95 33,15 34,35 35,55 36,75 HEIGItT FIG. 14. Vertical temperature profiles obtained on 11 November1990 on board ship tt. Poincard with lidar (+1 standard deviationin full line) and with a radio sounding obtained with a radar trajectography (dash lines represent temperature sonde and its uncertainty).radiosonde are obtained by a thermistance-type captor,which is intrinsically accurate to 0.2-0.5 K. However,comparisons of lidar temperature measurements andthose obtained by balloon soundings at 30 km showgreater differences. The principal drawback of balloonmeasurement is the inaccurate estimation of the altitude. Altitudes are determined from pressure using thehypothesis of hydrostatic equilibrium. The geopotentialaltitude that is calculated depends on the pressuremeasurement and its precision. The pressure sensorhas an inaccuracy of around 0.3 hPa. Given that andthe fact that this parameter varies exponentially in theatmosphere, this inaccuracy can lead to an incertitudeof several hundred meters (Fig. 13), which would meana difference in temperature of I-3 K. An error of severalkelvins in the estimated temperature at 30 km (1-6K), was experimentally obtained from several simultaneous soundings by different manufacturers (Nashand Schmidlin 1987). This includes the error due tothe information supplied by the pressure sensor (3 K),as well as from the measurement in itself (3 K). When the altitude of the balloon is radar monitored,however, a true geometric altitude is obtained that isaccurate to within a few dozen meters. Very fewsoundings have been carried out simultaneously andwithin close proximity by lidar and radar-monitoredballoons. They have only been made by the CEL stationand the ship H. Poincard. All the same, each time thesecomparisons show close concordance at 30-35 km witha difference of never more than 1 K (Fig. 14).e. Comparison between lidars From previous studies, we can see that the validationof lidar measurements is very difficult as no standardinstrument of equal precision exists. One solution forfurther validation consists in comparing temperaturesobtained by two lidar stations at the same moment. Itis obvious that errors in the hypothesis or errors related864 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME l0mO 20 ., . . ~5 ,,/... ~ ~:/ s0 n. :' "" a' ':/ 5 ~ ~ ..:== .== ,.': ~ ~_~.~= 0 / ', '" / ~ '-=" .- '~ ;, ~ ,,,' -10 , i -15 [] ~x -2(~ il;;;;ii~llllllllltl:~llll:lllllllll ',It :I 30 34,539 43,5 4852,557 61,5 66 70,575 79,584 88,593k rll HEIGHT (KM) FIG. 15. Difference between simultaneous vertical temperatureprofiles obtained at the CEL and with the lidar on board ship H.PoincarJ located at Brest (500 km north of the CEL) on 12 June1990 (black square). Full lines represent the +1 standard deviation.to the technique itself will not show in this type ofstudy, but as the stations are never absolutely identical,certain systematic errors, if they exist, should appear,such as: misalignment, incorrect estimations of background light, and incorrect estimation of photomultiplier nonlinearity. Lidar temperature profiles at a distance of 500 kmwere obtained simultaneously at the CEL station andthe OHP or at the CEL and at Brest H. Poincar~. Individual case studies show differences from one day tothe next between the sites that seem to be of a randomnature. An example taken on 12 June 1990, during aseason when wave activity is relatively low, is shownin Fig. 15a. The difference in the measurements carriedout at each station oscillates around a zero value as afunction of altitude. These disturbances, manifestedby the differences between the two profiles, are clearlycoming from the propagation of gravity waves. In thisparticular case, it is possible to discern an increase inthe amplitude of these exponential fluctuations and avertical wavelength of 8 km. This is concordant withthe characteristics of gravity waves. A statistical study of large quantities of measurements is certainly useful to estimate the differences thatmay exist between two lidar stations. It is also necessarythat the profiles be integrated over several hours inorder to eliminate most of the variance due to gravitywaves, which are the principal source of spatial variancefor scales of this type. The statistical study of 169 simultaneous profiles at the OHP and CEL stations (situated at the same latitude), leads to a very good concordance above 45 km considering the noise level atthese altitudes (90%). Below this altitude, a differenceof less than 2 K has been observed (Fig. 16); whichseems to indicate a higher temperature at OHP; thisfact may or may not have a geophysical interpretation.Anyhow, the result of this comparison has led to theconclusion that, if a systematic bias does exist at oneof the stations, it does not exceed this value. Largevariability is present in winter in both series, but thesimultaneous value presents a good agreement. However, systematic variations of several kelvins were observed during two summer periods (Fig. 17 ). With ourknowledge of the horizontal variability at these distances, in summer, we cannot attribute these variationsto a geophysical origin. After examination of differentpossible causes, the differences observed between thesetwo instruments during the summer of 1987 and inAugust 1988, could stem, respectively, from the increase in the divergence of the CEL laser during thisfirst period, and the use of incorrect alignment procedures at the OHP station in the second one. For these reasons, it would be desirable to carry outstudies between lidars at closer distances. This type ofstudy was recently made possible with the CEL lidarwhen the H. Poincar~ was transiting in the Gulf ofGascogne (180 and 100 km from the CEL station).These comparisons of simultaneous soundings integrated over several hours made it possible to eliminateonly part of the variance due to gravity waves. It clearlyseems that the variations in the measurements takenby these two lidars decrease with the distance betweenthem. Differences of less than several kelvins observedbelow 70 km correspond to oscillations that could beattributed to gravity waves. A difference of around 10 Kwas observed between these same two stations above70 km on 10 and 11 July 1990 (Fig. 18 and 19). Thiscan, however, be attributed to a predominant inversionlayer that was clearly visible over one of the two sites.The altitude of this anomaly is coherent with this interpretation. These case studies do not indicate anysystematic errors. The mobile station, currently located at the OHPstation, has allowed us for the first time in November1991 to verify the quite perfect similarity between90 '-.80 ~, - . . ~ ~ .%% - - ~% ! % - % ! % t ! I40 I ! % I30 , ,I -~0 -15 -~0 -$ 7O~ 6o~ so:l :1 I -II%%-I%II !5 10 1'5 20TEMPERATURE DEVIATION (KELVIN~ FIG. 16. Statistical comparison between 169 simultaneous profilesobtained from 1986 up to 1990 at the CEL and at OHP independentlyof the season. Temperature deviation (full line), statistical error with_+ 1 standard deviation (dotted lines), and standard deviation of thetemperature deviation (dashed lines) are represented.DV.('kMt~V:t~ 1993 KECKHUT ET AL. 865265260255250~245 240 ~235~30~'~ 225~ 220 215 2102O5;I :1o CELi 1986 1987 1988 1989 1990 TIME (Year)FIG. 17. Temperature at 35 km as a function of time given by both lidar located at the OHP and at the CEL.1991measurements taken by two lidars situated on thc samcsite every, time that they operate simultaneously. Thisrcsult is extremely important and it justifies the attribution of the previously obscrvcd variations betweenlidar measurements to natural and real atmospherictemperature inhomogeneities at distances of 100 km(Fig. 20).7. Conclusions This study proves the fidelity of its measurements.Furthermore, it appears that the evolution of the instrument has increased its possibilities without compromising accuracy. The validation of Rayleigh lidarmeasurements by the use of other instruments remainsdifficult because no standard instrument of equivalentsensitivity and accuracy exists. Generally speaking,there is a close concordance between different measurements, and comparison between Rayleigh lidarsseems to be a promising solution for checking new instruments. Variations in temperatures observed by thedifferent lidars when not collocated can in general beattributed to horizontal differences of geophysical origin. However, we observed two cases where either abad alignment procedure, or the degradation of thelaser divergence induced a large systematic error. Thusa careful procedure should be followed and a comparison between collocated instruments is advisable. In 50~ 4o ~ /~'~ ~o "',~5 20 -- / t~ ~o~ ~"'"~"" m~ i~.~a~B~ o ~-- -;; ~ - ~~ -lo ~~ -20 -30~ 40-50 ...... 30 ~,5 ~9 ~,5 ~ 52,5 57 61,5 ~ 70,5 75 ~,5 ~ 88,5 93 km HEIGHT (KM) FiG. 18. Same as Fig. 15 on t 1 July 1990 when the ship H. Poincard is lomt~ on the Gulf of Galore 180 km away from the CEL. 2o ha -5 30 34,5 39 43,5 48 52,5 57 61,5 66 70,5 75 79,5 84 88,5 93k m HEIGHTFIG. 19. Same as Fig. 15 on 10 July 1990 when the ship H. Poincar~ is located on the Gulf of Oascogne 100 km away from the CEL.866 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 1090 ; ~ ,' ]legend ~[-- Tohp-Tmobile ][---statistical noiseJ ~ 85 80 75'~ 66 ~ 6o360 40 3 3O -30 -20 -10 0 10 20 30160 180 200 220 240 260 280 300 TEMPERATURE TEMPERATURE (KELVIN) FIG. 20. Comparison between simultaneous profile obtained in the same site with .the permanentOHP lidar and the mobile lidar. The difference between both measurements is represented in theleft side compared with the total statistical noise.the future, with the mobile lidar, it will also be possibleto vary the distances between stations and thereby studythe horizontal structure as well as the temporal evolution of various geophysical phenomena. Climatological surveillance today in the upper atmosphere necessitates the use of an instrument thatcan provide absolute measurements of extreme precision. The Rayleigh lidar, which provides an absolutetemperature measurement and does not need adjustment or external calibration seems like an ideal candidate. Another major application of these results isthe use of a network of such lidars for validation ofsatellite data. Acknowledgments. We wish to acknowledge thecontribution of all the Rayleigh lidar team of the Service d'Arronomie and especially Frederic Fassina,Anne Gamier, Jacques Porteneuve, and Claude Souprayen for their help in improving the Rayleigh lidartechnique. APPENDIX Determination of an Experimental Value of the Variance Due to the Statistical Noise It is possible to experimentally estimate the variancein the measurement due to statistical noise over thecourse of one night, taking into consideration the variance of the signal between two altitudes, separated bya thickness az (300 m). To eliminate the componentthat could be attributed to short vertical wavelengths,we can refine the calculation by taking into equal consideration two consecutive recordings (At = 3 nm).The wave component varies in a coherent manner fromone recording to another, whereas the noise componentis incoherent. The following equation measures thevarianceames,2 which can be compared to the wellknown theoretical valuelYtheor2 .' ~ [A~p(t, z) -- &p(t+At, z)]2, (AI) 4 ~2mes(Z) = ~ t=lwith,a~,( t , z) oc(t, z) -- 1/2[$(t, z -- Az) + S(t, z + Az)] -- S(t, z) + l/2[S(t, z - Az) + S(t, z + Az)] 'and(A2) 1fft2heor(Z) = -- (A3) N(z) ' REFERENCESAikin, A. C., M. L. Chanin, J. Nash, and D. J. Kendig, 1991: Tem perature trends in the lower mesosphere. Geophys. Res. Left., 18, 416-419.Bain, W. C., and M. C. W. Sandford, 1966: Light scatter from a laserDECEMBER 1993 KECKHUT ET AL. 867 beam at heights above 40 km. J. Atmos. Terr. Phys., 28, 543 552.Chanin, M. L., and A. Hauchecorne, 1981: Lidar observation of gravity and tidal waves in the stratosphere and mesosphere. J. Geophys. Res., 86, 9715-9721.--, and ,1984: Lidar studies of temperature and density using scattcring, Map Handb., 13, 87-89. , and , and N. Smires, t985: Contribution to the CIRA model from ground-based lidar. Map Handb.. 16, 305-314. , N. Smires, and A. Hauchecorne, t987: Long-term variation of the temperature of the middle atmosphere at mid4atitude: Dynamical and radiative causes. J. Geophys. Res., 92, 933-941. , A. Hauchecorne, and N. Smires, 1990: Contribution to the new reference atmosphere from ground-based lidar CIRA 1986, Part IL Adv. SpaceRes., 10, 21 t-216.Clancy, R. T., and D. W. Rusch, 1989: Climatology and trends of mesospheric (58-90 kin) temperature based upon 1982-1986 SME limb scattering profiles, d. Geophys. Res., 94, 3377-3393.Cole, A. E., A. Court, and A, J. Kanmr, 1965: Model atmosphere. Handbook of Geophysics and Space Environments, S. L. Valley, Ed, McGraw.-Hilk 196-197.COSPAR International Reference Atmosphere 1990: Part II: Middle atmosphere models. Adv. Space Res., 10.Elterman, L. B., 1951: The measurement of stratospheric density distribution with the searchlight technique. J. Geophys. Res., 56, 509-520. , 1953: A series of stratospheric temperature profiles obtained with the searchlight technique. J. Geophys. Res., 58, 519-530.--~, 1954: Seasonal trends of temperature, density, and pressure to 67.6 km obtained with the searchlight probing technique. J. Geophys. Res., 59, 351-358.Finger, F. G., M. E. Gelman, J. D. Wild, M. L. Chanin, A. Hauche corne, and A. J. Miller, t993: Evaluation of NMC upper strato spheric temperature analyses using rocketsondc and lidar data. Bull. Amer. Meteor. Soc., 74, 789-799.Fiocco, G., and L. D. Smullin, 1963: Detection of scattering layers in the upper atmosphere 60-140 kin) by optical radar. Nature, 199, 1275-1276. , and G. Grams, 1966: Observation of the upper atmosphere by optical radar in Alaska and Sweden during the summer 1964; Teltus. 18, 34-38.Gitle, S. T., A. Hauchecorne, and M. L, Chanin, 1991: Semidiurnal and diurnal effects in thc middle atmosphere as seen by tidar. .I. Geophys. Res.. 96, 7579-7587.Haltdorson, T., and J. Langerholc, 1978: Geometrical form factors for the tidar function. Appl. Opt., ]17, 240-244.IIansen, G~, M. Serwazi, and U. Van Zahn, 1989: First detection of noctiluccnt cloud by planetary waves in the stratosphere. Geo ph.vs. Res. Left., 16, t445-1448.Itauchecorne, A., and M. L. Chanin, 1980: Density and temperature profiles obtained by lidar between 35 and 70 km. Geophys. Res. Lett., 7, 565-568. , and --., 1982: Mid latitude ground-based lidar study of stratospheric warmings and planetary waves propagation. J. At mos. Terr. Phys.. 44, 577-583. , and --, 1983: Mid latitude lidar observations of planetary waves in the middle atmosphere during winter of 1981-1982. d. Geophys. Res.. 88, 3843-3849. , --, and R. Wilson, t 987: Mesospheric temperature inversion and gravity wave breaking. Geophys. Re,'. Lea., 14, 933-936. --, and P. Keckhut, 1991: Climatology of the middle at mospheric temperature (30-90 km) and trends as seen by Ray leigh l idar above South of France. J. Geophys. Res., 96, 15 297 15 309, , P. Keckhut, and D. Nedeljkovic, 1992: Lidar monitoring of the temperature in the middle and lower atmosphere. Appl. Phys., B55, 29-34.Jenkins, D. B., D. P. Wareing, L. Thomas, and G. Vaughan, 1987: Upper stratospheric and mesosphedc temperature derived from lidar observation at Aberystwyth. J. Atrnos. Terr. Phys., 49, 287-298.Keckhut; P., and M. L. Chanin, 1989: Seasonal variation of the 11 year solar effect on the middle atmosphere; Role of the QBO. Map Handbook, 29, 33-38.---, and ---, 1992: Middle atmosphere response to the 27-day solar rotation as observed by lidar. Geophys. Res. Lett., 19, 809 812. --, and A. Hauchecorne, 1990: Stratospheric temperature measurement using Raman lidar. Appl. Opt., 29, 5182-5186.Kent, G. S., and R. W. Wright, 1970: Review of laser radar mea surements of atmospheric properties. J. Atmos. Terr. Phys., 32, 917-943.Lefr~re, J., J. Pelon, C. Cahen, A. Hauchecorne, and P. Flamant, 1981: Lidar survey of the post Mt. St. Helens stratospheric aerosol at Haute-Provence Observatory. Appl. Opt., 20, 1117.Lubken, F. J., W. Hilert, G. Lehmacher, U. Von Zahn, M. Bittner, D. Offermann, F. J. Schmidlin, A. Hauchecorne, M. Mourier, and P. Czechowsky, 1992: lntercomparison of density and tem perature profiles obtained by lidar, ionization gauges, falling spheres, datasondes and radiosondes during DYANA campaign. J. Atmos. Terr. Phys., submitted.Mac Cormick, P. D., E. C, Silverberg, S. K. Poultney, U. Van Vijk, C. O. Alley, and R. T. Bcttinger, t967: Optical radar detection ofbackseattering from the upper atmosphere. Nature, 215, 1262 1263.Nash, J., and F. J. Schmidlin, 1987: Final report of the WMO International Radiosonde lntereomparison. Report 30, WMO, 120pP.Remsberg, E. E., 1986: The accuracy of Nimbus 7 LIMS temperature in the mesosphere. Geophys. Res. Lett., 13, 311-314.Russell, P. B., T. J. Swissler, and U. Van Zahn, 1989: First detection of noctilucent cloud by lidar. Geophys. Res. Left., 116, 1445 1448.Sandford, M. C. W., 1967: Laser scatter measurements in the me sosphere and above. J. Atmos. Terr. Phys.. 29, 1657-1658.Schmidlin, F. J., 1981: Repeatability and measurement uncertainty of the United States meteorological rocketsonde. J. Geophys. Res., 86, 9599-9603.--, M. S. Lee, and W. Michel, 1991: The inflatable sphere: A technique for the accurate measurement of middle atmosphere temperatures. J. Geophys. Res., 96, 22 673-22 682.Thomas, G. E., 1984: Solar Mesosphere Explorer measurements of polar mesospheric clouds. J. Almos. Terr. Phys., 46, 819-824.Wilson, R., M. L. Chanim and A. Hauchecorne, 1990: Gravity waves spectra in the middle atmosphere as observed by Rayleigh lidar. Geophys. Res. Lett., 7, 1585-1588. ., and ,1991 a: Gravity waves in the middle atmosphere ~y Rayleigh lidar, Part 1: Case studies. J. Geophys. Res., 96, 5153-5167. --., and , 1992b: Gravity waves in the middle atmo 'sphere by Rayleigh lidar, Part 2: Climatology. J. Geophys. Res., 96, 5169-5183.