JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 93, NO. D5, PAGES 5267-5276, MAY 20, 1988

Sulfur Dioxide at the Venus Cloud Tops, 1978-1986

LARRYW. ESPOSITO,1 M. COPLEY,R. ECKERT,L. GATES,A. I. F. STEWART,• AND H. WORDEN

Laboratoryfor Atmosphericand SpacePhysics, University of Colorado,Boulder

Ultravioletspectroscopy from the PioneerVenus Orbiter showsa declinein the cloudtop abundance of SO2 from about 100ppb to about 10 ppb in the period1978-1986. A consistentdecline in polar haze has occurred over the same period, with the correlation coefficientbetween these two observablesof r - 0.8. Star calibrationsdetermine the instrumentsensitivity to within 10%, which rulesout the possi- bilitythat thisis an instrumentaleffect. Systematic errors could increase the SO2 abundanceto twicethe inferredvalues in later orbits.Tracking of SO2 featuresand powerspectral analysis give rotation periods for the longer-lived features of 3.6-5.2 days, consistentwith cloud-tracked winds observed at other wavelengths.The behavior of SO2 andpolar haze can be plausibly explained by episodicinjection of SO2 into the cloud top regions,for example,by active volcanism.

INTRODUCTION distance from the spacecraft to the planet and other geo- The discoveryof sulfur dioxide at the Venus cloud tops in metrical factors. Periapsis rose from 150 to 2500 km between 1978 [Barker, 1979; Conwayet al., 1979; Stewart et al., 1979] 1978 and 1986, and the spatial resolution in spectral mode has profoundly affected our understanding of the Venus correspondingly diminished from • 2 to • 75 km. The resolu- clouds and their chemistry [e.g., Winick and Stewart, 1980; tion of the images also degraded, from • 400 to • 800 km. Yung and DeMote, 1982]. Additionally, the variability of the CALIBRATION SO 2 has led Esposito [1984] to infer active volcanoes on For the spectral analysis that follows, it is essential to know Venus. In this paper we describe the data and model pro- cedureswhich establishthe long- and short-termvariability of the instrument sensitivity at the wavelengths of observation. sulfurdioxide on Venus,as observedby the ultraviolet spec- For this purpose we have preflight laboratory calibration; in- trometer on the Pioneer Venus Orbiter. flight bright star observation, as the PVO was en route to, and in orbit around, Venus; and observations of the planet Venus EXPERIMENT DESCRIPTION itself, at locations where the SO 2 absorption is insignificant. These measurements show that the instrument sensitivity has The Pioneer Venus Orbiter (PVO) began observations of been steadily declining since orbit insertion in 1978, becauseof Venus in 1978 [Colin, 1980]. The orbiter payload includes an aging of the detector tubes. The sensitivity decline is a strong ultraviolet spectrometer (UVS), which regularly carries out function of wavelength and the rate of decline is also a func- remote observationsof the cloud top region of the atmosphere tion of time, with instrument degradation somewhat slowed [Stewart et al., 1979; Stewart, 1980; Esposito et al., 1979; Es- down after measures were taken to reduce the light dose re- posito, 1980]. For the purpose of observing SO2, two modes ceived by the instrument during regular observations. are employed:In the spectralmode the spectrometergrating It is especially important to know the sensitivity as a func- is advanced as the spacecraftspin carries the instrument opti- tion of time, since we will make conclusions in the following cal axis across the planet (the PVO is a spin-stabilized space- discussion about temporal variations of the sulfur dioxide craft). The 4-ms integration period is synchronized with the from the spectral observations. 4-ms grating dwell period. The full spectral range of 1556 to The standard method we have used to track instrument 3605• is coveredin two segmentsof 256 words.Each seg- sensitivity is to take spectra of the brightest points in the polar ments takes 1 s to acquire and is collected on an individual haze [Trat, is et al., 1979]. The high albedo of these regions of scan of Venus. To minimize the variation of the illumination, Venus shows that the contribution of any absorber, including emission,and scatteringangles during the acquisitionof each SO 2, is quite small. The brightest spectrafrom latitude nearest segment,this mode is used only near PVO periapsis,when the 60':•N during a periapse observing season are compared to the planet'sdisk subtendsan angle much larger than the arc swept model of Kawabata and Hansen [1975] with an additional out by the optic axis in 1 s. In the wavelength mode the layer of enhanced polar haze [Kawabata et al., 1980] added to grating position is fixed during each scan: this yields a limb- the top of the atmosphere. The optical depth of the polar haze to-limb profile of planetary brightness.By alternating among layer is determined by polarimetry at the time of observation. several wavelengths in this mode, the UVS observations create These polar haze optical depths are given by K. Kawabata et a multicolor image of Venus. The basic data for this study of al. (unpublished data, 1982) and later, by personal communi- SO2 are the brightnessmeasured at wavelengths2 = 207 nm cations from L. D. Travis (1987). and 2 = 237 nm from these two modes of operation, along This method has the advantage that the calibrating data are with the associatedobserving geometry. collected on the same orbits as the spectral mode data. If the The spatial resolution of the observationsdepends on the brightness of the polar haze is constant, this directly gives the rate of decline of instrument sensitivity. However, the absolute •Alsoat Departmentof Astrophysical,Planetary, and Atmospheric sensitivity will depend on the accuracy of the photometric Sciences,University of Colorado, Boulder. model of Kawabata. Nonetheless, even if the brightness is not Copyright1988 by the AmericanGeophysical Union. constant, our analysis will determine the amount of SO 2 at Paper number 8D0082. other points on Venus relative to the polar haze. To reduce 0148-0227/88/008 D-0082505.00 the scatter, a least squares fit to these data gives the sensitivity

5267 5268 EsPOSITOETAL.' VENUS SO2, 1978--1986

207 nm

4.0 ' i i i

3.5-

3.0-

2.5 -

2.0 - o CLOUD SPECTRA z• CLOUD LIMB 1.5 - x ALPHA VIR ß BETA CEN 1.0 ---LEAST SQUARES FIT

0.5 • I I I I I I I I I I I I 0 2OO 400 600 800 1000 1200 1400 1600 1800 2000 2200 2400 26002800

ORBIT NUMBER

237 nm 4.5

4.0

3.5

•- 3.0

m 2.5 o CLOUD SPECTRA z• CLOUD LIMB '•' 2.0 . ß BETA OR I ß ALPHA LUP 0 1.5

• ,•• ßx BETAALPHA CENVIR 1.0 :"/ -----LEAST SQUARESFIT

0.5 I I I I I I I I I I I I I 0 200 400 600 800 1000 1200 1400 1600 1800 2000 2200 2400 26002800

ORBIT NUMBER Fig. 1. Sensitivityloss as a functionof time. Instrumentalsensitivity is comparedto preflightlaboratory calibration, S(LAB).Cloud spectra:observations of 60øNpolar haze,as describedin text.Cloud limb: observationsof clearRayleigh scatteringabove limb. Star calibrationswith individualstars noted. Dashed line is leastsquares fit to calibratingcloud spectra at (a) 207 nm, and (b) 237 nm.

lossas a function of time. These calibratingobservations, de- the PVO spin axis was precessedon two occasionsto observe noted "cloud spectra,"are shown in Figure 1 along with the the bright UV stars Spica (c•Vir) and Hadar (flCen) and to linear fit. accurately measure the responsevariability across the FOV. We have checkedthis calibration method by observations These resultsprovided reliable instrument sensitivitycalibra- of stars.The star observationsare muchmore infrequentthan tions and maps of instrument FOV responsethat allow accu- our spectral observations. This is because the instrument's rate comparison of discrete and extended sources.In addition, field of view (FOV) is normally fixed along a singletrack in they may be comparedto a stellar observationof Spica made the celestialsphere (---30øS eclipticallatitude, where only during cruise between Earth and Venus: this determines the yLup and flOri are available) except for a small wobble. A sensitivityloss over the period 1978-1985. problem in using the star observations to calibrate the instru- The stellar calibration results at the two wavelengthsused ment sensitivity is that stars are point sources,unlike the for SO2 analysisshow excellentagreement with the planetary Venus atmosphere.The instrumentresponse for sourceswhich observationsof the bright polar haze. At 2 = 207 nm the de- underfill the FOV dependssignificantly on the sourcelocation cline calculated from Venus is a factor of 4.7. For the star within the FOV. (That is, the instrumentresponse is not flat observationsthe declineis measuredas 4.34 (7.7% smaller).At across the FOV). We have discoveredthat in general the 2 = 237 nm the Venus observationsgive a decline of 5.0. The normal pointing information suppliedby the spacecraftis not stellar observationsgive 4.81 (3.8% smaller). Given the accu- accurate enough to determine exactly where a star is within racy of each of the calibrating methods (•< 10%), we conclude the FOV. that the instrument sensitivity as a function of time is deter- In order to resolvethese problems,during 1984 and 1985 mined within 10% by two sets of independentmeasurements ESPOSITOETAL.' VENUSSO2, 1978-1986 5269

*.TM FILES *.OA ON LINE

VAXTMIDL.PRO OAIDL.PRO OPEN TMOPEN.PRO FILES TMGET.PRO

DATA DISK IMAGE

YES MAKE.PRO FSP ß .DAT

SEEIT.PRO INFO.PRO LOCATE EXTRACT CREATE FLOCATE.PRC SPECTRA SPECTRAL BACK-UP BLOCKS DATA FILES

FMODEL.FOR

MODSET.LIS

MODEL SET TFPAG•CALCULATEINTENSITY • MODEL FPLOT.PRO•' I COMPARE I DATA AND MODELS SERTBLOCK.PRO• I' LISTBASE.PRO

CREATE PRINT pRINT/ DATA DATABASE BASE DATA / BASE/

Fig. 2. Flowchart of procedureto determineSO 2 abundance.Processing begins (top) with magnetictapes and ends (bottom left and right) with archival of results. For multicolor images (middle diamond) processingcontinues as shown in Figure 3. over the course of the mission and that the brightness of the to the model procedure. The SO2 absorption coefficient polar haze has not changed appreciably over the 8 years of averagedover the instrument band pass (1.3 nm)differs by a observation.The actual sensitivityused is given by the dashed factor of 100 at these two wavelengths[e.g., Thompsonet al., line in Figure 1. 1963]. This differenceallow determination of both SO2 abun- dance and scale height. MODEL PROCEDURE For the spectral data the planetary brightnessis compared Venus spectral and multicolor data arrive on a regular to a set of 10 brightness models, each calculated using the schedule in the form of magnetic tapes from the Pioneer appropriate observing geometry at two wavelengths. The Venus project at the NASA Ames Research Center. Once the model parametersare the SO2 abundanceat 40 mbar (essen- data are in hand at the University of Colorado, the data are tially the cloud top, see Esposito and Travis [1982]) and the handled by a sequenceof procedures that result in multicolor SO2 scale height (SO2 is assumed to have an exponential rectifiedimages of Venus, SO2 abundanceand scaleheight for distribution). The model names, SO2 abundance and scale each spectrum, and SO 2 abundance for every picture element height are given in Table 1. in the rectified images. These results are stored in various For the multicolor images, a subset of seven models (TL, databases by orbit, latitude, Venus phase, etc. The complete AB2, AA2, A2, B2, D2, E2; see Table 1) are compared to the procedure is sketchedin the flowchart shown in Figures 2 and data. These all have the SO2 scaleheight fixed at 2.5 km (the 3. most common value derived from the analyses of the spectral For every observation the brightness at 207 and 237 nm data). The SO 2 abundance has the parameter values 0, 3.125, and the instantaneousobserving geometry are the input values 6.25, 12.5, 25, 20, and 100 ppb at 40 mbar. Fixing the value of 5270 EsPOSITOETAL.' VENUS SO2, 1978-1986

DISKT•B.PRO•

VAXTMIDL BLOCKS calla DISKREC.PRO CALCULATE TIME SEPARATE LOCATEBLOCKS DATA• DISKRISC'PRO• :

TABLINREC GEOMETRIC [STEWART]FILL.PRO RECTIFICATION

.

MODSET.LIS DISKMOD• L.FOR• MODELf .• CALCULATE MODEL SET • IMAGES

DISKGEC•M.PRO

RECGEOM.PRO APOAPSlS GEOMETRY :; CREATEIMAGES •••

MODR•C.PRO ;

; GEOMETRIC ; RECTIFICATION ; OF MODELS

:•...... :..• DISKMAP.PRO'• [STEWART]FILL.PRO COMPARE FCAL.PRO DATAAND CREATE MAKEMAP.PRO MODELIMAGES CONC.MAP MANAGER.PRO CREATE MERCATORMAP DATA BASE

MDBASE.DAT

Fig. 3. Flowchartof processingof multicolorimages. (Top) Data enterfrom Figure 2 andprocedures end (bottom) with archival of results. the scaleheight at 2.5 km substantiallyreduces the compu- are variability of SO2 in the observed region which is too tational effort required in reducing this data set. As can be small to be resolved, random data errors, and the limited seenbelow, the derivedSO 2 abundanceis in goodagreement number of modelsto be compared.Of 1545 spectrataken by betweenthe multicolorimages and the spectra. the UVS in the first 2814 orbits (8 years)a satisfactoryfit is The modelingprocedure is similarto that usedby Esposito found for 1110 (72%). For each of the 157 orbits containing [1980]. SeeTable 2. The reflectedbrightness is calculated(in- spectra,we averagethe derived values (which are the SO2 cluding multiple scattering,vertical inhomogeneity,and the abundancesfrom the best fit models)to give global meansfor reported observing geometry) by the method of Markov SO2 abundance.Global meansare also calculatedby averag- chains [Esposito, 1979]. The best fit model is the one whose ing all the picture elementsfrom a multicolor image that are combinationof abundanceand scaleheight minimizesthe sum successfullymodeled (i.e., within 10% error). Combining the of the squares of the difference between model and data at the spectral data with the 195 multicolor images gives a total of two wavelengths.If the root-mean-square(rms) differenceis 352 orbits, for which we derive individual global mean greater than 10% of the observed brightness,no model fit is amounts of SO 2. These meansare plotted in Figure 4. deemedsatisfactory. The possiblecauses of unacceptablefits Figure 4 shows substantial scatter. We believe this scatter is ES?OSITOETAL.i VENUS SO2, 1978--1986 5271

TABLE 1. Models Used for Spectral Analysis 2. This SO 2 is rapidly convertedto new, small aerosolsof H2SO,•, seenas a haze and especiallybright polar regions. SO2 Abundance 3. These aerosols grow and fall out into the main cloud at 40 mbar, Scale Height, Model Name (ppb) km deck, which would give rise to decade-long periods during which neither small aerosolsnor SO 2 are seen. TL 0.0 Esposito and Travis [1982] showed that there is a negative AB2 3.125 2.5 correlation between the submicron haze at any given instant and the amount of SO2. The natural interpretation of this is a AA2 6.25 2.5 conservation of sulfur atoms as they are exchanged between A1 12.5 5.0 the gas (SO2) and the haze aerosol (H2SO4). Since the parti- A2 12.5 2.5 cles are produced from the SO2 gas, this would naturally ex- A3 12.5 1.67 A4 12.5 1.25 plain the observed long-term positive correlation; as the global amount of SO2 declines, so should the production of B1 25. 5.0 haze aerosols. Indeed, K. Kawabata et al. (unpublished data, B2 25. 2.5 B3 25. 1.67 1986) show the amount of submicron haze decreasedby an B4 25. 1.25 order of magnitude over the first 1000 days of the Pioneer Venus mission. D1 50. 5.0 D2 50. 2.5 Figure 6 compares our Pioneer Venus UVS data from the D3 50. 1.67 first 1600 orbits to seasonallyaveraged measurementsof polar D4 50. 1.25 haze optical depth from the Pioneer Venus UVS CPP (K.

E1 100. 5.0 Kawabata et al., unpublished data, 1986). The correlation is E2 100. 2.5 striking: both measures show a strong decline since the be- E3 100. 1.67 ginning of mission, leveling off at more historically typical E4 100. 1.25 values after 5 years. The correlation coefficient between these two measuresis 0.8. This strongly supports the hypothesisthat the polar haze is formed of small, recently created aerosols of real and indicates natural variability of Venus cloud top con- H2SO 4. ditions (similar to "weather" on Earth). The scatter can be reduced by averaging the inferred abundances from the spec- Cloud Top Rotation Rates tra over Pioneer Venus observational seasons (about 100 Tracking of cloud features has provided detailed measure- days). Each of these seasons represents a contiguous period ments of the wind velocity at cloud top altitudes [e.g., Travis when the planet disk visible to PVO is illuminated at perapse et al., 1979; Rossow et al., 1980; DelGenio and Rossow, 1982]. or apoaspe observing times. See Figure 5. We have examined the rotation rates of SO 2 featuresto com- pare with these with wind measurements. RESULTS

Abundanceof S02 TABLE 2. Model Atmospheres for Venus SO2 Abundance

From Figures 4 and 5 it is obvious that the amount of Atmospheric Property Adopted Characteristics sulfur dioxide visible at the Venus cloud tops has declined by about 1 order of magnitude since the start of the mission in Clear gas above clouds 5 mbar Aerosols 1978. This decline has two parts. A rapid decline in the first total optical depth 20.0 ratio of Rayleigh to aerosol 400 days of the mission has a characteristic decay time (l/e) of extinction, 0.035 at about 200 days. Following this is a slower decline, apparently 3650A still in progress, with a decay time of about 2000 days. Given physical characteristics: (1) the uncertainties in the instrument calibration, this latter spherical; (2) refractive index1.46 at 2650•;* (3) decay rate is marginally consistentwith a constant SO 2 abun- effective radius 1.05 dance and a 10% overestimate in the instrument sensitivity (4) effective variance decline.We note that the current abundanceof SO2 is consis- 0.07; (5) aerosol single tent with upper limits set before the Pioneer Venus encounter scattering albedo 0.98; [e.g., Owen and Sa•tan, 1972; Cruikshank and Kuiper, 1967]. uniformly mixed with gas SO 2 scale height from Table 1 mixing ratio at 40 mbar Correlation With Polar Haze from Table 1 maximum mixing ratio: 100 Travis et al. [1979] have noted several other unusual as- times values from Table pects of the Venus atmosphere at the time of PVO encounter. 1 The cloud images from the PVO Cloud Photopolarimeter uniformly mixed below 150 (CPP) clearly show bright polar regions. Polarimetry shows mbar Second absorber altitude 75 mbar that a haze of submicron particles covers the entire planet and optical depth 0.2 is especiallyevident over the poles. This haze of small particles single scattering albedo 0.0 can account for the bright polar regions. The last major oc- Clear gas below clouds total optical depth 64.0 currence of bright polar caps in 1959 was also associatedwith single scattering albedo submicron particles at the cloud tops [Dollfus et al., 1979]. It given by maximum SO2 mixing ratio is possible to understand these repeated occurrences in a straightforward way. A simple explanation is as follows. *Extrapolated to shorter wavelengths by the method of Hansen 1. SO2 is episodicallyinjected above the Venus cloud tops. and Hovenier [1974]. 5272 ESPOSITOET AL.' VENUSSO2, 1978-1986

DISK MODELING & FSPECTRA RESULTS WITHIN 10% ERROR lOO

E 80 o

"' 60

d z o o 40

o

z ++ •-+ .i- •: 20 • + + + •, + +

o ß • t;-•- 0 500 1000 1500 2000 2500 3000

ORBIT NUMBER Fig. 4. Mean SO2 amountfor orbitscontaining spectra (sunlit periapsis, crosses) and multicolorimages (sunlit apoapsis, diamonds).

The model procedure (described earlier, see Figure 3), is The procedure for using consecutive orbits is as follows. used to create a map for the portion of the planet which was SO: concentration maps are made from 4 or 5 consecutive viewed which representsthe SO2 concentrationfor the time of orbits, and then are converted to mercator projections with observation. A data base of these maps is used in two ways to Venus latitude as the vertical axis and Venus local solar time determinerotation rates from SO2 abundancevariations. The (LST) as the horizontal axis. This procedure usesthe geometry first method involves looking for features which persist for data obtained from the spacecraft. Each map is discretized more than 1 Earth day (one Pioneer Venus orbit) using a into pixels which are 12 min of LST wide and 3ø of latitude sequence of maps made for 4 or 5 consecutive days which tall. The map is then 120 pixels wide and 60 pixels tall, where show a complete rotation of the cloud tops. The second takes 120 pixels of LST correspond to one complete rotation of the the discrete Fourier transform of the sequenceof global SO2 cloud top. averages in Figure 4. If a SO: feature persistsfor more than 1 day it should be

SEASONAL AVERAGES OF FSPECTRA & DISKMODELING DATA 100 .....

80

60

40-

z

LU 20-

0 I I I I I 0 500 1000 1500 2000 2500 3000

ORBIT NUMBER Fig. 5. Mean SO2 amount averagedover each observingseason. Asterisks, sunlit periapsis,diamonds, sunlit apoapsis. Error bars show standard error of the mean. ESPOSITOET^L.' VENUSSO2, 1978-1986 5273

100 methods, finally choosing the relatively simple procedure pro- .,• posed by Deeming [1975]. In choosing this method, however, i- • PolarHaze _ x 80 ß ance we necessarilylose the possibility for the somewhat rigorous o error analysis available in the method of Ponman [1981]. o Ponman's method, however, works better for data which has "- 60 -

o no regularity in the time spacing, while our data has inherent regularity due to different factors, among which is the weekly E 40 . o schedulingof spacecraftcommands for the Pioneer Venus Or- biter. Ponman [1981] deals with the general caseof data which 20 is irregularly sampled, weighted, and integrated over time (possiblydifferent for each data point). Deeming [1975], however, assumes instantaneous observa- 0 I I I 0 500 1000 1500 2000 tions, which is better suited to our data: although the image of DAYS SINCE ORBIT INSERTION the planet is taken over •-,2 hours, a rectification procedure (part of the normal processingin Figure 2) is applied to each Fig. 6. Intercomparisonof derived SO 2 (solid line, this work) and image to create another image corresponding one point in derived polar haze optical depth (dashed line) from PVO Cloud Photopolarimeter (L. D. Travis, personalcommunication, 1987). time, giving an "instantaneous" observation. Deeming [1975] defines a discrete Fourier transform F(o) of f[t(k)], for arbitrary time spacing t(k) of N data points, as possibleto track the feature on successivedays as the cloud follows:

top rotates. To searchfor persistentfeatures, overlapping ad- N jacent mercator maps are shiftedby integer numbers of pixels F(v) = • fEt(k)] exp [2•zt(k)] along the LST axis, relative to the previous map. Between k=l adjacent maps (e.g. first and second, second and third, third The spectral window and fourth, but not fourth and first), the overlappingportions of the maps are extracted and a least squaresfit determines N the correlation of SO 2 abundancein the overlapping portion W(v)= y'. exp [2r•ivt(k)] k=l and the proportionality constantbetween the measuredSO 2 on consecutive days. For each value of shift, i.e., from a shift of also known as a "data window," contains the "pathology" of 1 pixel to the shift value at which the maps no longer overlap, the data spacing, including aliasing and related effects. F(v) a correlation coefficient and a proportionality constant are represents the convolution of the "true" Fourier transform of calculated.The total range of periods sampled is 0.5 days to the function which the data represent with the spectral 120 days, which we considersufficient for observingcloud top window. motions of SO2 features.The shift value which givesthe high- The problem of aliased peaks in the discrete Fourier trans- est correlation with a proportionality constant closestto unity form places a constraint on the maximum frequency for which determines the rotation period. reliable information can be calculated. Aliasing occurs at fre- For PV orbits 525-528, the derived cloud top rotation quencies above a Nyquist frequency which is calculated from period is 4.8 days and the correspondingconstant of pro- the data spacing. For equally spaced data there is only one portionality was 0.98. This is close to periods calculated by Nyquist frequency: v0 -- 1/2At, where At is the constant time Rossow et al. [1980]. However, other consecutive orbit se- spacing. The results of aliasing are a Fourier transform at quencesgive divergent values from 2 to 6 days and correlation • > u0 which has a significant contribution from the transform coefficientssubstantially below 1.0. Thus these results do not at t;- t;0. For equal spacing, this aliasing is complete, and seemto be definitive, but do not contradict previous measure- frequenciesabove the Nyquist frequency are indistinguishable ments. from their corresponding frequenciesbelow the Nyquist fre- For the Fourier transform method, we use the global quency.The data spacingof the available SO 2 concentrations, average SO 2 concentration for each available orbit between although unequal, is in integer multiples of a smallest unit orbits 474 and 574, yielding 27 data points. Linear trends in spacing (1 orbit -- 1 day). This leads to the possibility of sev- the data are removed by subtracting the mean and first eral Nyquist frequencies,with the maximum as the Nyquist moment, determined by a least squares fit to the data. The frequencyfor equal spacing, 1/2 orbits-• This is then the same procedurewas applied to the maximum value of SO2 highest frequency calculated in the discrete Fourier transform. concentration in any latitude bin for each orbit. Those maxi- Normalization of the spectral window represents another mum SO 2 data points have the same time distribution as the procedural difficulty. Ideally, the spectral window should be global SO2 averages. square integrable to unity. However, since this is only possible Because the data were not obtained every day, we have in an approximate sense, it is easier and more useful to "nor- unequal spacing in the time domain. We note that the actual malize" the spectral window by setting W(0) - 1. This is done data spacing is integer numbers of orbits and is not random. simply by dividing the original function W(v) by the number Becauseof a relatively small number of data points and the of data points N (over which W(u) is summed), as Deeming lack of randomness in the spacing, we chose not to impose proposes.Then, for equal time spacing,the spectral window is equal spacing through interpolation, as this would introduce unity at integer valuesof v and zero elsewhere.For unequal new and unreliable data points. Instead, we analyze the un- spacing,W(u) 2 may be large for nonintegralvalues of v, but weighted,unequally spacedSO 2 abundances. lessthan unity. These secondarypeaks in the spectral window In determining the best approach to a discrete Fourier produce artifact peaks in the discrete Fourier transform. This transform on unequally spaced data, we considered several is becausethe discreteFourier transform representsthe convo- 5274 ESPOSITOET AL.' VENUSSO2, 1978-1986

POWER SPECTRUM FROM GLOBAL AVERAGED SO2 POWER OF SPECTRAL WINDOW FOR DATA SPACING

1200 i I i 1.0 ! i i i

1000 0.8

800 er- 0.6 -

0.4 g 400600 i

0.2 200

0 I 0 O. 1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 FREQUENCY (DAY -1) FREQUENCY (DAY -1) Fig. 7. Power spectrumof mean $0 2 for orbits containingspectra Fig. 9. Power of spectral window for actual time spacing of orbits between 474 and 574. Peak at v = 0.25 + 0.03. containing spectra474-574. lution of the true Fourier transform with the spectral window The same analysis was performed on SO• concentrations and, if the spectral window has peaks which are artifacts of the derived from modeling F spectra.This data set of all available data spacing, the effects of these peaks will then be present in SO• concentrationsbetween orbits 2160 and 2255 also had the convolution. the mean and first moment removed. The result of the discrete The Fourier transform of the data is normalized differently Fourier transform was a power spectrumwith a peak of 0.22, from the spectral window, so that the transform is symmetric correspondingto a period of •4.5 days. The spectralwindow and approximately obeys Parseval's theorem; that is, average for this data set is similar to the earlier spectral window, with power is nearly equal in the time and frequncy domains. The a largesecondary peak at 0.43(3/7 day-•). FORTRAN code used is very close to that proposed by These analysesshow rotation rates for SO• features com- Deeming, with minor adjustment for the different normal- parable to Venus mean cloud top rotation (•4.8 days) and ization (we dividethe power by N insteadof N2). rotation of the dark "Y" feature (4-5 days), as determined by The two calculated power spectra are shown in Figures 7 Rossow et al., 1980. and 8. The power of the spectral window for the time spacing DISCUSSION of both data setsis shown in Figure 9. The main peak in both frequencypower spectrais at 0.25 + .03 (Earth days)-•, giving Sourcesof Error and Uncertaintyin SO2 a rotation period of 4.0 + 0.5 Earth days for the cloud top. In Abundance the spectral window the first artifact peak at o = 0.14 repre- The data analyzed here representthe longestuniform spec- sentsa weekly periodicity (schedulingof spacecraftcommands troscopicdata set for the planet Venus. Identical Venus obser- is on a weekly basis). The second artifact peak at o = 0.43 vations have been made on a weekly basis for more than 8 represents a mean sampling rate of 3 orbits per week. This arises because an average of three out of every seven orbits years,and thoseobservations have beenreduced and analyzed have the spectral or multicolor data which we can use to using a single establishedprotocol. This analysis, although calculateSO2 concentration.The primary peak in the spectral optimized to measureSO2 abundance,has a number of imper- window, at o = 0, has the value 1.0, as expected. fections, which we now discuss. The total declineof instrumentsensitivity over the length of the mission is now known within 10%, because of careful star POWER SPECTRUM FROM MAXIMUM SO2 calibrations.However, we have modeled this as a purely linear 3500 i I i i • decline. Some evidence shows that the rate of decline is less in

3000 recentyears. This meansthat the abundanceof SO2 in the last 1000 days may be underestimated by up to 50%. Likewise,

2500 any other deviations from linear behavior have been interpre- ted as SO 2 variability. This means that short-term trends 2000 should not be overinterpreted. A secondcharacteristic of the analysisis the use of a simple 1500 two-parametermodel (SO2 abundanceat 40 mbar and scale height at 40 mbar). Model results are only calculated for dis- 1000 crete pairs of these parameters. Venus is certainly more com- plicated than this! For example, the assumption of a simple 500 exponential distribution with a constant scale height is cer-

I l tainly wrong. As the mixing ratio of SO2 declines,the photo- 0.1 0.2 0.3 0.4 0.5 chemicallifetime of SO2 increasesuntil diffusion dominates. FREQUENCY(DAY -1) Eventually, a constant mixing ratio is establishedin the strato- Fig. 8. Power spectrum of maximum concentration averaged over sphere,causing the SO2 scaleheight to increasewith altitude. 10 ø latitude bins from orbits 474 to 574. Such an effect would be evident from observations at different ESPOSITOET^L.: VENUS SO2, 1978-1986 5275

zenith angles. However, our abundance determination show It is well known that major volcanic eruptions give rise to a no strong dependenceon zenith angle ['see Esposito, 1984]. haze layer of submicron particles in the Earth's atmosphere. Thus while ignoring this effect provides a source of error in The mass of aerosols in the Earth's stratosphere after the our determinations, we have insufficient information to correct eruptionsof E1 Chich6nin 1982 approached10•3g [Thomas for it. We guard against some problems by not using data et al., 1983]. Vertical optical depths measured in the Earth's which do not lie within 10% rms of one of the model spectra. polar regions exceeded0.1. The total mass of the Venus polar The models were originally tuned to match real data in the haze inferred[Kawabata et al., 1980] is 2 x 10•4g; the optical first 200 orbits. After this, the model set was fixed. As Venus depth of the Venus polar haze exceeds1.0. Thus the amount of has changed over the years, the model set has not. This has aerosols injected into the Venus middle atmosphere is greater the advantage of uniformity, but a disadvantage in that the by at least an order of magnitude than that associated with later orbits are not as well spanned by the model set. the most recent volcanic episodeson Earth. The models include a number of assumptions about the Phillips and Malin [1983] have suggestedvolcanic activity Venus atmosphere. For example, the amount of the second, may be more vigorous on Venus than on Earth becauseof the unknown UV absorber is assumed constant and to have a flat lack of organized plate tectonicson Venus. The more vigorous absorption spectrum over the wavelengths of interest (see volcanic activity on Venus might be due to larger volcanoes,a Table 2). The submicron haze varies only in the polar region greater number of volcanoes, or more frequent eruptions. The model in a slow way that accounts for the broad long-term last two of these possibilitiescould account for a higher fre- trendsmeasured by the CPP. As the SO2 abundancedeclines, quency of SO2 injectionsinto the visible Venus atmosphere.It these other scattering componentshave become more impor- should be noted that the injection of SO 2 to cloud heights tant, thus decreasingthe reliability of our SO2 abundance requires a substantial explosive eruption. Theoretical expecta- determinations. tions of lower volatile inventory and the higher Venus surface The intrinsic variability of SO2 also provides some limit on temperature would make such explosionsless likely on Venus. the accuracy of determining its mean global abundance. Our Recently, Taylor and Cloutier [1986] have criticized the in- resultsshow that the mean amount of SO2 visibleon consecu- terpretation of Pioneer Venus observations as evidence for tive days can vary by a factor of 4. Horizontal variability lighting and the more general idea of active volcanoes on provides another limitation. Spacecraft images clearly show Venus. Their arguments against volcanoesrepeat some of the structure in the absorption features smaller than our best cautions enumerated by Esposito [1984]. The case for vol- image resolution (about 400 km square), and occasionally canoes erupting now is clearly still open. However, the vol- smaller than our spectralmode resolution (as large as 80 km). cano hypothesisexplains the major aspectsof the SO 2 behav- This unresolved structure can cause none of the models to fit ior, the polar haze decline, and the recurrence of bright Venus the observationor an underestimateof the mean SO2 abun- polar caps. This hypothesis fits well with electromagnetic, geo- dance. chemical, topographic, and tectonic evidence showing geologi- The general agreementof the image and spectral abundance cally recent volcanic activity [-seeEsposito, 1984]. determinations (see Figure 5) showsthat horizontal structures with scalesresolved by the spectralmode observationsbut not by the imaging mode do not contribute to significanterrors in TABLE 3. Seasonal Mean SO2 Abundance our derived abundances.However, scale lengths shorter than Derived SO2 this may give rise to incorrect abundances. Mean PVO Abundance at Altogther, however, we believe that observations over a orbit Date 40 mbar, ppb season(see Table 3) approximate the actual global means and that the data quality does not justify a model with substan- Sunlit periapsis seasons tially more parameters or allowance for unresolved horizontal 1 8 Dec. 12, 1978 82 variability. In the interests of uniformity we have not tinkered 2 191 June 13, 1979 32 with the model set as a function of time. Because of these 3 433 Feb. 11, 1980 10 considerations the systematic errors in the later determi- 4 619 Aug. 15, 1980 14 5 871 April 25, 1981 6 nations are probably larger than those in the first 400 orbits. 6 1067 Nov. 7, 1981 21 A number of these systematicerrors go in the same direction, 7 1326 July 24, 1982 10 so that SO2 abundancemay be underestimatedby 50-100% 8 1527 Feb. 10, 1983 5 in the last 1000 orbits. In the worst case this could be consis- 9 1771 Oct. 12, 1983 18 10 11 tent with no significantdecline in SO2 abundanceafter orbit 1961 April 19, 1984 ll 2216 Dec. 29, 1984 14 1500. 12 2455 Aug. 25, 1985 9 13 2681 April 8, 1986 5 Actire Venus Volcanoes

Esposito[1984] has interpreted theobserved SO2 injection Sunlight seasons apoapsis and resulting haze formation as due to episodic volcanism. 1 100 March 14, 1979 51 Radar studies by the Pioneer Venus Orbiter [Masursky et al., 2 320 Oct. 21, 1979 21 1980]have detected topographic features on Venusthat re- 3 519 May 7, 1980 23 4 759 Jan. 2, 1981 20 sembleterrestrial volcanic landforms. Prinn [1985] argues that 5 971 Aug. 3, 1981 13 the observedamount of SO2 in the lower atmosphereis out of 9 1871 Jan. 20, 1984 9 equilibrium with the surface composition measured by the 10 2077 Aug. 12, 1984 4 latestVenera landers [Moroz, 1983].This currentimbalance 11 2318 April 10, 1985 11 12 2527 Nov. 5, 1985 3 requires a geologically recent source of sulfur compounds at 13 2771 July 7, 1986 9 the surface. 5276 ESPOSITOET AL..'VENUS SO2, 1978--1986

However, we will not know for certain that volcanoes are Esposito, L. W., and L. D. Travis, Polarization studies of the Venus erupting until we have a direct sighting or observebefore-and- UV contrasts: Cloud height and haze variability, Icarus, 51, 374- 390, 1982. after topography. The best opportunity for this is from the Esposito,L. W., J. R. Winick, and A. I. Stewart, Sulfur dioxide in the radar imaging on the Magellan mission, to be launched to Venus atmosphere: Distribution and implications, Geophys. Res. Venus later in this decade. Lett., 6, 601-604, 1979. Hansen, J. E., and J. W. Hovenier, Interpretation of the polarization CONCLUSIONS of Venus, d. Atmos. Sci., 31, 1137-1160, 1974. Kawabata, K., and J. E. Hansen, Interpretation of the variation of Pioneer Venus spectroscopyof the Venus cloud tops shows polarization over the disk of Venus, d. Atmos. Sci., 32, 1133-1139, a decline by about a factor of 10 in the amount of SO2 since 1975. 1978. A consistentdecline is also seen in the amount of polar Kawabata, K., D. Coffeen, J. Hansen, W. Lane, M. Sato, and L. haze, which was also unusually abundant in 1978. Star cali- Travis• Cloud and haze properties from Pioneer Venus polarimetry, d. Geophys.Res., 85, 8129-8140, 1980. brations determine the instrument sensitivity well enough to Moroz, V. I., Summary of preliminary results of the Venera 13 and rule out the decline as an instrumental effect. Systematic errors Venera 14 missions,in Venus edited by D. M. Hunten, pp. 45-68, are probably larger in the later orbits and may be as large as University of Arizona Press, Tucson, 1983. 50-100% for the SO2 abundance.A plausibleexplanation for Masursky, H., E. Eliason, P. G. Ford, G. E. McGill, G. H. Pettengill, G. C. Schaber, and G. Schubert, Pioneer Venus radar results:Ge- the observed behavior is injection of SO2 into the Venus ology from images and altimetry, d. Geophys.Res., 85, 8232-8260, middle atmosphere by a volcanic explosion [Esposito, 1984]. 1980. The assumed injection can account for the current geo- Owen. T., and C. Sagan, Minor constituents in planetary atmo- chemical imbalance between the surface and atmosphere, spheres: Ultraviolet spectroscopyfrom the Orbiting Astronomical Observatory, Icarus, 16, 557-568, 1972. noted by Prinn [1985] and also the failure of steady state Phillips, R. J., and M. C. Malin, The interior of Venus and tectonic photochemical models to match both the small-scaleheight of implications, in Venus, edited by D. M. Hunten, pp. 159-214, Uni- SO2 and the upper limits on 0 2. Since SO2 is the major versity of Arizona Press, Tucson, 1983. precursor of the Venus clouds and is an important absorber of Ponman, T., The analysis of periodicities in irregularly sampled data, Mon. Not. R. Astron. Soc., 196, 583-596, 1981. solar radiation, steady state models of the chemistry and dy- Prinn, R. G., The photochemistry of the atmosphere of Venus, in namics of the cloud top region must be used cautiously. The Photochemistryof Atmospheres,edited by J. Levine, Academic, Or- abundances of other chemically active constituents, such as lando, Fla., 1985. C12, CO, and polymorphicsulfur, must also experiencecou- Rossow, W. B., A.D. DelGenio, S.S. Limaye, L. D. Travis, and P. H. Stone, Cloud morphology and motions from Pioneer Venus pled episodicvariations. images,d. Geophys.Res., 85, 8107-9128, 1980. Stewart, A. I., Design and operation of the Pioneer Venus Orbiter Acknowledgments. Larry Travis supplied the information on polar ultraviolet spectrometer,IEEE Trans. Geosci.Remote Sens.,GE-18, haze optical depths. We have had helpful discussionswith Y. L. Yung 65-70, 1980. and H. Masursky. The Pioneer Project Office allowed us to turn the Stewart, A. I., D. E. Anderson, L. W. Esposito, and C. A. Barth, UV spacecraft to observe stars. This work has been supported by the spectroscopyof Venus: Initial resultsfrom the Pioneer Venus Or- Pioneer Venus Project Office and the NASA Planetary Atmospheres biter, Science, 203, 777-779, 1979. Program (NAGW-389). Taylor, H. A., and P. A. Cloutier, Venus: Dead or Alive? Science,234, 1087-1093, 1986. REFERENCES Thomas, G. E., B. M. Jakosky, R. A. West, and R. W. Sanders, Satellite limb-scanning thermal infrared observations of the E1 Barker, E. S., Detection of SO2 in the UV spectrumof Venus, Geo- Chich6n stratosphericaerosol: First results,Geophys. Res. Lett., 10, phys. Res. Lett., 6, 117-120, 1979. 997-1000, 1983. Colin, L., The Pioneer Venus program, d. Geophys.Res., 85, 7575- Thompson, B. A., P. Harteck, and R. V. Reeves,Jr., Ultraviolet ab- 7598, 1980. sorption coefficientsof CO 2, CO, 0 2, H20, N20, NH 3, NO, SO2, Conway, R. R., R. P. McCoy, and C. A. Barth, IUE detection of sulfur and CH4 between1850 and 4000•, d. Geophys.Res., 68, 6431- dioxide in the atmosphere of Venus, Geophys.Res. Lett., 6, 629-631, 6436, 1963. 1979. Travis, L. D., D. L. Coffeen, J. E. Hansen, K. Kawabata, A. Lacis, W. Cruikshank, D. P., and G. P. Kuiper, An upper limit for the abun- A. Lane, S.S. Limaye, and P. H. Stone, Orbiter cloud photo- dance of SO2 in the Venus atmosphere,Commun. Lunar Planet. polarimeter investigation, Science,203, 781-785, 1979. Lab., 6, 196, 1967. Winick, J. R., and A. I. F. Stewart, Photochemistryof SO2 in Venus' Deeming, T. J., Fourier analysis with unequally-spaced data, Astro- upper cloud layers,d. Geophys.Res., 85, 7849-7860, 1980. phys. Space Sci., 36, 137-158, 1975. Yung, Y. L., and W. B. DeMore, Photochemistry of the atmosphere DelGenio, A.D., and W. B. Rossow, Temporal variability of ultra- of Venus: Implications for atmospheric evolution, Icarus, 51, 199- violet cloud features in the Venus stratosphere, Icarus, 51, 391-415, 247. 1982. 1982. Dollfus, A.• M. Auri•re, and R. Santer, Wavelength dependence of M. Copley, R. Eckert, L. W. Esposito, L. Gates, A. I. F. Stewart, polarization, XXXVII, Regional observations of Venus, Astron. d., and H. Worden, Laboratory for Atmospheric and Space Physics, 84, 1419-1436, 1979. Campus Box 392, University of Colorado, Boulder, CO 80309. Esposito, L. W., An adding algorithm for the Markov chain formal- ism for radiative transfer, Astrophys. d., 233, 661-663, 1979. Esposito, L. W., Ultraviolet contrasts and the absorbers near the Venus cloud tops, d. Geophys.Res., 85, 8151-8157, 1980. (Received October 1, 1987; Esposito, L. W., Sulfur dioxide: Episodic injection showsevidence for revised January 22, 1988; active Venus volcanism, Science, 223, 1072-1074, 1984. accepted January 22, 1988.)