Tropospheric OH in a three-dimensional chemical tracer model: An assessment based on observations of CH 3 CCl 3

eScholarship provides open access, scholarly publishing services to the University of California and delivers a dynamic research platform to scholars worldwide. Abstract: The three-dimensional global distribution of OH over a year is calculated as a function of temperature, ultraviolet irradiance, and densities of H 2 O, CO, O 3 , CH 4 , and NO t , (defined as NO + NO 2 + NO 3 + 2N 2 O 5 + HNO 2 + HNO 4 ). The concentration of OH is computed within a chemical tracer model (CTM) with an accuracy comparable to that of a detailed photochemical model. Distributions of CO, NO t , O 3 , CH 4 , and the density of O 3 column were specified on the basis of observations. Meteorological fields were derived from the general circulation model developed at the Goddard Institute for Space Studies. The numerical method for parametrization of chemistry is described in Spivakovsky et al. (this issue). The CTM is used to simulate the global distribution of CH 3 CCl 3 . The computed distribution of OH implies a lifetime of 5.5 years for CH 3 CCl 3 (obtained by relating the global burden of CH 3 CCl 3 to the global loss, integrated using simulated three-dimensional distributions). Analysis of the long-term trend in CH 3 CCl 3 as defined by observations suggests a lifetime of 6.2 years (consistent with Prinn et al. (1987)), indicating that model levels of OH may be The three-dimensional global distribution of OH over a year is calculated as a function of temperature, ultra- violet irradiance, and densities of H20, CO, 03, CH4, and NOt (defined as NO + NO2 + NO3 + 2N205 + HNO2 + HNO4). The concentration of OH is computed within a chemical tracer model (CTM) with an accu- racy comparable to that of a detailed photochemical model. Distributions of CO, NOt, 03, CH4, and the density of 03 column were specified on the basis of observations. Meteorological fields were derived from the general circulation model developed at the Goddard Institute for Space Studies. The numerical method for parametrization of chemistry is described in Spivakovsky et al. (this issue). The CTM is used to simulate the global distri- bution of CH3CC13. The computed distribution of OH implies a lifetime of 5.5 years for CH3CC13 (obtained by relating the global burden of CH3CC13 to the global loss, integrated using simulated three-dimensional distribu- tions). Analysis of the long-term trend in CH3CC13 as defined by observations suggests a lifetime of 6.2 years (consistent with Prinn et al. (1987)), indicating that model levels of OH may be too high by about 13%. This estimate for the lifetime depends on industry data for global emissions and on the absolute calibration of observations. It is argued that seasonal variations of CH3CC13 provide an independent test for computed OH fields that is insensitive to the uncertainties in the budget of CH3CC13. The annual cycle of CH3CC13 from about 25øS to the South Pole is dominated by seasonal changes in OH. Observed seasonal variations of CH3CC13 indicate that the OH field south of 20øS+4 ø should be scaled by 0.75 _+0.25 from computed values, consistent with the result based on long-term trends. Reactions involving non-methane hydrocarbons were not included in the current model. These reactions could account for lower concentrations of OH than computed. Seasonal variations of CH3CC13 in the tropics and in the northem mid4atitudes are dominated by effects of transport. If use of CH3CC13 is phased out (as envisioned by the Montreal protocol), the dynamically driven seasonal variations of CH3CC13 will decrease dramatically, whereas the chemically driven variations will remain proportional to the concentration of CH3CC13; then the annual cycle of CH3CC13 in northem mid-latitudes will provide a measure of OH as does at present the annual cycle in southern mid-latitudes. The influence of chemistry on the lati- tudinal distribution of CH3CC13 is small and at present does not provide a constraint for the globally averaged OH or for the latitudinal distribution of OH. However, ff emissions of CH3CC13 were to cease, the tropical depression in the concentration of CH3CC13 caused by high levels of OH in the tropics may provide an additional means to test OH models. variations in southern mid-latitudes at present. In addition, the tropical dent in the latitudinal distribution of CH3CC13, caused by high concentrations of OH in the tropics, will also provide a measure of OH, since the effect of will no longer be obscured by interannual variations in transport and emissions. Consistent calibration of measurements and fine latitudinal resolution, including that in the equatorial region, will be required.

They introduced a CTM designed to run independently of the GCM using transport fields recorded in a prior GCM simulation.Subsequent work at GFDL and other laboratories examined global distributions of Os [Mahlman et el., 1980;Levy et el., 1985], N20 [Levy et al., 1982], CO [Pinto et al., 1983], halocarbons [Golombeck and Prinn, 1986], NOx (NO + NO2) [Levy andMoxim, 1987, 1989], and CO2 [Fung et el., 1983].These studies focused on the effects of transport, with the exception of the early study of CO which used monthly and longitudinal averages of concentrations of OH.
The present paper belongs to a series intended to develop a global CTM with realistic simulation of photochemistry as well as meteorology.The CTM uses transport fields from the GCM developed at Goddard Institute for Space Studies [Hansen et el., 1983]; numerical algorithms for simulating advection are described by Russell and Lerner [1981] and Prather [1986].The CTM has been used to study the global dispersion of species emitted mainly in industrialized regions at northem mid-latitudes, chlorofluorocarbons (CFCs) [Prather et al., 1987], and a•Kr [Jacob et al., 1987].Transport from continents to the oceans and ventilation of the boundary layer was tested in the simulation of the distribution of ZZ2Rn [Jacob and Prather, 1990;Balkanski and Jacob, 1990;Balkanski et el., 1989].
The present study introduces the representation of tropospheric chemistry that allows for calculation of chemical reaction rates within a CTM simulation with the accuracy comparable to that of the derailed chemical model.(The numerical method for parametrization of chemistry is described by Spivakovslcy et al. [this issue].)The CTM is described in section 2. The global distribution of tropospheric OH is calculated as a function of time, atmospheric composition, temperature, and solar irradiance (section 3).Sections 4-6 examine the options for testing the accuracy of computed fields for OH.
In section 4 we use observed temporal trends of CHzCClz [Prinn et el., 1987] in order to evaluate the globally integrated OH concentration (weighted by frequency of reaction with CH3CClz), assuming accurate knowledge of rams for emissions and the current absolute calibration of the observational data.We discuss a procedure, proposed by Prinn et el.[1987], that was intended to give an estimate of both the loss rate and the absolute calibration, for a given level of emissions.We show that a family of values for these parameters is consistent with presently available data for CHzCClz and conclude that in practice, one has to assme the absolute calibration in order to determine the loss rate of CHzCCI•.
Section 5 examines the feasibility of using the observed latitudinal distribution of CH•CCI• as a test for computed concentrations of OH [cf.Prinn et el., 1987].Model results indicate that the latitudinal distribution of CH•CClz is determined primarily by transport, not by chemistry.We show that at present the latitudinal distribution of CHzCCI• as defined by the ALE data provides little information on either the globally integrated OH concentration or on the distribution of OH with latitude.
Section 6 focuses on seasonal variations of CHzCC13 as a potential test for computed concentrations of OH.Observations at middle latitudes display annual cycles that have been attributed to seasonal variations in atmospheric transport [Prinn et el., 1983] and in the concentration of OH [Khalil and Rasmussen, 1984b;Fraser et el., 1986].Our analysis suggests that annual cycles of CH•CCI• are controlled, or strongly influenced, by dynamical processes at latitudes northward of 25øS.We show that seasonal variations of CH•CCI• at southern mid-and high-latitudes provide a test for computed concentrations of OH.The resuks of this test are insensitive to errors in absolute calibration or in the rams assumed for global emissions of CH•CCI•.

Dynamics
The CTM solves the three-dimensional continuity equation for chemically reactive tracers.The dynamical module is described by Prather eta/.[1987].It uses wind fields, surface pressures, and convective mass fluxes recorded at 4-hour intervals from the GCM [Hansen et el., 1983].The GCM has a resolution of 4 ø in latitude by 5 ø in longitude.Height is resolved into 9 layers using o coordinates.The two highest layers reach into the stratosphere, extending from approximately 150 to 10 mbar.For the present study, transport fields derived from the 4øx5 ø GCM were averaged over spatial elements of 8 ø latitude by 10 ø longitude and into 8hour time intervals.Advection was treated using the slopes method of Russell and Lerner [1981].The simulation of the interhemispheric transport, examined previously using distributions of CFCs [Prather et al., 1987] and SSKr [Jacob et al., 1987], is tested further as discussed below.

Chemistry
The chemical module computes the change in tracer concentration due to chemical reactions during a time step.Rates of chemical production and loss vary spatially and temporally reflecting variations in background concentrations, temperature, and solar irradiance.The irradiance in turn is a function of cloudiness, surface albedo, solar zenith angle, and the density of the overhead ozone column.The chemical module is designed to use these quantities as input, with either prescribed fields or fields simulated by the CTM.In the present work, climatologies were specified for O:•, NOt (defined as NO + NO2 + NO• + 2N2Os + HNO2 + HNO4), CO, CI-I4, and ozone column using observations as described in Appendix A. Distributions of temperature, cloud cover, surface albedo, and water vapor were adopted from the GCM.Water vapor concentrations above 500 mbar were specified using observations, since concentrations of H20 derived by the GCM for this altitude region are known to be too high (see Appendix A for details).The 24-hour average OH field was recalculated every 5 days.
In previous studies we used an efficient photochemical model to describe the chemistry of the troposphere [Wofsy, 1978;Logan et el., 1978Logan et el., , 1981;;Prather et el., 1984].If this model were used as a chemical module in the CTM, computation of the OH field for 1 year with a 5-day time step would take more than 50 hours of the CPU time on an Amdahl-V6 computer.Conceivably, for the present study the OH field could be computed once and then recycled within the CTM simulations, since there is no feedback between CH•CCI• and OH.However, this approach would preclude the study of sensitivity of the computed field to concentrations of Os, NOt, CO, and H20, described in section 3. Furthermore, we intend to use the present model for simulations of CO, NO,, and Os, i.e., for studies requiring computation of chemical rates within the CTM simulation based on tracer distributions at a given time step.
In order to reduce the time required for computation of OH, we replaced the detailed chemical model with a set of explicit expressions describing the functional relationship between input parameters and the concentration of OH as computed in the full model [cf.Dunker, 1986;Marsden et el., 1987].We expressed the 24hour average concentration of OH as a high-order polynomial in N independent variables including temperature, radiation conditions, and concentrations of Os, NO,, CO, and water vapor.The mathematical procedure for deriving the polynomials is described by Spivakovsky et el.[this issue].As shown there, the polynomials represent tropospheric OH with the accuracy comparable to that of the full chemical model.The computational cost, however, is reduced by more than a factor of 600.A simulation of 1 year for CH3CC13 with the present CTM takes about 85 min on Amdahl-V6 with less than 5 min spent for computation of OH.
The procedure for determining stratospheric loss of CH 3 CC13 is described in Prather et al. [1987].Both the photodissociation of CH3 CC13 and its destruction by OH are treated as a first-order loss process.See Appendix B for more detail.

TROPOSPHERIC OH A S CALCULATED IN THE THREE-DIMENSIONAL MODEL
The zonally averaged distribution for tropospheric OH is shown as a function of latitude and pressure for four seasons in Figure 1 and Table 1.Concentrations are highest in the tropics and in mid-latitudes in summer.Elevated levels of 03, NOt, and H20 in the northern hemisphere lead to the enhancement of OH, mitigated to some extent by higher concentrations of CO.
The shape of the vertical distribution of OH reflects mainly that of the rate of OH production by reaction

O3+hv --> O2+O(1D) (2)
To a good approximation, the rate of (1) is proportional to the densky of 03, to the mixing ratio of H20, and to the ratio J/k, where J is the frequency of ( 2       ture, we present information on rates for production and loss for the important reaction paths (see Table 2) in addition to concentrations of OH.The three-dimensional OH field from the present study (8 ø latitude by 10 ø longitude, for 7 and 9 vertical layers for the mid-latitudinal and tropical troposphere, respectively) computed with a 5-day time step for a year is available from the authors as ASCII files on the 5 1/4" floppy disks.

Comparison With Observations
Observed and calculated concentrations of CHzCC13 at the five ALE/GAGE stations [Prinn et al., 1987]   However, this optimization problem appears to be ill posed.There is a continuum of near-optimal pairs (L, Fc) that correspond to values of S near the minimum.Figure 5  Concentrations are presented as increments with respect to the concentrations at Tasmania (41øS, 145øE), to remove a temporal trend in concentration.Some uncertainty is associated with interpretation of model results at the ALE sites north of the equator, in the regions with steep horizontal gradients (see Figure 7).The horizontal grid used in the CTM cannot resolve concentration differences along these gradients to better than 2-3 ppt, and uncertainty of comparable magnitude arises from exclusion of "polluted events" in the computation of background concentrations at the ALE stations in the model.In the ALE/GAGE program, the influence of pollution at Adrigole was identified by noting occasions when the concentration of short-lived industrial tracers was unusually high; samples exceeding a certain threshold were excluded from the computation of monthly means in defining the background [Prinn et al., 1987;Cunnold et al., 1986].A similar procedure may be implemented in simulations by introducing a short-lived tracer emitted in industrialized areas, if the additional effort appears warranted.At present, we use a simple method to compute monthly mean "background" concentrations for the model at ALE sims taking the monthly median (50th percentile) to represent the background level [cf.Prather et al., 1987].This approach rests on two assumptions: (1) the median is not significantly affected by pollution events; and ( 2 In this section we compare observed and simulated seasonal variations for CH3CC13 at all sites for which there are sufficient data to define the annual cycle (section 6.2).Observations of CFCs are used to confirm the accuracy of the model in reproducing the dynamical component of the annual cycle.We show that the seasonal variation of OH is the dominant influence on the annual cycle of CH3CC13 south of about 25øS, and we use observed seasonal variations of CH3CCI• to estimate the accuracy of the computed OH field.This assessment proves to be insensitive to uncertainties in either absolute calibration or in the magnitude of the source, as shown in section 6.1.In section 6.3 we discuss processes that in the model determine annual cycles of CH5CC13 and CFCs.We begin by outlining the techniques used here for the study of seasonal variations.

Variations
The annual cycle of a gas is represented by the average concentration for each month, over the length of the record.Prior to averaging, a long-term trend is removed [Fraser et al., 1986; Ent-/ng, 1987; Pearman and Beardsmore, 1984;Thompson et al., 1986].Methods for defining the long-term trend vary from one study to another [Chatfield, 1984;Durbin, 1963], but the differences are inconsequential as long as the residuals compose a stationary time series, with mean values for individual years dose to ZerO.
We define a time-dependent concentration • incorporating variations with periods longer than a year as where t is in years, q,(t) is a polynomial, andf(t) is given by for a data record of N years.First, the coefficients of q,(t) axe defreed using the least squares fit to the time series; the degree n is chosen as the highest for which the best fit polynomial is void of inflection points within the span of the record.The coefficients ot k, Ilk in J•0 are then chosen to give the least squares fit to the residuals obtained by subtracting q•(O from the time series.The function J•t) accounts for variations with frequency 1 through (N-1) cycles over the span of the record, and q,(t) represents those with frequencies lower than half a cycle per N years.
The definition of a ct•a used here guarantees that the sum of residuals [c(t)-ct•a(t)] is equal to zero for the entire record, but not for an individual year.We evaluate the deviation of ct

Comparison of Observed and Simulated Annual Cycles
We examine the relative importance of chemistry and transport in the annual cycles of CH3CCI3 by comparing model simulations for three tracers: 1) MCL/OH: CH5CC13 with the loss frequency calculated on the basis of the standard OH; 2) MCL/50%OH: CH3CC13 as in MCL/OH, but with OH everywhere divided by 2; and 3) MCL/ASEAS: a contrived tracer, identical to CH3CCI3 in all but one respect: the associated loss frequency in each grid box is at all times equal to the annual mean for the box in the simulation with the standard OH.Model runs were performed using two   To summarize our analysis of CFC observations at Tasmania, we conclude that most of the 5840 data do not exhibit a significant annual cycle, consistent with Fraser et al. [ 1985a].In contrast to these, the 5880 data display significant annual cycles, which are reproduced by the model.The 5840 instrument has a lower signal-to-noise ratio and may therefore fail to detect small seasonal variations [cf.Fraser et al., 1985a].However, it is conceivable that the derived annual cycles of CFCs at Tasmania could be artificially enhanced by recalibration.We conclude that existing data define an upper limit for possible seasonal variations in CFCs at southern mid-latitudes.The model is either accurate in simulating the annual cycle of CFCs or it coincidentally reproduces the magnitude and phase of an artifact.In any case the model does not underestimate the amplitude of seasonal variations in CFCs as defined by available observations.The same conclusion should apply to the dynamical component of the seasonal variations of CH3CC13.
Prather [1985,1988] estimated seasonal variations in emission rates for various halocarbons using ALE data colleCted during pollution episodes at Adrigole.He assumed that emission rates for CFC-11 are uniform with season, based on the diverse uses of this compound.Seasonal panems for enhancements of concentrations of CFC-12 and CH3CC13, relative to that of CFC-11, were identified and attributed to seasonal variations in emission rates.The annual modulation of CH3CC13 release was found to be less than 20%, with maximum release in spring, minimum in summer.Seasonal changes in emission rates of this magnitude have no effect on computed annual cycles in the southern hemisphere, and a small effect in the northern hemisphere, as shown in section 6.2.2.
Our analysis suggests that computed fields for global OH can be tested by comparing observed and computed seasonal variations in CH3CC13 at southern mid-latitudes, where the annual cycle of CH3CC13 is determined by seasonal variations in the loss rate.Figure 17   Observed variations at Samoa (12øS, 171øW) appear to be significant at the 95% confidence level for CFC-11 (P) and CFC-12, and marginally detectable for CFC-11 (S) as shown in Figure 12g.Observed concentrations of CFCs are higher in January-April than in July-October, consistent with model results, as shown in Figure 23.For CH3CC13 the model predicts that chemical and dynamical components of seasonal variations should be out of phase (Figure 24), resulting in small net variations (about 1 ppt); the observed annual cycle of CH•CCI• appears to be insignificant (Figure 12g).Model results are not inconsistent therefore with the observed seasonal behavior of halocarbons at Samoa.CHzCC13 attributed to chemistry is twice as high for northern mid-latitudes as for southern mid-latitudes (Figure 25), reflecting both higher concentrations of CH3CC13 and higher concentrations of OH in the north.Concentrations of OH in summer are higher in the north mainly due to higher concentrations of NOt and O3 and to larger areas of land (higher surface albedo).If the global burden of CH3CC13 continues to rise, the amplitude of the seasonal oscillation of CH3CC13 (in pptv) south of about 25øS (where the seasonality is associated mainly with chemical loss) may be expected to increase proportionally.The predicted temporal change in the seasonal amplitude may be observable and could provide a useful check on the validity of the analysis presented here.If, on the other hand, emissions of CH3CC13 are phased out, measurements of CH3CC13 may offer new possibilities for testing OH models.After the cessation of release, we expect dynamically driven variations in CH3CC13 diminish on a time scale of 1 year, whereas the concentration will decay on a longer time scale of 6 years.During this latter period the seasonal variations in northern mid-latitudes will reflect mainly the seasonality of OH, and therefore provide a test for OH models, as do variations in southern mid-latitudes at present.In addition, the tropical dent in the latitudinal distribution of CH3CC13, caused by high concentrations of OH in the tropics, will also provide a measure of OH, since the effect of chemistry will no longer be obscured by interannual variations in transport and emissions.Consistent calibration of measurements and fine latitudinal resolution, including that in the equatorial region, will be required.

Processes Regulating Annual Cycles of
Our analysis suggests that monitoring programs should make a special effort to preserve information present in the frequency range 1-6 year -• (time scales 2-12 months), with particular attention to elimination of step changes associated with recalibration.If additional monitoring stations are to be considered in the future, sites in the subtropics would appear to be particularly interesting.A2.Nitrogen oxides.The CTM requires concentrations of a family of oxidized nitrogen species, NOt (NO + NO2 + NO3 + 2N205 + HNO2 + HO2NO2).Concentrations of NOt, constant throughout a year, were chosen to reproduce representative profiles for NO [Ridley et al., 1987;Davis et al., 1987;Torres and Buchan, 1988;Fehsenfeld et al., 1988].The values of NO t corresponding to observations of NO were derived using the full photochemical model (see Figure A1).
Methane and carbon monoxide.The methane field was assumed to be uniform in each hemisphere, with values of 1.7 ppm in the north and 1.6 ppm in the south.Values for CO in the CTM were based on measurements at a few sites in each hemisphere [Khalil andRasmussen, 1983, 1984c;Seller et al., 1976Seller et al., , 1984;;Khalil and Rasmussen, 1982;Fraser et al., 1986], and from latitudinal surveys obtained using aircraft and ships [Seller, 1974;Heidt et al., 1980;Seiler and Fishman, 1981;Schmidt et al., 1982].Annual mean values for CO are given in Table A3 with latitude resolution of 10 ø.Uniform vertical profiles were used over the oce- tots shown in Table A3.Monthly mean values were derived using the seasonal scaling factors in Table A4 for latitudes 30ø-90ø; seasonal factors for 00-30 ø were interpolated from the values in Table A4 and

Fig. 1 .
Figure 1.The monthly mean concentrations in excess of 106cm -3 extend in the summer hemisphere from the equator to subpolar latitudes, while concentrations in the winter hemisphere are below 106cm -3 poleward of 20 ø.Summer concentrations at 45 ø exceed winter concentrations by more than a factor of 5.The influence of

Fig. 3 .
Fig. 3. OH concentrations for 45øS and 15øN in January and July, computed using the present full photochemical model (solid lines), and reported by Crutzen and Gidel, [1983] (dot-dashed lines), and by Logan et al. [1981] (dashed lines).The values reported by Crutzen and Gidel, which represent daylight averages, were rescaled to obtain 24-hour averages.
with the standard OH model (solid lines) and with OH reduced by 25% (dashed lines), bracket the observed rates at all five stations.Lifetimes of CH 3 C•C13, computed as a ratio of global mass to global loss, for these simulations are 5.5 and 7.1 years, respectively.The best fit to the observed trends would be obtained for a lifetime of 6.2 years, corresponding to reduction of the standard OH concentrations by 13%.The close agreement between model and observations must be regarded with caution, since the comparison depends on the absolute calibration of the observations and on the estimated magni-tude of the global source.The absolute concentration of CHaCC13 was revised by 20% from values initially reported [Rasmussen and Lovelock, 1983], and further adjustments can not be ruled out.It is likely that the global emission rate (Appendix B) has been underestimated, as discussed by Khalil and Rasmussen [ 1984a] and Prinn et al. [1983, 1987].These authors quoted ranges for the t; In Fc + In o(ti)--ln c(co,ti,L (3) i=l with respect to L and Ft.Here o(tl) denotes the observed concentration at time tl, N is the nmber of observations, and co denotes the initial condition.Concentrations c

Fig. 4 .Fig. 5 .Fig. 6 .
Fig. 4. Concentrations of CHzCC13 (pptv) at the ALE sites.Observations (monthly means) are designated by triangles.Results of simulations are represented by monthly medians of surface concentrations calculated for a grid box containing the site [Prather et al. 1987].Solid lines correspond to the simulation with the standard OH.Dashed lines correspond to the simulation with the standard OH reduced everywhere by 25%.

Figure 9
Figure 9 compares annual averages of observed (triangles) and simulated (circles) concentrations of CH3CC13 at the ALE stations for 1979-1984.Concentrations are presented as increments with respect to the concentrations at Tasmania (41øS, 145øE), to remove a temporal trend in concentration.Some uncertainty is associated with interpretation of model results at the ALE sites north of the equator, in the regions with steep horizontal gradients (see Figure7).The horizontal grid used in the CTM cannot resolve concentration differences along these gradients to better than 2-3 ppt, and uncertainty of comparable magnitude arises from exclusion of "polluted events" in the computation of background con-

Fig. 7 .Fig. 9 .
Fig. 7. Global distribution of CH3CCI 3 in pptv at the surface for January and July of 1980.Locations of the ALE/GAGE sites are denoted by solid squares.
•,•a from an ideal baseline by the magnitudes of the ratios R = S•/{J• where St• is a sum of the residuals for an individual year, and ch• is a square root of the variance (with respect to ct•a) over the span of the record.The magnitude of R does not exceed 0.2 for any data analyzed in this study, and most commonly is less than 0.1.It can N-1 be shown easily that .•f(t+i) --0, i.e., the mean annual cyclewould not change if the function J•t) were not included in the definition of ctm,a.However, values of IRI would be larger.The functional form for ct•a is sufficiently flexible to give a good fit without artificially distorting the annual cycle.Coefficients of the polynomials q,(t) for all observations analyzed in this study are given in Table3.The mean deviations of the data from ct•na must be examined for statistical significance, since superposition of random variations will in general produce nonzero mean residuals.The statistical significance of an annual cycle may be assessed using a correlogram (i.e., the autocorrelation coefficient plotted against the time lag[Chatfield, 1984]).A sinewave-like pattern with the frequency of 1 cycle per year is characteristic of correlograms for the time series describing seasonally varying processes.Autocorrelation coefficients for a random time series of N independent observations are expected to be smaller than I5 = 2/x/'ff for 95% of time lags.An annual cycle is not considered significant at the 95% level unless the autocorrelation coefficient exceeds 15 for a time lag of 12 months.Figure11illustrates an application of this test.The upper panel shows the average of the residuals by month over the span of the record, for two time series, CFC-11 at Tasmania (5880 P) (asterisks) and CHzCCla at Samoa (diamonds).Monthly residuals have similar magnitudes at the two stations.However, the corresponding correlograms (lower panel) show that the annual cycle is not statistically significant for CHzCClz at Samoa, while there is a well der'reed annual cycle for CFC-11 at Tasmania.Correlograms for all time series of observations used in this study are given in Figure12.In section 6.2 we show that seasonal variations of CHzCC13 at southern mid-latitudes are determined mainly by oscillations in the loss rate.The amplitude A,•, of these variations when measured in absolute terms (e.g., pptv), is a function of the absolute concentration of CH3CCI3, and thus, for the computed concentration, a function of the calibration standard and rams for emissions and chemical loss, as shown in Figure13a.In contrast to A,h, the amplitude A•l of variations in the ratio [c(t)-c•a(O]/%•a(t), is insensitive to the assumed strength of the source and absolute calibration (Figure13b).We assess the accuracy of the computed loss rate for CH3CCI3, and thus the accuracy of the computed OH

Fig. 17 .Fig. 18 .Fig. 22 .
Fig. 17.Annual cycle in [c(t) -c•-na]/ctr•d for CHaCCIa in southern midqatitudes at Tasmania (a) and at Palmer station (b).Circles and crosses correspond to the mean annual cycle derived from observations with the 5880 and 5840 instruments, respectively [Fraser et al., 1985a, b, 1987; Cronn et al., 1986]ß Dotted lines represent individual years of observations.Solid, dashed, and dot-dashed lines show the annual cycle derived from the simulations MCL/OH, MCL/ASEAS and MCL/50%OH, respectively (see Fig. 14 for the definition of these simulations).The simulations are shown for SSD.Note that the MCL/50%OH curve is almost equidistant from the MCL/OH and MCL/ASEAS curves.This is not the case for the variations in concentration shown in Fig. 14, where the decrease in amplitude caused by reduction in the loss rate was offset by accelerated growth of concentration assorated with the lower loss rate.Also note a change between Figures 14 and 17 in disposition of observations versus the MCL/OH simulation, reflecting the fact that simulated concentrations for standard OH (MCL/OH) are slightly lower than those observed (see Fig. 4).

Fig. 24 .
Fig. 24.Annual cycle of CH3CCI3 simulated for Samoa (solid line).The dashed line corresponds to the MCL/ASEAS simulation.The dotted line represents the chemical component, defined as a difference between the MCL/OH and MCL/ASEAS simulations.

Fig. 23 .
Fig. 23.Annual cycle of (a) CFC-11 and (b) CFC-12 at Samoa.Observations [Cunnold et al., 1987] are represented by circles and crosses for the porosil and silicone columns, respectively.Dotted lines denote variations for individual years of observations.Solid and dot-dashed lines correspond to the simulations for the grid box, containing Samoa at the surface, computed with the FSD and SSD, respectively.Dashed lines show an annual cycle simulated for the grid box with the center located 26.5 ø west of Samoa (FSD).

- 2 Fig. 25 .
Fig. 25.Annual cycles for a zonal mean at 44øN and 44øS.Solid and dashed lines represent MCL/OH and MCL/ASEAS simulations, respectively.Dotted lines correspond to the chemical component in the variatitrs defined as a difference between MCL/OH and MCL/ASEAS.

Fig. 26 .Fig. 27 .Fig. 28 .Fig. 29 .
Fig. 26.Annual cycles at the surface along model zones 8 ø wide centered at 44øN, 12øN, 12øS, and 44øS for MCL/ASEAS, given with a 20 ø step in ltrgitude (dashed lines).Solid lines represent annual cycles for ztrally averaged concentrations.Dotted lines for 12øN correspond to the grid boxes over the Atlantic.Variations for each grid box are given with respect to the base line characteristic of this box, and the annual cycle for the zonal mean represents variations with respect to the cmma for ztrally averaged concentrations.
-dimensional distribution of OH was computed using a new parameterization of tropospheric photochemistry.We have developed a CTM for studying distributions of chemically reactive tracers.The CTM was applied to simulate the distribution of CH3CC13 and to identify the features of the distribution, in addition to the long-term trends, that can be used to evaluate the computed OH.The model reproduces observed spatial and temporal variations in CH3CCI3, as defined by ALE/GAGE observations.The computed OH field implies a global lifetime for CH3CC13 of 5.5 years (obtained by relating the global burden of CH3CCI3 to the global loss, integrated using simulated three-dimensional distributions).Consistent with Prinn et al. [1987], the best fit to the observed long-term increase of CH3CC13 is obtained with a lifetime of 6.2 years, implying that the model overestimates the global abundance of OH by 13%.This result is subject to uncertainties associated with data for global emissions and the absolute calibration of the ALE observations.An independent assessment of the accuracy of the OH field was obtained by analyzing the annual cycles of CH3CCI3 at Tasmania and at Palmer, where the seasonal cycle is determined mainly by seasonal variations in OH.The best fit to the observed seasonal variation was found with the OH field scaled by a factor of 0.75_+0.25,consistent with the estimate obtained from analysis of the long-term trend.This result is insensitive to the assumed emission scenario and to the absolute calibration.On the seasonal scale the whole distribution of CHzCClz south of about 25øS fluctuates as a coherent entity; therefore this estimate for the OH field applies to the concentration of OH averaged over the southern hemisphere from the subtropics to the pole (weighted by frequency of reaction with CH3CCI3).It appears that the present simulation overestimates the global concentration of OH by about 15-25%.We expect that inclusion of reactions with nonmethane hydrocarbons could lead to a reduction in computed concentrations of OH by 10%-40%.Definite conclusions are not yet possible, however, given the deficiencies in the existing data base for nonmethane hydrocarbons, particularly for unsaturated species.CIIF2VIICAL TRACER MODEL We showed that the observed latitudinal distribution of CH3 CC13 at present does not provide a constraint for the concentration of OH and that parameterization of tropical mixing based on data for CFCs [Prather et al. 1987] and ssKr [Jacob et al., 1987] is consistent also with observations of CH3CC13.The interhemispheric difference in the concentration of CH3CC13 remained essentially in a steady state from 1978 to 1985, reflecting an approximately constant rate of emissions over this period.Most of the observed seasonal variations in CFCs and CH3 CC13 can be reproduced assuming constant emissions throughout a year.Annual cycles of CH3CC13 north of about 20øS are determined by seasonality of transport processes.The absolute amplitude (in pptv) of the chemical component in seasonal variations of CH3 CC13 is almost twice as high in the northern hemisphere as in the southern hemisphere, reflecting both higher concentrations of CH3 CC13 and higher concentrations of OH in the north.
Dynamical influences are complex for the region of Barbados and indeed for most of the north Ariantic.A station at Barbados alone can not provide a representation of northern subtropics sufficient for a comprehensive study of interhemispheric exchange or seasonality of transport processes; a northern Pacific subtropical site would add significantly to our understanding of the factors that influence the distribution of tracers in the tropics.A subtropical site at about 30øS would provide an important means for verifying OH models.It is intended that future applications of the CTM will focus on global distributions of CO, CI-h, NO•,, and 03.The chemical module developed for the present study allows for efficient computation of photochemical rams within a CTM simulation (as functions of changing tracer concentrations), essential for most of these studies.If chemical rams are recalculated at every time step, as many applications require, a CTM run using our parameterization of chemistry takes about twice the computer time of a calculation for a conservative tracer.Studies of global distributions of CO and CH4 using the present CTM are currently in progress at Harvard and Goddard Institute for Space Studies.
APPENDIX A: INPUTS TO THE CHEMICAL MODULE Data for temperature, cloud cover, and water vapor necessary for computing concentrations of OH and loss rates for CH3CC13 were recorded in the GCM simulation as 5-day means.Zonal mean temperatures from the GCM agree with observations [0ort, 1983] within 2-3 K. Water vapor was taken from the GCM below 500 mbar, and was derived from observation above.In the CTM simulation, values for temperature, cloud cover, and water vapor below 500 mbar were updated every 5 days.Seasonal means were used for water vapor above 500 mbar.Distributions of 03, CO, NOt, and CH4, described below, were derived from observation.Annual-mean profries were used for NO t and CI-h.Values for concentrations of O3 and CO and for density of the O3 column were updated every month.Ozone.The locations of the stations used to derive the global distribution of tropospheric O3 are given in ozonesonde and surface stations used to derive the ozone climatology.The locations of stations, sources of data, and illustrations of the seasonal and vertical distributions are given by Chatfield and Harrison [1977a,b], Logan [1985], Logan and Kirchhoff [1986], Oltmans and Komhyr [1986], Angle and Sandhu [1986], and Oltmans et al Fig. A 1. Vertical profiles of NOt (in ppt) used in the CTM, for ocean and land.The mid-latitude profiles were used for 90ø-30øN, and the tropical profiles were used for 30øN-90øS.
a value of 1.0 at the equator.Water vapor.Zonal mean concentrations of H20 are within about 20% of observed values [0ort, 1983] in the lower troposphere.The specific humidity in the model tends to be too low below 700 mbar, and too high above, particularly in the tropics (see Figure A2).Model values for the upper troposphere (above 400 mbar) are too high by over a factor of 2, based on comparisons with data given by Oort [1983] and H. J. Mastenbrook [see Logan et al., 1981].The specific humidity at 500 mbar and above was specified, using the global radiosonde network [0ort, 1983] at 500 mbar and more limited data above 500 mbar: Fig. A2.Comparison of •nal m•n v•ues for spe•ic humidity f•m ß e GCM (dashed) and from Oort [1983] (soUd).Resets a• sho• for su•er •d w•ter for 9• and 500 mbar.

(
in Florida) while industry estimates do not.About 3% of emissions are released in the southern hemisphere.This fraction has fluctuated from 2.7% to 3.1% in the period 1979-1986 (P.Midgley, private communication, 1988).We assumed that the release pattern was constant over this period.Initial conditions.The simulations of CH3CC13 were performed for the period of 6 years, starting at a model time of January 1, 1979.The field of initial concentrations was obtained in a special 1 year initialization run.For that run, uniform tropospheric concentrations of 100 and 72 ppt were assumed for northern and southern hemispheres, respectively, for January 1, 1978.Mixing ratios in the stratospheric layers were reduced by factors of 0.67 and 0.25, relative to tropospheric values.These factors were derived from observed stratospheric profiles [World Meteorological Organization, 1982, 1986].Emissions for the initialization run were assumed 490x106kg y-X.Computed concentrations for January 1, 1979, were scaled by an uniform factor, to give a global burden of CH 3 CC13 of 2,007x106kg.