ROCOZ-A ozone measurements during the Stratospheric Ozone Intercomparison Campaign (STOIC)

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 100, NO. D5, PAGES , MAY 20, 1995 ROCOZ-A ozone measurements during the Stratospheric Ozone Intercomparison Campaign (STOIC) Robert A. Barnes ManTech, Incorporated, Wallops Island, Virginia Chester L. Parsons 1 NASA Goddard Space Flight Center, Wallops Flight Facility, Wallops Island, Virginia Arthur P. Grothouse Computer Sciences Corporation, Wallops Island, Virginia Abstract. We present a set of ROCOZ-A (rocket ozonesonde) ozone measurements during the October/November 1988 (pre-stoic) and the July/August 1989 Stratospheric Ozone Intercomparison Campaign (STOIC) in southern California. ROCOZ-A and its associated electrochemical concentration cell (ECC) ozonesondes participated in the comparisons as established techniques for the validation of lidar and microwave instruments that have been proposed for the Network for the Detection of Stratospheric Change (NDSC). For the proposed network instruments, STOIC has provided a picture of their performance characteristics in 1989 and has given an estimate of their future performance in the NDSC. For ROCOZ-A, STOIC has added new information on its accuracy and precision. It is this continuing characterization that gives ROCOZ-A its value in comparisons. The STOIC comparisons show a shift of 5-6% in ROCOZ-A ozone densities (ROCOZ-A higher) from October/November 1988 to July/August This shift has been seen in comparisons with the Stratospheric Aerosol and Gas Experiment II (SAGE II), ECC ozonesondes, and the Jet Propulsion Laboratory (JPL) lidar. The source of this shift has not been determined. Until this new error source is resolved, we recommend that the previously quoted accuracy estimate for ROCOZ-A ozone measurements be increased from 5-7% to 8-10%. About 2% of the difference between ROCOZ-A ozone measurements and those from the proposed network instruments in 1989 appears to be due to differences in atmospheric ozone between the two STOIC sites. A correction for these site-to-site differences brings the ROCOZ-A ozone measurements within 10% of all of the other STOIC instruments, and the average agreement (ROCOZ-A 6% higher) becomes consistent with the historical set of ROCOZ-A comparisons. The STOIC comparisons have shown structures in stratospheric ozone that cannot be resolved by ROCOZ-A with its 4-km vertical resolution. In addition, comparisons with nighttime measurements from the Millitech microwave above 45 km show a divergence from the daytime ROCOZ-A values that agrees with the general characteristics of the diurnal cycle in upper stratospheric ozone that is predicted by photochemical models. Evidence of this ozone cycle is also seen in ROCOZ-A comparisons with the SAGE II zonal mean above 45 km but not in comparisons with the nighttime measurements from the JPL lidar. The effects of diurnal ozone change in the upper stratosphere and mesosphere will be an important consideration in future comparisons of NDSC instruments. Introduction In this paper we present the measurements from ROCOZ-A (rocket ozonesonde) during two instrument comparison campaigns in southern California. The two campaigns were designed to compare proposed instruments for the Network for the Detection of Stratospheric Change (NDSC) with established measurement techniques. The first intercompar- Deceased September 28, Copyright 1995 by the American Geophysical Union. Paper number 94JD /95/94JD ison was conducted at two sites in southern California in October/November 1988 [McDermid et al., 1990a]. Although it was not formally named, for this paper we have chosen to call the 1988 intercomparison "pre-stoic." The second phase of comparisons was conducted in July/August 1989 and has been called the Stratospheric Ozone Intercomparison Campaign (STOIC). STOIC provides a measure of the continuity of the old and the new measurement techniques. These comparisons also provide a means of examining the measurements from ROCOZ-A and its associated instru- ments. A portion of pre-stoic has been presented in the literature by McDermid et al. [1990a], giving an overview of the October/November 1988 ozone comparison. These re- 9209

2 9210 BARNES ET AL.: OZONE MEASUREMENTS DURING STOIC Table 1. Launch Information for the California Measurement Set ROCOZ-A ROCOZ-A Launch Solar Zenith Datasonde Launch ECC Launch Flight Date Location Time, UT Angle, deg Time, UT Time, UT 352 Oct. 31, 1988 Mugu Nov. 1, 1988 Mugu Nov. 2, 1988 Mugu Nov. 5, 1988 Mugu July 20, 1989 SNI July 24, 1989 SNI July 26, 1989 SNI July 27, 1989 SNI July 28, 1989 SNI Aug. 2, 1989 SNI Mugu, Point Mugu (34.2øN, 119.1øW); SNI, San Nicolas Island (33.3øN; 119.5øW). ROCOZ-A, rocket ozonesonde; ECC, electrochemical concentration cell. suits show excellent agreement between instruments, with ozone values grouped within 5% of the mean from 20 to 50 km. For the July/August 1989 STOIC campaign, ROCOZ-A measurements averaged from 5 to 10% higher than other instruments. As explained below, the ROCOZ-A measurements in October/November 1988 were consistent with previous comparisons with the Stratospheric Aerosol and Gas Experiment II (SAGE II) and electrochemical concentration cells (ECCs). Evidence indicates a 5-6% upward shift in ROCOZ-A ozone measurements from 1988 to To date, this shift has not been explained. Until this additional error source is identified, we recommend that the accuracy estimate for ROCOZ-A ozone densities be set at 8-10%. Flight Summary The set of ROCOZ-A measurements presented here comes from two field campaigns. Both campaigns were conducted at the U.S. Navy Pacific Missile Test Center (PMTC) at Point Mugu, California. Details of the locations and launch times for the instruments are given in Table 1. The first campaign (pre-stoic) was conducted at PMTC's mainland rocket range (34.2øN, 119.1øW) in October/ November The PMTC rocket range at San Nicolas Island (33.3øN; 119.5øW), some 80 km south of Point Mugu, was chosen for the July/August 1989 ROCOZ-A measurement series during STOIC. Problems with radar acquisition of the rocket payloads at San Nicolas Island are responsible for four partial profiles in the STOIC measurement set. Each daily set of flights included atmospheric soundings from four instruments: a ROCOZ-A ozonesonde, a Super- Loki datasonde, a standard U.S. meteorological radiosonde, and an ECC ozonesonde. The daily flight schedule for these measurements had a balloon launch within about 2 hours of the ROCOZ-A flight and a Super-Loki launch immediately preceding or following the ROCOZ-A measurements. The exact schedule of launches on each day was dictated by the availability of the rocket range and the radars (see Table 1). For the measurements in the California campaigns presented here, the ECC ozonesondes were calibrated in the laboratory against a Dasibi ozone photometer that is routinely checked at the National Institute of Standards and Technology (NIST). The calibrations were single comparisons of ozone in zero-grade air at partial pressures of about 160 nbar by each ECC and the Dasibi instrument. The ratios of the Dasibi measurements to the ECC values provided individual correction factors that were applied to the flight values from the ECCs. The ECC results were not normalized to the concurrent Dobson and Brewer total ozone measure- ments at Table Mountain. A detailed description of the ECC soundings during the July/August 1989 STOIC campaign is given by Barnes and Torres [this issue]. The ECC is carried aloft, accompanied by a standard U.S. meteorological radiosonde that measures temperature and pressure from the ground to the burst altitude of the balloon. Temperatures were also measured with rocket-borne Super- Loki datasondes [Miller and Schmidlin, 1971; Schmidlin, 1981], which were radar tracked from apogee down to about 20 km. The datasonde temperatures are merged with the balloon values at about 22 km. Above this altitude, pressures are calculated with the hypsometric equation using the datasonde temperatures and radar altitudes plus a tie-on pressure from the radiosonde. The C band tracking radars at Point Mugu and San Nicolas Island have estimated accuracies and precisions of 30 m. Estimates of the accuracies and precisions of the measurements from the ROCOZ-A flight series have been documented in the literature. The most complete set of estimates can be found in Tables 8 and 9 of Barnes et al. [1989]. These are the same values that are found in the STOIC overview paper [Margitan et al., this issue]. For ROCOZ-A ozone densities the measurement repeatability (l r) have been estimated to be 3-4%. The accuracy of ROCOZ-A ozone densities has been estimated to be 5-7%. As reported below, there is an apparent shift of ROCOZ-A ozone densities of 5-6% from October/November 1988 to July/August The set of ROCOZ-A measurements from the two STOIC campaigns is given in Tables 2 through 7. These values are presented as 2-km averages in the same manner as previous ROCOZ-A results [Barnes et al., 1987, 1989]. Table 2 gives ROCOZ-A results as ozone column amounts, which are the fundamental ozone measurements by the instruments. As before, ozone column amounts are not presented for the ECC ozonesondes (Table 2), since these values require extrapolations above the burst heights of the balloons. The ozone mixing ratios in Table 4 are calculated directly from the ozone number densities in Table 3 and the atmospheric densities in Table 7. In addition, the balloon-borne pressure measurements during STOIC were made with hypsometers, in addition to standard aneroid pressure sensors. Hypsometers have been shown to give excellent pressure measurements at altitudes

3 BARNES ET AL.: OZONE MEASUREMENTS DURING STOIC 9211 Table 2. Ozone Column Amounts From the California Measurement Set Ozone Column Amounts, molecules cm -3 Altitude, km ROCOZ-A Data () 4.91 () 5.22 () 5.29 () 6.64 () 6.53 () 6.62 () 6.63 () 6.65 () 6.32 () (17) 8.12 (17) 8.40 (17) 8.37 (17) (17) (17) 6.87 (17) 6.92 (17) 6.84 (17) 6.95 (17) (16) 9.32 (16) 9.59 (16) 9.90 (16) (16) 7.28 (16) 7.69 (16) 7.54 (16) 7.65 (16) Associated ECC Data See Table 1 for the dates and times of the flights. Read 5.24 () as 5.24 x 10 above 20 km [Parsons et al., 1984; Hilsenrath et al., 1986]. Overall, the performance of the balloon instruments at higher altitudes (above 24 km) seems significantly improved, when compared with their performance during previously reported measurement campaigns. Auxiliary pressure and temperature values from the Super-Loki datasondes in Tables 5 through 7 have been extended above the upper measurement altitudes for the ROCOZ-A ozonesondes to cover the altitude ranges of other instruments. The datasonde results have been used in separate comparisons of stratospheric and mesospheric temperatures with the mobile lidar from the Goddard Space Flight Center [McGee et al., this issue]. The ROCOZ-A and ECC ozone densities from the pre- STOIC and STOIC campaigns are shown in Figure 1. The profiles come from averages for each instrument and show the 95% confidence limits about the mean for each set. The confidence limits for the October/November 1988 measure- ments (Figure l a) are significantly larger than their counterparts for July/August A part of this increase derives from the smaller sample size for the October/November measurements. However, most of the change is due to increased atmospheric variability in October/November The July/August 1989 measurement period was planned, based on satellite climatology showing low-ozone variability during this period at northern midlatitudes (A. J. Miller, NOAA, Washington, D.C., private communication, 1988). The change in the distribution of stratospheric ozone between pre-stoic and STOIC is shown in Figure 2. Figure 2a gives the average values for the ROCOZ-A measurements in October/November 1988 and in July/August Figure 2a shows little change in ozone between the two campaigns in the altitude region near 40 km. Above and below this region, there is a decrease in stratospheric ozone during October/November, when compared with July/August. Similar seasonal changes in stratospheric ozone have been reported in the set of ROCOZ-A ozone measurements from Wallops Island, Virginia [Barnes et al., 1989], and those measurements have been tabulated in the same manner as the results presented here. The climatology from the Wallops data set is neither refined nor sophisticated. However, it does show a region of low annual variability near 40 km with greater seasonal changes at altitudes above and below this null point. The October/November and July/August profiles from the Wallops data set are shown in Figure 2b. These profiles show stratospheric ozone changes similar to Figure 2a. The measurements in Figure 2b show a second null in stratospheric ozone change near 24 km. This region of low

4 9212 BARNES ET AL.: OZONE MEASUREMENTS DURING STOIC Table 3. Ozone Number Densities From the California Measurement Set -3 Ozone Number Densities, molecules cm Altitude, km ROCOZ-A Data (12) 2.35 (12) 2.92 (12) 3.23 (12) 3.97 (12) 3. (12) 3.65 (12) 3.74 (12) 3.65 (12) 3.38 (12) (11) 7.94 (11) 8.22 (11) 8.24 (11) 9.61 (11) 9.95 (11) (11) (11) (11) (11) (10) 8.25 (10) 8.60 (10) 9.00 (10) (10) 6.67 (10) 6.72 (10) 7.21 (10) Associated ECC Data (11) 7.87 (11) 1.01 (12) 7.86 (11) 9.64 (11) 1. (12) 1.57 (12) 1.26 (12) 1.41 (12) 1.12 (12) (11) (11) (12) (11) 8.95 (11) (11) (11) 7.66 (11) (12) (12) (12) 1.42 (12) 2.05 (12) 1.48 (12) 1.81 (12) (12) 1.53 (12) (12) See Table 1 for the dates and times of the flights. Read 2.84 (12) as 2.84 x variability has been reported in the Wallops data set [Barnes et al., 1989]. The California measurements in Figure 2a do not show this null. This difference in lower-stratospheric ozone measurements does not seem to be geophysical; rather, as discussed below, we feel that the difference results from instrumental differences between October/November 1988 and July/August Temperature measurements from the California measurement set are shown in Figure 3. The figures show the 95% confidence limits about the means for both the rocket and the balloon measurements during the two phases of the STOIC campaign. Again, the October/November 1988 temperatures show greater atmospheric variability in the lower stratosphere than their July/August 1989 counterparts. In the upper stratosphere and lower mesosphere the increased profile-to-profile differences in the datasonde measurements [Miller and Schmidlin, 1971; Schmidlin, 1981] account for much of the variability in both Figures 3a and 3b. Method of Comparison Comparison Data Sets We have chosen to make comparisons of the average data sets from the instruments, rather than looking at individual instrument profiles or at arbitrarily defined reference profiles. The ROCOZ-A program has been set up to perform one-on-one comparisons with satellite instruments [Barnes, 1988; Cunnold et al., 1989; Barnes et al., 1991]. These comparisons below have been designed to examine both absolute and structural differences between the individual instruments. As explained below, the comparisons have also been designed to give estimates of the size of differences that are statistically significant in the results of the comparisons. We have chosen to examine the characteristics of ROCOZ-A during STOIC, using one-on-one comparisons with the blind data from other instruments. In the examination of the shift in ROCOZ-A ozone profiles we feel that it is important to use the 1989 measurements from other instruments that agree most closely with those in This provides consistency in the profiles that we use to determine the characteristics of ROCOZ-A ozone measurements. Since the ozone profiles from each instrument have been stored in the database using differently spaced vertical references for each instrument, we have put each data set on the same set of altitudes by interpolation. And since ozone varies exponentially with altitude above the ozone maximum, we have interpolated the logarithm of the ozone density from each instrument to standard 100-m altitude

5 BARNES ET AL.: OZONE MEASUREMENTS DURING STOIC 9213 Table 4. Ozone Mixing Ratios From the California Measurement Set Ozone Mixing Ratios, ppmv Altitude, km ROCOZ-A Associated Data ECC Data See Table 1 for the dates and times of the flights; ppmv, parts per million by volume. intervals. Along with the ozone densities the blind data set for each instrument includes estimates of the precision (profile-to-profile repeatability) and range (vertical resolution) of each measurement. These values have been placed on the same 100-m grid using linear interpolation versus altitude. Comparison Procedure We have created average data sets from the individual 100-m gridded profiles for each instrument, giving an average and standard deviation at each altitude. The standard deviations from these averaged data sets can be used to give an estimate for the precision of the comparison. In effect, this can give an estimate of the sizes of the smallest differences that can be determined from the comparison. The standard deviations for the data sets are, of course, the products of both instrumental and atmospheric variability. As presented below, the 95% confidence limits for the comparisons between instruments can be placed over the average differences. For the altitudes where the 95% confidence limits intersect zero, there is no statistically significant differences between the two instruments. However, the presentation of the comparisons in the following figures may allow a look at structural differences between instruments that cannot be found from a strict statistical analysis. It may be possible to obtain information from the following figures beyond the statement that the instruments do or do not agree within their respective error estimates. In these cases the conclusion that there is information in the comparisons beyond the strict formalism of statistical analysis is subjective. In most of the comparisons below, the standard deviations for the data sets from each instrument are roughly the same, passing the F-test [Youden, 1951; Snedecor and Cochran, 1980] at the 5% level. This criterion has been assumed in all cases, even when the standard deviations do not pass the F test. It allows a fairly simple calculation of the 95% confidence limits for the differences between data sets [Youden, 1951; Snedecor and Cochran, 1980]. A flowchart that de- scribes the procedure for selecting the proper test statistics for comparisons of averages is given in Figure 8.4 of Dowdy and Wearden [1983]. Comparisons With Other Data Sets Comparisons With Electrochemical Concentration Cell (ECC) Ozonesondes The dates and times of the ECC soundings for this comparison are listed in Table 1. The ECC ozone densities in

6 9214 BARNES ET AL.: OZONE MEASUREMENTS DURING STOIC Table 5. Atmospheric Temperatures From the California Measurement Set Atmospheric Temperatures, øk Altitude, km ROCOZ-AData Associated ECC Data See Table 1 for the dates and times of the flights. this comparison are summarized, as 2-km averages, in Table 3. We believe that the structure from 20 to 23 km in the October/November 1988 comparison in Figure 4a comes from a mismatch in the vertical resolutions of the rocket and balloon instruments. The spike below 23 km comes from an atmospheric structure that was seen by the balloon instruments but not by the rockets. The ECC, with its 300-m vertical resolution [Barnes and Torres, this issue], is capable of resolving vertical structures in atmospheric ozone that cannot be seen by ROCOZ-A, with its 4-km vertical resolution [Barnes et al., 1986]. The 10% shift between the instruments from 21 to 23 km does not appear to be due to an outlier in one of the data sets, since the 95% confidence limits for the comparison remain constant at about 8% over the entire altitude range of the comparison. The presence of a persistent atmospheric structure seems much more likely than an altitude-dependent measurement artifact in four ECC instruments. In addition, the comparisons in Figures 4a and 4b show a 5-6% shift between the rocket and the balloon instruments between October/November 1988 and July/August Strictly speaking, the 95% confidence limits for the two comparisons overlap at almost every altitude, indicating no statistical difference between the two comparisons. How- ever, an independent comparison with SAGE II, presented below, also shows a shift between instruments between pre-stoic and STOIC. As shown below, the altitude-dependent shape of the July/August 1989 comparison in Figure 4b is also duplicated in comparisons with the JPL lidar and with the Millitech microwave. All of the comparisons, except for the microwave, show the same sharp tail between 20 and 21 km that is seen in Figure 4b. These comparisons indicate the presence of an atmospheric structure below 21 km that cannot be seen by ROCOZ-A with its 4-km vertical resolution. Overall, the comparisons in Figures 4a and 4b are consistent with previous ROCOZ-A/ECC comparisons [Barnes et al., 1987, 1989]. In general, the rocket instrument has averaged about 5% higher than the balloon sondes at altitudes below 25 km. However, for the previous comparisons the 95% confidence limits were greater than 5%. From a statistical standpoint, ROCOZ-A/ECC comparisons cannot resolve instrument differences of less than about 8%. Still, the ECC comparisons retain an overall agreement at the 10% level or better, with ROCOZ-A having measurements about 5% higher than the ECCs. This is the historical level of agreement of ROCOZ-A with other instruments [Barnes et al., 1989].

7 BARNES ET AL.: OZONE MEASUREMENTS DURING STOIC 9215 Table 6. Atmospheric Pressures From the California Measurement Set Atmospheric Pressures, mbar Altitude, km ROCOZ-A Data ! Associated ECC Data See Table 1 for the dates and times of the flights. Comparisons With Stratospheric Aerosol and Gas Experiment (SAGE) II Since the use of ROCOZ-A in the validation of SAGE II ozone measurements [Cunnold et al., 1989], comparisons with SAGE II have become a primary part of the characterization of ROCOZ-A. The validation comparison, shown in Figure 5a, links ROCOZ-A measurements with SAGE II in the spring and summer of 1985, the first year of SAGE II measurements. The planning of the comparison dates for the two phases of STOIC included the incorporation of SAGE II overpasses near southern California. As a result, STOIC has provided an opportunity to continue the link between these rocket and satellite instruments. The comparison in Figure 5a was derived from five pairs of ROCOZ-A/SAGE II profiles, with three pairs at northern midlatitudes and two at the equator [Cunnold et al., 1989]. The overall agreement between instruments was good, with ROCOZ-A averaging within 1% of the SAGE II over the 34 comparison altitudes. The greatest difference between instruments (ROCOZ-A 6% higher) occurred at 45 km. This agreement between instruments has been duplicated in an equatorial comparison of ROCOZ-A and SAGE II in March/ April 1985 [Barnes et al., 1991]. The equatorial comparison included sets of 7 ROCOZ-A and 14 SAGE II profiles taken over a period of 3 weeks. Two of the closely paired rocket/ satellite measurements from Barnes et al. [1991] were also part of the SAGE II ozone validation. The comparison in Figure 5b includes six ROCOZ-A and three SAGE II measurements during July/August A listing of the dates and locations for the SAGE II profiles is given in Table 8, along with the minimum and maximum altitudes for the SAGE II ozone measurements. Table 8 also gives summary information for all of the comparison profiles used in this paper. We believe that Figure 5b shows a relative shift between ROCOZ-A and SAGE II when compared with the SAGE II ozone validation in 1985 (Figure 5a). This change resembles the shift of ROCOZ-A with the ECCs in Figures 4a and 4b. The SAGE II and ECC comparisons are both consistent with a shift of about 5-6% in the ROCOZ-A profiles. In addition, the comparison in Figure 5b also shows a sharp tail at 20 km, in agreement with the ECC results in Figure 4b. The ROCOZ-A/SAGE II comparisons in Figure 5 do not include the October/November 1988 SAGE II measurements that would make the optimum comparison in place of Figure 5a. A latitude-dependent ozone gradient in the SAGE II ozone densities [McDermid et al., 1990a; McGee et al., 1990] has made the selection of an appropriate SAGE II data

8 9216 BARNES ET AL' OZONE MEASUREMENTS DURING STOIC Table 7. Atmospheric Densities From the California Measurement Set Atmospheric Densities, molecules cm -3 Altitude, km ROCOZ-A Data () 1.98 () 2.00 () 1.93 () 2.03 () 1.99 () 1.98 () 1.97 () 1.98 () 2.00 () (17) (17) (17) (17) (17) 7.54 (17) 7.44 (17) 7.49 (17) 7.53 (17) 7.61 (17) (16) 7.99 (16) 8.19 (16) 8.10 (16) 9.38 (16) 9.20 (16) 9.03 (16) 9.11 (16) 9.09 (16) 9.36 (16) (16) (15) 9.62 (15) 9. (15) 9.46 (15) (15) 9.29 (15) 9.34 (15) 9.13 (15) Associated ECC Data (19) 2.05 (19) 2.06 (19) 2.00 (19) 2.00 (19) 1.99 (19) 1.99 (19) 1.99 (19) 1.99 (19) 2.00 (19) () 8.71 () 8.93 () 8.74 () 8.93 () 8.84 () 8.90 () 8.82 () 8.83 () 8.71 () (17) (17) (17) (17) (17) 7.47 (17) 7.46 (17) 7.49 (17) 7.56 (17) 7.57 (17) See Table 1 for the dates and times of the flights. Read 1.98 () as 1.98 x 10 GO SOUTHERN CRLIFORNIR - OCT/NOV MERN 95 CONFIDENCE SOUTHERN CRLIFORNIR - JUL/RUG MERN + 9S CONFIDENCE ¾ 48 W - ROCOZ-R - ROCOZ-R ha, - ECC OZONE DENSITY (MOLECULE/CM 3 ) b OZONE DENSITY (MOLECULE/CM 3 ) o Figure 1. ROCOZ-A (rocket ozonesonde) and electrochemical concentration cell (ECC) ozone density profiles from the pre-stoic and the Stratospheric Ozone Intercomparison Campaigns (STOIC). (a) Average profiles from October to November 1988 with 95% confidence limits. (b) Average profiles from July to August 1989 with 95% confidence limits.

9 BARNES ET AL.' OZONE MEASUREMENTS DURING STOIC e,,,,,a ROCOZ-fi fiveraged OZONE PROFILES SOUTHERN CALIFORNIA 68 ROCOZ-A AVERAGED OZONE PROFILES WALLOPS ISLAND, Vfi 4o v 48 ROCOZ-A -e - OCT/NOV [988 -ar- - JUL/RUG 1989 _.J ROCOZ-A -e- - OCT/NOV [983/5 - - JUL/RUG 1983/5 5 TM OZONE DENSITY (MOLECULE/CM 3) b OZONE DENSITY (MOLECULE/CM 3) I 13 5 Figure 2. ROCOZ-A ozone density profiles from pre-stoic and STOIC compared with published ROCOZ-A results. Both sets of profiles show more stratospheric ozone in the summer than in the autumn with a node near 40 km. (a) Average profiles from the pre-stoic and STOIC campaigns, (b) Average profiles from Wallops Island, Virginia, set for this period impractical. Instead, the October/ November 1988 SAGE II data are presented at six altitudes in Figure 6. The figure shows a significant gradient in stratospheric ozone between 30 ø and 35øN for altitudes below 45 km. The dates and locations of SAGE II profiles in Figure 6 are listed in Table 8. The latitude-dependent gradient in the SAGE II measurements is present in both the southward (sunrise) SAGE II sweep from October 29 to November 2 and the northward (sunset) SAGE II sweep from November 3 to 8. The ROCOZ-A values in Figure 6 fall at or below the SAGE II measurements at all six altitudes. This agreement is consistent with the ROCOZ-A/SAGE II comparison at the other altitudes not shown in Figure 6. The ROCOZ-A/SAGE II comparison in October/November 1988 is fundamentally the same as the 1985 SAGE II ozone validation. However, the July/August 1989 ROCOZ-A/SAGE II comparison shows a 5-6% bias (ROCOZ-A higher) between the rocket and the satellite instruments. Combined with the ECC comparison, this indicates a shift in the ROCOZ-A densities from October/November 1988 to July/August Comparisons With the Jet Propulsion Laboratory (JPL) Lidar Figure 7 gives comparisons of the full sets of ROCOZ-A and JPL lidar measurements during pre-stoic and STOIC. Figure 7a compares the set of four ROCOZ-A measurements in October/November 1988 with the ten JPL lidar profiles from the same period (see Table 8). Figure 7b compares the six ROCOZ-A profiles with fourteen JPL measurements in July/August Comparisons are made between the blind data sets. The positive shift of the ROCOZ-A measurements from Figure 7a to Figure 7b is about 8-10%. This shift has nearly twice the magnitude of the ROCOZ-A change relative to the ECCs and SAGE II. The 95% confidence limits for the comparisons, particularly those in Figure 7a, are such that a 5-6% change in ROCOZ-A, with respect to the JPL lidar, cannot be ruled out. Figure 7b also shows the sharp "tail" reported in the ECC and SAGE II comparisons. Again, we feel that this tail represents a structure in the lower stratosphere that cannot be seen in the rocket profiles with their 4-km vertical resolution. 68 SOUTHERN CALIFORNIA - OCT/NOV SOUTHERN CALIFORNIA - JUL/RUG 1989 AN E 540 w - 38 v48 38 MEAN + 95 CONFIDENCE 28 ONDE m/i I I I I I I TEMPERATURE (DEGREES K) Q: 28 ', - DATASONDE ', - RADIOSONDE TEMPERRTURE (DEGREES K) Figure 3. Datasonde and radiosonde temperature profiles from the pre-stoic and STOIC campaigns. (a) Average profiles from October to November 1988 with 95% confidence limits. (b) Average profiles from July to August 1989 with 95% confidence limits.

10 92 BARNES ET AL' OZONE MEASUREMENTS DURING STOIC OF ROCOZ-8 FROM ECC OZONESONDES DURING PRE-STOIC - OCTOBERYNOVEMBER 1988 OF ROCOZ-R FROM ECC OZONESONDES DURING STOIC - JULYyRUGUST 1989 : 38 _1 RO - ROCOZ-R (4 PROFILES) EC - ECC OZONESONDE (4 PROFILES) RO - ROCOZ-R m (6 PROFILES) 38 EC - ECC OZONESONDE (6 PROFILES) 0 S - 0! CONFIDENCE 0 S CONFIDENCE Figure 4. A Comparison of ROCOZ-A and ECC ozone profiles in the altitude region where measurements overlap. ROCOZ-A shows a positive shift with respect to the ECCs from 1988 to See text for details. (a) The percent difference of ROCOZ-A from ECCs in October/November 1988 with 95% confidence limits. (b) The percent difference of ROCOZ-A from ECCs during July/August 1989 with 95% confidence limits. The structure near 30 km in Figure 7b is not considered to be geophysical in nature. It appears only in comparisons with the original (blind) data set for the JPL lidar. The region near 30 km marks the transition between the high-gain and the low-gain channels for the lidar, where the very large numbers of returning photons in the high-gain channel can cause an undercount in the photon-counting system. These large photon fluxes in photomultiplier tubes can be treated statistically [Evans, 1955], and the transition altitude for the switch between channels can be changed to remove this structure from the lidar data set [McDermid and Walsh, 1990]. The structure near 30 km was removed by changes in the lidar data reduction [McDermid and Walsh, 1990] and is not found in comparisons with the refined data set from the JPL lidar. There is a substantial increase in the 95% confidence limits for the ROCOZ-A/JPL comparisons at altitudes above 40 km. This change results from the increased profile-to-profile differences for the lidar measurements at these altitudes. The repeatability of the ROCOZ-A profiles remains nearly constant from 40 to 50 km (see Figure 1). The confidence limits for the comparison in Figure 7b are large enough to mask the skew in the comparison from 40 to 50 km. The confidence limits in Figure 7a are greater than in Figure 7b, due in part to the smaller data sets. However, a portion of the skew in Figure 7b is considered to be instrumental. The empirical background correction for the JPL lidar measurements [McDermid et al., 1990b] was modified after the blind STOIC comparisons, and the refined lidar profiles show increased ozone amounts at altitudes above 40 km [McDermid and Walsh, 1990]. The use of the OF ROCOZ-R FROM SAGE II DURING SAGE II VALIDATION OF ROCOZ-A FROM SAGE II DURING STOIC - JULY/AUGUST 1989 s8 - j MEASUREMENTS FROM RO m ROCOZ-R = SA - SAGE II H R0 - FROM- 48 RO - ROCOZ-A m (6 PROFILES) 38 SA - SAGE II H (3 PROFILES) - SA - 8 S X CONFIDENCE 8 S _ X CONFIDENCE Figure 5. A Comparison of ROCOZ-A ozone profiles with Stratospheric Aerosol and Gas Experiment (SAGE) II in 1985 and ROCOZ-A shows a positive shift with respect to SAGE II from 1985 to See text for details. (a) The percent difference of ROCOZ-A from SAGE II during the SAGE II validation program [from Cunnold et al., 1989] with 95% confidence limits. (b) Percent difference of ROCOZ-A from SAGE II during July/August 1989 with 95% confidence limits.

11 BARNES ET AL.: OZONE MEASUREMENTS DURING STOIC 9219 Table 8. Instrument Sample Sets ROCOZ-A ECC JPL Lidar Microwave SAGE II Date Min, Max, Min, Max, km km km km Min, Max, Min, Max, km km km km Min, Max, Lat, Long, Sunrise/ km km deg deg Sunset Oct. 25, 1988 Oct. 26, 1988 Oct. 27, 1988 Oct. 28, 1988 Oct. 29, 1988 Oct. 30, Oct. 31, Nov. 1, Nov. 2, Nov. 3, Nov. 4, 1988 Nov. 5, Nov. 6, 1988 Nov. 7, Nov. 8, Nov. 9, 1988 October/November 1988 Campaign July/August 1989 Campaign July 20, July 21, July 22, July 23, July 24, July 25, July 26, July 27, July 28, July 29, July 30, July 31, Aug. 1, Aug. 2, øN 106.1øW sunrise øN 130.3øW sunrise øN 110.4øW sunrise øN 134.7øW sunrise øN 114.9øW sunrise øN 119.6øW sunrise øN 124.3øW sunrise øN 1.8øW sunset øN 123.6øW sunset øN 104. IøW sunset øN 128.4øW sunset IøN 108.9øW sunset øN 133.2øW sunset øN 113.6øW sunset IøN 1.2øW sunset øN 116.1øW sunrise øN 120.7øW sunrise IøN 125.3øW sunrise The table entries include the maximum and minimum altitudes for each profile. The SAGE II entries also include the location of the measurements. On some days in October/November 1988 there were more than one SAGE II measurement. refined JPL data set removes most of the skew above 40 km in the comparison in Figure 7b. However, the application of the new data reduction procedures to the 1988 lidar data should also significantly change the comparison in Figure 7a and the results of McDermid et al. [1990a] for altitudes above 40 km. Comparisons With the Mi!!itech Microwave The 1989 STOIC campaign marked the first field comparison for the Millitech ozone microwave [Parrish et al., 1990]. For these comparisons the microwave used auxiliary temperature profiles from the Goddard lidar [McGee et al., 1990] in its data reduction scheme. Revisions to two of the lidar temperature profiles required changes to two of the microwave ozone profiles from the blind data set. These changes were minor and the blind and refined data sets for the Millitech microwave are essentially the same. The blind comparison in Figure 8a gives results identical to a comparison with the refined data set. The primary ozone values from the microwave measurements are mixing ratios versus pressure. These are the same ozone units for the ultraviolet-visible and infrared remote sensors planned for the NDSC. For the STOIC comparison the mixing ratios were converted into densities and the pressures were converted into altitudes using temperaturepressure-altitude values from the balloon sondes and the Goddard lidar. These conversions add an error source to the comparisons. For satellite comparisons, this added error can be of the order of 5-10% of the ozone values in the upper stratosphere [Olivero and Barnes, 1991]. For an additional comparison, the ROCOZ-A ozone densities (versus altitude) were converted into mixing ratios (versus pressure). The results of this comparison of mixing ratios show differences that are fundamentally the same as Figure 8a. This is expected, since the auxiliary ROCOZ-A temperature and pressure profiles are very close to those used by the microwave. Both data sets use the standard U.S. meteorological radiosonde to around 20 km. Above that altitude the datasonde measurements agreed well with the values from the temperature lidar [McGee et al., this issue]. Over the altitude range from 20 to 45 km (see Figure 8a), ROCOZ-A averaged 8% higher than the microwave. Below the ozone density maximum at 25 km there is a skew between the two instruments with ROCOZ-A values about 12% higher at 20 km. Similar differences below the density

12 9220 BARNES ET AL OZONE MEASUREMENTS DURING STOIC S 35.5 KM 0.Z4 S. mu Zl.d 0 1 SRGE II ROCOZ- m.ee I I I I SRGE II d ROCOZ-R Z 0. LRTITUDE (DEGREES NORTH) LRTITUDE (DEGREES NORTH) 30. x 45.5 KM 24. x 30.5 KM lb. u >- w i- j I. dg I;:'. m u Zl.i c::1o Z N _ SRGE II b ROCOZ-R SRGE II ROCOZ-R z o N o LRTITUDE (DEGREES NORTH) LRTITUDE (DEGREES NORTH) s x u KM - & KM 30. u w mu 2.4 zw z w o Z 0 N 0 SRGE II ROCOZ-R SRGE II 10 z ROCOZ-R o N 20 ; ; CATZTUDE (DEGREES NORTH) CATZTUD (DEGREES NORTH) Figure 6. A Comparison of ROCOZ-A with SAGE II at six altitudes during October/November The SAGE II gradient in midstratospheric ozone with latitude has made comparisons of averages difficult. ROCOZ-A does not appear to be significantly high with respect to SAGE II during this period. See text for details. (a) SAGE II and ROCOZ-A values at 50.5 km. (b) Same at 45.5 km. (c) Same at 40.5 km. (d) Same at 35.5 km. (e) Same at 30.5 km. (f) Same at 25.5 km. maximum can be seen in the other 1989 comparisons presented above. Above 45 km the structure in Figure 8a may be of geophysical rather than of instrumental origin. The microwave measurements in this comparison were taken during the night, while the ROCOZ-A profiles were taken near local noon. There is a diurnal cycle in mesospheric and upper stratospheric ozone, with greater ozone amounts predicted by photochemical models at night [Herman, 1979; Allen et al., 1984]. The models show about a 6% difference in ozone at 50 km, increasing to a 25% difference at 55 km (see, for example, Herman [1979]. The difference curve in Figure 8a follows the pattern predicted by the models. Above 45 km the difference curve shows a divergence between instruments, with the nighttime (microwave) measurements becoming increasingly greater than the daytime (ROCOZ-A) values. The magnitude of the divergence is approximately correct, although the divergence does not continue above 50 km. However, these comparisons cannot rule out the possibility that there may be an instrumental rather than a geophysical source for the skew between instruments above 45 km. Comparisons With the SAGE II Zonal Mean Since SAGE II measures at sunrise and sunset, it also experiences the effects of the diurnal ozone cycle in the upper stratosphere and the mesosphere [Chu, 1989]. In a comparison of SAGE II with the Solar Mesospheric Ex-

13 BARNES ET AL.' OZONE MEASUREMENTS DURING STOIC 9221 OF ROCOZ-8 FROM JPL LIDSR DURING PRE-STOIC - OCTOBER/NOVEMBER 1988 OF ROCOZ-R FROM JPL LIDRR DURING STOIC - JULY/AUGUST 1989 v w 48 RO m ROCOZ-R (4 PROFILES) D 30 JL - JPL LIDRR (10 PROFILES) v 48 RO m ROCOZ-R (6 PROFILES) D 30 JL - JPL LIDRR (14 PROFILES) 10 0 S -to 0 lb + 95 CONFIDENCE 0 S -to X CONFIDENCE Figure 7. A Comparison of ROCOZ-A and the Jet Propulsion Laboratory (JPL) lidar during the pre-stoic and STOIC campaigns. ROCOZ-A shows a positive shift with respect to the lidar from 1988 to See text for details. (a) The percent difference of ROCOZ-A from the JPL lidar in October/ November 1988 with 95% confidence limits. (b) The percent difference of ROCOZ-A from the JPL lidar in July/August 1989 with 95% confidence limits. plorer (SME) ultraviolet spectrometer at 48 km, SAGE II sunset ozone values were found to be 5% higher than the midafternoon SME ozone measurements [Rusch et al., 1990]. This is consistent with results from the photochemical model [Rusch et al., 1990] that predicted SAGE II measurements to be equivalent to measurements at night. The diurnal ozone cycle should also be seen in ROCOZ- A/SAGE II comparisons. The presence of the predicted skew above 45 km in the SAGE II ozone validation (see Figure 5a) is a matter of personal interpretation. However, the shape of the comparison in Figure 5b shows no sign of the predicted diurnal effect. This comparison was made with only three SAGE II measurements, which may be too small a number to examine structure in the SAGE II profiles. To create a larger SAGE II data set, we have used the SAGE II zonal mean for 30ø-40øN for July/August The zonal mean contains over 80 profiles with nearly equal numbers of sunrise and sunset ozone measurements. There is a striking similarity to the difference curves in Figures 8a and 8b. The 95% confidence limits for the SAGE II zonal mean comparison in Figure 8b are very small, due to the large number of profiles in the zonal mean data set. The shapes of the two difference curves are essentially identical, with a 3% shift between the two figures (SAGE II higher than the microwave). With regard to the diurnal ozone cycle in the upper stratosphere and the mesosphere, the SAGE I! zonal mean shows the same expected divergence from ROCOZ-A above 45 km that is shown in the microwave comparison (Figure 8a). The comparison with the SAGE II zonal mean should be OF ROCOZ-R FROM MICRONAVE DURING 60 STOIC - JULY/AUGUST 1989 OF ROCOZ-R FROM SAGE II ZONAL MEAN - JULY/AUGUST x 40 RO - ROCOZ-R x 40 RO - ROCOZ-R D 30 _J (6 PROFILES) MM - MICRONAVE (6 PROFILES) D 30 (6 PROFILES) SR - SAGE II ZONAL MEAN N JUL/AUG o IO O S -IO O IO + 95Z CONFIDENCE O S -IO O IO + 95Z CONFIDENCE Figure 8. A Comparison of ROCOZ-A with microwave measurements and the SAGE II zonal mean during July/August The two comparison profiles are exceptionally close, both in shape and in absolute magnitude. See text for details. (a) The percent difference of ROCOZ-A from the full 14-profile microwave data set with 95% confidence limits. (b) The percent difference of ROCOZ-A from the SAGE II 30ø-40øN zonal mean for July/August The error bars give the 95% confidence limits.

14 9222 BARNES ET AL.: OZONE MEASUREMENTS DURING STOIC viewed with caution, since the comparison implies that the local ROCOZ-A and microwave measurements are representative of the global set of measurements for the 30ø-40øN latitude band and for the July/August 1989 time period. For this reason we consider the comparison in Figure 8b to be secondary to the other instrument comparisons presented here. The comparisons in Figures 8a and 8b give evidence for the presence of a diurnal ozone cycle above 45 km. However, the blind JPL lidar comparison in Figure 7b shows a skew with ROCOZ-A in the opposite direction from the model predictions. For the comparison with the refined JPL lidar profiles the skew with ROCOZ-A above 45 km is nearly removed, but the refined lidar measurements still do not show the nighttime slope shown by the Millitech microwave or by the SAGE II zonal mean. There is no complete consistency in the higher-altitude ozone measurements from STOIC. Finally, the idea of a diurnal ozone cycle in the upper stratosphere is based on the calculations of photochemical models. The models require the use of measurements to verify their predictions [Olivero and Barnes, 1991] in the same manner that the measurements require models to explain possible differences between instruments. In the upper stratosphere and mesosphere the connection between model predictions and measurements is circular. Eventually, improvements in both the models and the measurements may bring them into a harmonious partnership. Concluding Remarks For the microwave and lidar instruments, STOIC has provided an important development opportunity in their preparation for operation with the NDSC. In the time since STOIC the microwave and lidar instruments have all been modified. For them, STOIC provided a "snapshot" of their performance characteristics in It gives an estimate of their future performance in the NDSC. The participation of ROCOZ-A in STOIC provided comparison profiles to aid these instruments in their continuing development. The characterization of these proposed network instruments is the primary goal of the STOIC campaigns. The participation of the rocket ozonesonde in STOIC has also added informa- tion on the accuracy and precision of the ozone profiles from ROCOZ-A. It is the continuing characterization of ROCOZ-A profiles that gives the instrument value in comparisons, such as STOIC. The STOIC comparisons show a shift of 5-6% in ROCOZ-A ozone densities (ROCOZ-A higher) from the October/November 1988 campaign to the July/August 1989 campaign. A detailed examination of the instrument calibrations, telemetry records, and data reduction has not revealed the source of this shift. Until this new error source is determined, we recommend that the accuracy estimate for ROCOZ-A ozone measurements be increased from 5-7% to 8-10%. The STOIC results are consistent with the results of previous comparisons with in situ ozone instruments, where ROCOZ-A values agreed with other instruments at the 10% level or better [Hilsenrath et al., 1986; Watson et al., 1988; Barnes et al., 1989]. There are model-based predictions of diurnal changes in upper stratospheric and mesospheric ozone [Herman, 1979; Allen et al., 1984; Rusch et al., 1990] with greater ozone amounts predicted at night. The pattern of this day-night effect has been demonstrated during STOIC in the upper altitude comparisons between ROCOZ-A and the Millitech microwave and between ROCOZ-A and the SAGE II zonal mean. However, the results from the nighttime measurements from the JPL lidar showed ozone scale heights in the upper stratosphere with the pattern predicted for daytime measurements and with the pattern shown by the ROCOZ-A profiles. Until these conflicting results are resolved, there is no complete consistency in the higher-altitude ozone measurements at STOIC. References Allen, M., J. I. Lunine, and Y. L. Yung, The vertical distribution of ozone in the mesosphere and lower thermosphere, J. Geophys. Res., 89, , Barnes, R. A., Changes in SBUV ozone profiles near Natal, Brazil, from 1979 to 1985, J. Geophys. Res., 93, , Barnes, R. A., and A. L. Torres, Electrochemical concentration cell ozone soundings at two sites during the STOIC campaign, J. Geophys. Res., this issue. Barnes, R. A., A. C. Holland, and H. S. Lee, An improved rocket ozonesonde (ROCOZ-A), 2, Preparation of stratospheric ozone profiles, J. Geophys. Res., 91, 14,521-14,531, Barnes, R. A., A. C. Holland, and V. J. H. Kirchhoff, Equatorial ozone profiles from the ground to 52 km during the southern hemisphere autumn, J. Geophys. Res., 92, , Barnes, R. A., M. A. Chamberlain, C. L. Parsons, and A. C. Holland, An improved rocket ozonesonde (ROCOZ-A), 3, Northem midlatitude ozone measurements from 1983 to 1985, J. Geophys. Res., 94, , Barnes, R. A., L. R. McMaster, W. P. Chu, M.P. McCormick, and M. E. Gelman, Stratospheric Aerosol and Gas Experiment II and ROCOZ-A ozone profiles at Natal, Brazil: A basis for comparison with other satellite instruments, J. Geophys. Res., 96, , Chu, D. A., The interpretation of SAGE II ozone measurements in the lower mesosphere, Ph.D. thesis, Georgia Inst. Technol., Atlanta, Ga., Cunnold, D. M., W. P. Chu, R. A. Barnes, M.P. McCormick, and R. E. Veiga, Validation of SAGE II ozone measurements, J. Geophys. Res., 94, , Dowdy, S., and S. Wearden, Statistics for Research, p. 196, John Wiley, New York, Evans, R. D., The Atomic Nucleus, p. 785, McGraw-Hill, New York, Ferrare, R. A., et al., Lidar measurements of stratospheric temperature during STOIC, J. Geophys. Res., this issue. Herman, J. R., The response of stratospheric constituents to a solar eclipse, sunrise, and sunset, J. Geophys. Res., 84, , Hilsenrath, E., et al., Results from the balloon ozone intercomparison campaign (BOIC), J. Geophys. Res., 91, 13,137-13,152, Margitan, J. J., et al., Stratospheric Ozone Intercomparison Campaign (STOIC) 1989: Overview, J. Geophys. Res., this issue. McDermid, I. S., and D. T. Walsh, Results from the JPL-TMF ozone lidar system during STOIC '89, paper presented at the Meeting on Optical Remote Sensing of the Atmosphere, Opt. Soc. of Am., Incline Village, Nev., Feb McDermid, I. S., et al., Comparison of ozone profiles from groundbased lidar, electrochemical concentration cell balloon sonde, ROCOZ-A rocket ozonesonde, and Stratospheric Aerosol and Gas Experiment satellite measurements, J. Geophys. Res., 95, 10,037-10,042, 1990a. McDermid, I. S., S. M. Godin, and L. O. Lindqvist, Ground-based laser DIAL system for long-term measurements of stratospheric ozone, Appl. Opt., 29, , 1990b. McGee, T. J., P. Neuman, R. Ferrare, D. Whiteman, J. Butler, J. Burris, S. Godin, and I. S. McDermid, Lidar observations of ozone changes induced by subpolar air mass motion over Table Mountain, California (34.4øN), J. Geophys. Res., 95, 20,527-20,530, McGee, T. J., Ferrare, R. A., D. N. Whiteman, J. J. Butler, J. F.

15 BARNES ET AL.: OZONE MEASUREMENTS DURING STOIC 9223 Burris, and M. A. Owens, Lidar measurements of stratospheric ozone during the STOIC campaign, J. Geophys. Res., this issue. Miller, A. J., and F. J. Schmidlin, Rocketsonde repeatability and stratospheric variability, J. Appl. Meteorol., 10, , Olivero, J. J., and R. A. Barnes, Satellite ozone comparisons: Effects of pressure and temperature, J. Geophys. Res., 96, , Parrish, A., B. J. Connor, and J.-J. Tsou, A microwave system for remote sensing of ozone, paper presented at the Meeting on Optical Remote Sensing of the Atmosphere, Opt. Soc. of Am., Incline Village, Nev., Feb Parsons, C. L., G. A. Norcross, and R. L. Brooks, Radiosonde pressure sensor performance: Evaluation using tracking radars, J. Atmos. Ocean. Technol., 1, , Rusch, D. W., M.P. McCormick, R. T. Clancy, and J. M. Zawodny, A comparison of SME and SAGE II ozone densities near the stratospause, J. Geophys. Res., 95, , Schmidlin, F. J., Repeatability and measurement uncertainty of the United States meteorological rocketsonde, J. Geophys. Res., 86, , Snedecor, G. W., and W. G. Cochran, Statistical Methods, 7th ed., pp , Iowa State University Press, Ames, Watson, R. T., Ozone Trends Panel, M. J. Prather, Ad Hoc Theory Panel, and M. J. Kurylo, and NASA Panel for Data Evaluation, Present state of knowledge of the upper atmosphere 1988: An assessment report, NASA Ref. Publ., NASA-RP-1208, 208 pp., Aug Youden, W. J., Statistical Methods for Chemists, pp , John Wiley, New York, R. A. Barnes, ManTech, Incorporated, Building F-160, Wallops Island, VA A. P. Grothouse, Computer Sciences Corporation, P.O. Box 37, Wallops Island, VA C. L. Parsons, NASA Goddard Space Flight Center, Code 972, Wallops Flight Facility, Wallops Island, VA (Received September 8, 1994; revised September 16, 1994; accepted September 22, 1994.)