Stratospheric water vapor trends over Boulder, Colorado: Analysis of the 30 year Boulder record

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 116,, doi: /2010jd015065, 2011 Stratospheric water vapor trends over Boulder, Colorado: Analysis of the 30 year Boulder record Dale F. Hurst, 1,2 Samuel J. Oltmans, 2 Holger Vömel, 3 Karen H. Rosenlof, 4 Sean M. Davis, 1,4 Eric A. Ray, 1,4 Emrys G. Hall, 1,2 and Allen F. Jordan 1,2 Received 16 September 2010; revised 16 November 2010; accepted 23 November 2010; published 26 January [1] Trend analyses are presented for 30 years ( ) of balloon borne stratospheric water vapor measurements over Boulder, Colorado. The data record is broken into four multiple year periods of water vapor trends, including two that span the wellexamined but unattributed period of stratospheric water vapor growth. Trends are determined for five 2 km stratospheric layers (16 26 km) utilizing weighted, piecewise regression analyses. Stratospheric water vapor abundance increased by an average of 1.0 ± 0.2 ppmv (27 ± 6%) during with significant shorter term variations along the way. Growth during period 1 ( ) was positive and weakened with altitude from 0.44 ± 0.13 ppmv at km to 0.07 ± 0.07 ppmv at km. Water vapor increased during period 2 ( ) by an average 0.57 ± 0.25 ppmv, decreased during period 3 ( ) by an average 0.35 ± 0.04 ppmv, then increased again during period 4 ( ) by an average 0.49 ± 0.17 ppmv. The diminishing growth with altitude observed during period 1 is consistent with a water vapor increase in the tropical lower stratosphere that propagated to the midlatitudes. In contrast, growth during periods 2 and 4 is stronger at higher altitudes, revealing contributions from at least one mechanism that strengthens with altitude, such as methane oxidation. The amount of methane oxidized in the stratosphere increased considerably during , but this source can account for at most 28 ± 4%, 14 ± 4%, and 25 ± 5% of the net stratospheric water vapor increases during , , and , respectively. Citation: Hurst, D. F., S. J. Oltmans, H. Vömel, K. H. Rosenlof, S. M. Davis, E. A. Ray, E. G. Hall, and A. F. Jordan (2011), Stratospheric water vapor trends over Boulder, Colorado: Analysis of the 30 year Boulder record, J. Geophys. Res., 116,, doi: /2010jd Introduction [2] Atmospheric water vapor plays a critical role in the Earth s radiation budget as the predominant attenuator of outgoing longwave radiation. It is the most important gaseous source of infrared opacity in the atmosphere, accounting for about 60% of the natural greenhouse effect for clear skies [Kiehl and Trenberth, 1997]. Atmospheric water vapor also provides the largest positive feedback in model projections of climate change [Held and Soden, 2000]. [3] Small changes in stratospheric water vapor abundance have a large impact on the Earth s radiation budget [Forster and Shine, 2002]. A recent study by Solomon et al. [2010] 1 Cooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder, Colorado, USA. 2 Global Monitoring Division, NOAA Earth System Research Laboratory, Boulder, Colorado, USA. 3 Meteorologisches Observatorium Lindenberg, Deutscher Wetterdienst, Lindenberg, Germany. 4 Chemical Sciences Division, NOAA Earth System Research Laboratory, Boulder, Colorado, USA. Copyright 2011 by the American Geophysical Union /11/2010JD estimated that the 10% decrease in lower stratospheric water vapor during counteracted 25% of the global surface temperature increase that would have been caused by well mixed greenhouse gases and aerosols over the last decade. Since stratospheric water vapor is an important driver of decadal global surface climate change [Solomon et al., 2010], monitoring trends in its abundance is an essential part of climate change detection and prediction. [4] Water vapor abundance in the northern midlatitude lower stratosphere fluctuates as variations in tropical lowerstratospheric water vapor (e.g., the seasonal cycle) are advected upward and poleward [Mote et al., 1996]. Since the mean ages of stratospheric air masses generally increase with altitude [e.g., Hall and Plumb, 1994; Boering et al., 1996], signals originating in the tropical lower stratosphere, including longer term water vapor trends, tend to dissipate with increasing altitude in both the tropical and extratropical stratosphere. Stratospheric water vapor abundance is also influenced by methane oxidation above the tropopause [Bates and Nicolet, 1950; le Texier et al., 1988]. This photochemical water vapor source is strongest at higher altitudes in the tropics and falls off sharply at lower altitudes. Though not specifically a function of mean age, the contributions of this source to midlatitude stratospheric water vapor tend to 1of12

2 increase with altitude. These two major influences on midlatitude water vapor abundance have opposing altitude dependences, making observations of any altitude dependent stratospheric water vapor changes helpful in identifying the causal mechanism(s). [5] Oltmans et al. [2000] reported that stratospheric water vapor increased at a rate of ppmv yr 1 (1 1.2% yr 1 ) over Boulder, Colorado (40 N, 105 W) during based on 190 balloon flights of the NOAA frost point hygrometer (FPH). The trend strength diminished somewhat with altitude from 16 to 28 km, suggesting a tropical water vapor increase as the source, but several studies of long term changes in cold point temperatures near the tropical tropopause found cooling instead of warming trends for this period [e.g., Simmons et al., 1999; Randel et al., 2000; Zhou et al., 2001; Seidel et al., 2001; Randel et al., 2004; Fueglistaler and Haynes, 2005]. The substantial 0.27 ppmv CH 4 added to the troposphere during , even if completely oxidized to water vapor in the stratosphere, could elicit only about half of this stratospheric water vapor increase [Kley et al., 2000; Rosenlof et al., 2001; Randel et al., 2004]. [6] A recent reanalysis of the Boulder water vapor data set by Scherer et al. [2008] reduced the originally reported stratospheric trends to ppmv yr 1 ( % yr 1 ). These authors applied two frost point temperature calibration corrections to the same data set analyzed by Oltmans et al. [2000] that slightly increased stratospheric water vapor mixing ratios prior to 1987 and reduced them after The magnitudes of the corrected trends diminished with altitude, similar to the findings of Oltmans et al. [2000]. Scherer et al. [2008] noted a sudden drop in stratospheric water vapor beginning in 2001 that Randel et al. [2006] attributed to anomalously cold tropical tropopause temperatures and an increase in tropical upwelling. [7] Here we present new trend analyses of the modern 30 year record ( ) of stratospheric water vapor measurements over Boulder by the balloon borne NOAA FPH. The record of 336 balloon flights is divided into four distinct time periods of multiple year water vapor trends by a two phase regression model [Lund and Reeves, 2002], then is analyzed using measurement uncertainties as statistical weights in piecewise continuous fitting procedures. 2. FPH Measurements and Their Uncertainties [8] The original design of the NOAA FPH has been improved upon throughout the years [Mastenbrook and Oltmans, 1983; Oltmans et al., 2000; Vömel et al., 1995; Oltmans and Hofmann, 1995], yet the fundamental measurement principle and calibration procedure remain the same. The measurement principle is to maintain a thin, stable layer of frost (ice) on a temperature controlled mirror as air flows through the hygrometer at L s 1 (4 6 ms 1 ). Changes in frost coverage, detected by a photodiode that measures the frost surface reflectivity of a small infrared beam, are countered by rapidly adjusting the amount of mirror heating against persistent cryogenic cooling. A stable frost layer implies equilibrium between the ice and overlying water vapor. Under these equilibrium conditions the ice surface temperature (frost point temperature) and partial pressure of water vapor in the air stream are directly related. The water vapor partial pressure is calculated using the Goff Gratch formulation of the Clausius Clapeyron equation [Goff, 1957], then divided by dry atmospheric pressure to determine the volume mixing ratio. The only calibration required is that of a small thermistor embedded in the mirror, and this is done using NIST traceable standards. No water vapor calibration scale or water vapor calibration standards are required, as these are difficult to create and maintain. Together, the well established measurement principle and calibration procedure of the NOAA FPH contribute to the sustainable accuracy of these water vapor measurements over long periods of time. [9] The NOAA FPH was launched from Boulder approximately 500 times between April 1980 and February 2010 to produce 314 water vapor vertical profiles with reliable stratospheric data. From November 2004 through February 2009, 22 of the 85 water vapor profiles in the Boulder record were obtained using the cryogenic frost point hygrometer (CFH). This instrument spawned from the NOAA FPH, using the same fundamental measurement principle, with substantial efforts to reduce instrument size and weight, improve frost layer stability and eliminate the need for a sun shield [Vömel et al., 2007]. Many similar improvements were also made to the NOAA FPH, but since FPH modifications were not exactly the same as for the CFH, both the design and operation of the two instruments diverged in subtle ways. [10] Over this 52 month period the 22 CFH flights were intermingled with 63 FPH flights. The number of CFH profiles in a given month exceeded the number of FPH profiles only twice. During June 2008 to February 2009 there were five flights with both instruments on the same balloon. For these flights the mean difference between stratospheric water vapor measurements by the two instruments was 0.1 ± 0.3 ppmv, about 2% of the mean stratospheric water vapor mixing ratio and well within the measurement uncertainties. [11] Given the tremendous dynamic range and variability of water vapor number density in the atmosphere from the surface to 28 km, the FPH measurement quality can vary significantly during each flight. There were three primary difficulties that caused some flight data to be of lower quality: poor instrument performance, often resulting from substandard frost layer control, contamination of high altitude ascent measurements, and a low vertical resolution of measurements during descent if the balloon burst. Variable flight conditions, including the presence of low altitude and highaltitude clouds, can also affect the quality of instrument performance from one flight to the next. The stratospheric water vapor measurements for each flight were carefully evaluated and only flights with reliable data were retained for this analysis. [12] Overall, an average measurement precision in the stratosphere of better than 4% is calculated from the data analyzed in this work (see section 3). Measurement precision has improved in recent years through better control of the frost layer. Modifications to the frost control electronics during , necessitated by component obsolescence, compromised the measurement precision during a number of flights and significantly lowered the availability of reliable stratospheric profile data. [13] During ascent, the FPH trails in the wake of the balloon, where any outgassing of water vapor from the balloon can contaminate the measurements. There may be intermit- 2of12

3 tent contamination if the instrument payload, hanging 30 m below the balloon, swings in and out of the balloon wake. This is typically observable above 24 km as the balloon warms up after passing through the cold tropopause. Until 2009, similar contamination from a sun shield, mounted just above instrument to minimize sunlight reaching the frost measuring photodiode, was also a potential source of contamination during ascent. The high altitude ascent measurements used in this work have been carefully scrutinized for the effects of contamination and excluded if suspect. Exclusion of the contaminated ascent data is very straightforward if there are high quality measurements available for descent, when the instrument payload leads the balloon and contamination is not a concern. [14] A burst balloon causes the payload to fall uncontrolled through the stratosphere at rates >20 m s 1, slowing only after the parachute provides sufficient drag. Both the vertical resolution and quality of FPH data during uncontrolled descent are degraded because of the excessive velocity of the instrument. The desire for uncontaminated high altitude water vapor data prompted Mastenbrook [1966] to design and deploy a valve system that slowly releases helium from the balloon at a preset pressure and efficiently turns balloon ascent into a controlled descent at rates of 4 6ms 1. Several versions of this valve system have been used throughout the years with varying degrees of success, but overall 75% of balloons fitted with valves and launched from Boulder did not burst (85% for the flights studied here). Thanks to this valve system the 30 year stratospheric data record is comprised of near equal numbers of reliable ascent and descent water vapor measurements, otherwise ascent measurements would predominate. [15] It is difficult to quantify the total uncertainties of FPH measurements during each flight given the multitude of errors that can potentially influence measurements. Vömel et al. [1995] reported that the FPH performance is slightly variable from instrument to instrument, depending on the quality of several factors, and estimated that measurement accuracy in the stratosphere is better than 10%. Over the years, component modifications, changes to the frost layer control logic, and analog to digital upgrades of data telemetry and frost control circuitry may have affected the accuracy of measurements. The significant transformation from analog to digital telemetry in 1991 resulted in no evidence of shifts in the measurement data [Oltmans et al., 2000]. Unfortunately we cannot make similar claims for all instrument changes over the years, nor can we objectively quantify any such biases from the information we have available for each flight. Our use of the same measurement principle and calibration procedure over the entire 30 year period has undoubtedly reduced the potential for measurement biases to drift over the long term. [16] In this paper we desire to objectively weight the data obtained during each flight, using a measure of actual instrument performance, before performing trend analyses. Simple temporal or spatial averaging of lower precision measurements can yield a very good average result, but the large uncertainty of that average should reduce its significance in quantitative analyses like trend determinations. Weighting places greater quantitative emphasis on higher quality data and reduces the potentially anomalous influences of poorer quality data. Scherer et al. [2008] assessed data quality in the Boulder record and analyzed trends after excluding lowerquality profile data. Here, we employ quantitative weights based on in flight measurement variability as a proxy of measurement uncertainties during a flight. Other sources of measurement error undoubtedly contribute to total measurement uncertainties, but these cannot be objectively quantified on a flight to flight basis for the past 30 years. In section 5 we will discuss the quantitative effects of significantly larger FPH measurement uncertainties on the water vapor trends determined here. 3. Comparison With Previous Trend Analyses, [17] Previous reports of the stratospheric water vapor increase over Boulder during [Oltmans et al., 2000; Kley et al., 2000; Scherer et al., 2008] were based on simple linear regression fits of the average mixing ratios measured in 2 km altitude layers during each flight, from 16 to 28 km (Q = 380 to 750 K), hereinafter referred to as 2 km averages. For direct comparison to these previous studies we use the same corrected data set analyzed by Scherer et al. [2008] of 190 flights from April 1980 through April 2000, specifically for the km and km altitude layers. [18] We apply a statistical weight to each 2 km average based on an estimate of the in flight measurement variability that is typically caused by unstable control of the frost layer. Weights are calculated independently for each flight and 2 km layer as the reciprocal of twice the standard deviation (s) of the water vapor mixing ratios measured within the layer. This calculation assumes negligible natural water vapor variability within each layer that is arguably small except for the km layer where there can be significant vertical gradients. The 2 s values averaged 0.34 ppmv (8%) for the km layer, 0.18 ppmv (4%) for both the km and km layers, and 0.16 ppmv (3.5%) each for km, km, and km. Over the full data record, the 2 s values ranged from <0.1 ppmv to >1 ppmv, reflecting substantial variations in measurement uncertainties during flights and from one flight to the next. [19] Several 2 km averages were identified as statistical outliers and excluded prior to the fitting of the water vapor time series (Figure 1). Outlier testing was performed by creating a smoothed representation of the data in each 2 km layer (Figure 2) and computing the absolute values of residuals. Gaps in the smoothed representations were filled by interpolation and extrapolation (not shown). Data points with residuals >2.5 times the mean of absolute residuals for the entire time series were excluded from further analysis (Figure 1). [20] Trend analyses were performed using weighted linear least squares regression of the individual flight 2 km averages. Trends of ± ppmv yr 1 for the km layer and ± ppmv yr 1 for the km layer were determined for the period (orange lines in Figures 1b and 1e). These values are 20 50% lower than the ± and ± ppmv yr 1 trends presented by Scherer et al. [2008] (green lines in Figures 1b and 1e). Our trend slopes and uncertainties are reduced by the exclusion of statistical outliers and the use of weighted fits. These results are presented here for comparison purposes 3of12

4 Figure 1. (a f) Water vapor mixing ratio averages (filled black circles) for six 2 km altitude layers over Boulder, Colorado. Open gray circles are statistical outliers that were excluded from the analyses. Vertical error bars (black and gray) span the 95% confidence intervals of the 2 km averages. Piecewise continuous trend lines (red) and quadratic curves (blue) are shown for the four periods defined by the three changepoint dates (magenta vertical lines). Linear trends for reported by Scherer et al. [2008] (green lines) and determined here using weighted least squares regressions (orange lines) for km (Figure 1b) and km (Figure 1e) are included for comparison purposes. only, as we now take a new approach to trend analyses of the Boulder water vapor record. 4. New Trend Analyses [21] The historical definition of as a single period of net water vapor increase was retained in the previous section for the purpose of comparison. Fitting a straight line to these 20 years of data is probably not the best analytical approach to trend determination since some residuals exceed 1 ppmv (e.g., Figure 1e). Instead, we prefer to break the modern 30 year record into distinct trend periods, each spanning at least several years of a general trend in water vapor. This was accomplished using the revised two phase regression model of Lund and Reeves [2002] to identify undocumented changepoint dates in the time series for each 2 km layer. [22] The analysis was initiated by identifying, for each altitude layer, the changepoint date just prior to the drop in stratospheric water vapor (Figure 2) that Randel et al. [2006] attribute to an anomalous decline in tropical tropopause temperatures. Residuals from linear regression fits to the data before and after the identified changepoints were then analyzed for additional changepoints, and in a recursive manner six statistically significant changepoint dates were identified for each altitude layer. The results, though not entirely consistent between all six layers, and at times finding changepoint dates too close in time for this analysis of multiple year water vapor trends, identified two additional changepoint dates near the years 1990 and 2006 that were common to at least four of the six 2 km layers. These changepoint dates were confirmed by examining smoothed derivatives of the moving averages (including interpolated values) for long term transitions from negative to positive values (and vice versa). Altitude layers with undetected changepoint dates near 1990 and 2006 were assigned changepoint dates based on their smoothed derivatives. [23] Both the changepoint and derivative analyses suggest that the entire data set could be divided into many trend periods, some as short as 2 years. The benefits of doing this, namely decreased residuals from fits, are outweighed by fit uncertainties that generally inflate with shrinking amounts of data. It is also advantageous to this study that trends for each of the altitude layers are determined over similar 4of12

5 Figure 2. Moving averages of the 2 km water vapor mixing ratio averages in each of the six altitude layers. The averaging window had a width of ±1 year and a threshold of 12 data points to compute an average. Colored vertical bars define the four trend periods for each altitude layer. Moving averages were not calculated for the first and last years of the record. No interpolated or extrapolated values are shown. periods of time so the results can be compared and interpreted. Since these analyses produce 3 consistent changepoints for the majority of the 6 altitude layers, we break the data record into four trend periods. The changepoint dates for each altitude layer as depicted in Figures 1 and 2 average ± 0.8, ± 0.7, and ± 0.4, where would represent 1 July We thus designate the approximate date intervals , , , and as periods 1 4, respectively. [24] Trends within each of the four periods were determined using weighted, piecewise continuous regression fits of the 2 km averages after outliers were removed. Again, the statistical weights were based on estimates of in flight measurement variability as described above. Linear and secondorder polynomial (quadratic) functions were independently employed in the fitting algorithms. Quadratic fits generated c 2 values 10% lower and residuals 3% smaller than those for linear fits, indicating a greater goodness of fit for the polynomials. For these reasons we choose the quadratic fits over linear fits for trend determinations. [25] Water vapor trends at altitudes km for periods 1 3 do not conform to trends at the lower altitudes for either the linear or quadratic fits. This is likely the result of several lengthy intervals of sparse or missing data during periods 1 and 2, and a paucity of reliable data between January 2000 and September 2001 (Figure 1f). The km trend values are highly suspect and are excluded from further analysis. We are able to analyze with greater confidence the km data after 2001 when there are fewer gaps, but unfortunately these years span only period 4 in its entirety. [26] Though water vapor mixing ratios in the midlatitude lower stratosphere vary seasonally [e.g., Mote et al., 1996], previous analyses of the Boulder water vapor data ignored this attribute because many seasonal cycles were included in the 20 year record. In this current work we have fit data periods as short as 4.1 years and therefore consider it prudent to investigate the potential impacts of seasonal cycles on the derived trends. [27] Monthly averaged water vapor vertical profiles over the entire 30 year record (Figure 3) reveal that most of the water vapor seasonality is mixed out above 19 km (Q > 450K). Monthly averages for km ( K) vary negligibly with season, but at km ( K) there are significant seasonal changes from a February minimum of 3.3 ppmv to an August maximum of 6.2 ppmv (Figure 3). Both the magnitude and phase of this cycle suggest it derives from rapid transport of the seasonal cycle in the tropical lower stratosphere to the midlatitudes. Below 16 km (380 K) the midlatitude seasonal cycle amplitude increases dramatically (Figure 3), with high northern summer mixing ratios (>8 ppmv) signifying that air masses could have entered the stratosphere without passing through the tropical tropopause, either isentropically through the tropopause break with subsequent poleward advection [e.g., Ray et al., 1999] or vertically through the warmer extratropical tropopause. For these reasons we exclude data below 16 km from this analysis of water vapor in the stratospheric overworld. [28] The effects of water vapor seasonality on the derived trends for km were investigated by removing the 30 year average seasonal cycle from the data (not shown). Specifically, we constructed a representative average seasonal cycle around the mean 30 year mixing ratio at km, then subtracted the appropriate seasonal cycle amplitude (based on the day of year the data were obtained) from each km average. The deseasonalized data were then fit in the same manner as the original data to produce trends that do not statistically differ from the original trends. 5. Water Vapor Growth Rates and Growth [29] We define growth as the net change in water vapor abundance over a given period of time, based on the quadratic 5of12

6 Figure 3. Monthly averaged vertical profiles of stratospheric water vapor over Boulder, Colorado. Each average profile is based on individual soundings in the specified month during The seasonal cycle is evident for altitudes <19 km. trend curves (Figure 1). Growth uncertainties are calculated from the 95% confidence limits of trend curve endpoint mixing ratios, as these bound the minimum and maximum growth values possible for the trend period. Growth rates (Table 1), computed as growth divided by the length of trend period, have the same relative uncertainties as growth values (Table 2). Since the best piecewise continuous regression fits to the 2 km averages were quadratic (i.e., nonlinear), the growth rates presented here represent average rates of water vapor change over trend periods rather than singular trend slopes. All growth rates and growth values derived here, for every period and 2 km altitude layer from 16 to 26 km, are statistically different from zero at the 95% level of confidence. Growth rates and growth values for periods 1 3 are not available for the km layer because the trend analysis produces dubious results for the reasons mentioned above. [30] Before discussing the water vapor trends we examine the sensitivity of growth rates to the choice of changepoint dates. Quadratic piecewise regression fits were performed while each set of changepoint dates was sequentially shifted, both backward and forward in time, by 1 5 flight dates. This procedure generated 11 different trend curves for each altitude layer during each trend period, from which we calculated a mean trend for each layer and period. Without exception these mean trends were not statistically different (95% confidence) from the growth rates determined using the original changepoint dates. [31] Water vapor growth rates for periods 1, 2, and 4 are positive for all five 2 km stratospheric altitude layers (16 26 km), but for period 3 are uniformly negative (Table 1). Growth rates for period 1 ( ) average ± ppmv yr 1 and diminish with altitude, from distinctly positive (0.057 ± ppmv yr 1 )at18 20 km to weakly positive (0.007 ± ppmv yr 1 )at24 26 km. Period 2 ( ) growth rates average ± ppmv yr 1 and increase with altitude from ± ppmv yr 1 at km to ± ppmv yr 1 at km. Growth rates for period 3 ( ) are negative and more consistent across the 5 altitude layers, averaging ± ppmv yr 1. For period 4 (mid 2005 through February 2010) growth rates increase with altitude from ± ppmv yr 1 at km to ± ppmv yr 1 at km and average ± ppmv yr 1. On average, the magnitudes of growth rates for periods 3 and 4 are about twice those for periods 1 and 2, respectively. [32] The same observations presented above for water vapor growth rates are generally applicable to water vapor growth values. Growth values for periods 1, 2, and 4 are all positive and exhibit statistically significant altitude gradients of 0.06 ± 0.02 ppmv km 1, 0.08 ± 0.02 ppmv km 1, and 0.03 ± 0.02 ppmv km 1, respectively. Growth values for period 3 are all negative and much more uniform with altitude than those for the other three periods (Table 2). [33] One deviation from the growth rate observations above is that the magnitudes of average growth values for periods 3 and 4 are not twice those for periods 1 and 2 (Table 2), as was the case for growth rates (Table 1), simply because each of periods 1 and 2 was twice as long as either period 3 or 4. However, even through period 3 was only 5 years in duration, large negative growth rates reversed 40 50% of the previous 20 years positive growth. Figure 2 reveals very abrupt water vapor decreases for the lower Table 1. Growth Rates of Stratospheric Water Vapor Over Boulder, Colorado a Period Dates b km km km km km km Altitude Interval (0.016) c (0.011) (0.010) (0.007) (0.007) (0.011) (0.011) (0.012) (0.010) (0.007) (0.024) (0.024) (0.026) (0.029) (0.024) (0.028) (0.026) (0.028) (0.029) (0.029) (0.027) (0.009) (0.008) (0.008) (0.006) (0.005) (0.009) (0.008) (0.008) (0.008) (0.007) a Growth rates and their uncertainties were calculated by dividing water vapor growth and growth uncertainties, determined from quadratic piecewise continuous regression fits, by the length of the appropriate trend period. Growth rates for the km layer during periods 1 3 are not presented because they do not conform to growth rates at the lower altitudes due to sparse data and lengthy gaps in the record. b Dates are the average changepoints that define water vapor trend periods for the six different altitude intervals. A date of 1 January 1999 would be represented by c Water vapor growth rates and their 95% confidence limits (in parentheses) are given in ppmv yr 1. 6of12

7 Table 2. Growth in Stratospheric Water Vapor Over Boulder, Colorado a Period Dates b km km km km km km Altitude Interval (0.13) c 0.51 (0.10) 0.38 (0.10) 0.22 (0.07) 0.07 (0.07) (0.12) 0.42 (0.12) 0.48 (0.12) 0.77 (0.11) 0.89 (0.08) (0.13) 0.35 (0.13) 0.29 (0.14) 0.39 (0.13) 0.37 (0.11) (0.14) 0.21 (0.13) 0.54 (0.13) 0.64 (0.13) 0.62 (0.12) 0.59 (0.11) (0.17) 0.93 (0.16) 0.87 (0.15) 0.99 (0.13) 0.96 (0.10) (0.26) 0.80 (0.24) 1.12 (0.25) 1.25 (0.23) 1.20 (0.19) a Growth was determined from quadratic piecewise continuous regression fits. Growth values for combined periods 1 2 and 1 4 are sums of growth during the individual periods, with uncertainties calculated by combining in quadrature the 95% confidence limits of individual period growth values. Growth values for the km layer during periods 1 3 are not presented because they do not conform to growth at the lower altitudes due to sparse data and lengthy gaps in the record. b Dates are the average changepoints that define water vapor trend periods for the six different altitude intervals. c Water vapor growth values and their 95% confidence limits (in parentheses) are given in ppmv. altitudes (16 22 km) at the start of period 3; about 90% of the overall drop occurred within the first 2 years. At higher altitudes the period 3 decreases were much more gradual, yet the overall water vapor declines during this period were similar at all altitudes. Positive growth rates during the even shorter ( 4.5 years) period 4 were on average 50% stronger than the negative period 3 growth rates, restoring mixing ratios at all altitudes to near or above their values at the start of period 3. There are indications period 4 growth at the lowest altitudes slowed or leveled off after 2006 (Figures 1a 1b), but strong growth above 20 km did continue into early 2010 (Figures 1c 1f). [34] Growth values for , calculated as sums of growth during periods 1 and 2, range from 0.73 ± 0.17 ppmv to 0.99 ± 0.13 ppmv (Table 2) and have a positive altitude gradient that is barely statistically significant (0.022 ± ppmv km 1 ). Fractions of growth that occurred during period 2 ( ) increase considerably with altitude from 40% to 90%, yet the five 20 year growth values are fairly consistent, averaging 0.89 ± 0.10 ppmv (Table 2). The linear trends of Scherer et al. [2008] translate to growth values of 0.65 ± 0.10 ppmv at km and 0.57 ± 0.13 ppmv at km (Figures 1b and 1e) which are 60 70% of growth values based on these new quadratic piecewise fits. Visually, the 20 year linear trends (green and orange lines in Figures 1b and 1e) at these altitudes are constrained low by weak increases during the first years of the record and fail to capture the subsequent stronger upturns in water vapor like the piecewise quadratic fits do. [35] Growth values for the entire 30 year period ( ) range from 0.71 ± 0.26 to 1.25 ± 0.23 ppmv and average 1.02 ± 0.24 ppmv (Table 2). Relative to mean water vapor mixing ratios these values represent increases of 20 ± 7% to 31 ± 6%, and an average increase of 27 ± 6% for km. The 30 year growth values also exhibit a statistically significant altitude gradient of 0.07 ± 0.04 ppmv km 1. Intuitively, the positive altitude gradients of water vapor increases during periods 2 and 4, and over the 30 year record all require significant contributions from a source whose strength increases with altitude, such as the oxidation of stratospheric methane. [36] As mentioned earlier, the statistical weights used in piecewise fitting procedures are based on measurement variability and do not include estimates of systematic errors. Here we increase measurement uncertainties by including a potential 5% systematic error in each 2 km average to illustrate how the introduction of substantial measurement biases would affect water vapor growth values and their altitudinal behavior. The value of 5% is deemed to be a realistic estimate of potential systematic measurement errors in the stratosphere based on observed flight to flight mixing ratio differences and the estimate of Vömel et al. [1995] that FPH accuracy in the stratosphere is better than 10%. This illustration should not be misinterpreted to imply that systematic errors of 5% are embedded in all stratospheric FPH measurements. Rather, systematic errors of varying values, some possibly near 5%, may be present in the Boulder record. A uniform bias error of 5% is employed here because there is no objective way to quantify, in retrospect, the systematic errors for each and every FPH flight. [37] The illustrative 5% systematic error estimates were combined in quadrature with the measurement variability values to yield total measurement uncertainties 25% to 150% greater than the measurement variability values themselves. Note that in many cases the added systematic errors greatly overshadow the measurement variability, making fit weights much more uniform and less effective in reducing the potentially anomalous impacts of flights with high measurement variability. [38] The introduction of 5% systematic measurement errors changes growth values because the relative fit weights within each 2 km interval are altered. Changes to the 31 growth values in Table 2 average 0.02 ± 0.14 ppmv, with an average absolute change of 0.11 ± 0.09 ppmv. 30 year growth values are changed by 0.11 to 0.18 ppmv ( 16 to 20%), with an average change of 0.05 ± 0.12 ppmv ( 5 ± 11%). Four of the illustrative growth values, three for period 1 and one for period 4 are not statistically different from zero because the additional 5% errors greatly expand the uncertainties of growth values. All other growth values remain statistically significant, including those for periods 1 2 combined ( ) and periods 1 4 combined ( ). Without exception the original growth values (Table 2) lie within the uncertainty ranges of the illustrative growth values and 26 of the 31 illustrative growth values lie within the uncertainty ranges of the original (Table 2) values. [39] With the larger uncertainties of the illustrative growth values come increased uncertainties in the altitude gradients of water vapor growth. For periods 1, 4, and 1 4 combined, the altitudinal gradients of growth lose their statistical significance because of larger error bars while gradients for periods 2 and 1 2 combined remain significantly positive at 7of12

8 0.04 ± 0.03 and 0.06 ± 0.05 ppmv km 1, respectively. The lack of altitude dependence for period 3 illustrative growth values is consistent with the original growth values. [40] Inclusion of a uniform 5% systematic error in the uncertainty values for all 2 km averages greatly impacts the uncertainties of growth values and their altitudinal gradients. However, nearly all the water vapor growth values computed with these large systematic errors remain statistically significant. Altitude gradients in growth for periods 2 and 1 2 combined also remain statistically significant, but the altitude gradients for periods 1, 4 and 1 4 combined lose their statistical significance due to the much larger uncertainties. 6. Water Vapor Growth from Methane Oxidation [41] The only important in situ sources of stratospheric water vapor are oxidation of stratospheric methane (CH 4 ) and hydrogen (H 2 ). The observed relationship between CH 4 and H 2 O mixing ratios below 30 km in the northern midlatitude stratosphere implies that CH 4 and H 2 oxidation reactions produce approximately 2 molecules of H 2 O for each CH 4 molecule destroyed [le Texier et al., 1988; Dessler et al., 1994; Hurst et al., 1999; Zöger et al., 1999]. [42] Stratospheric water vapor production may be enhanced following growth in tropospheric CH 4 or H 2 if some of the additional tropospheric methane or hydrogen is oxidized in the stratosphere. We calculate these enhancements following the approach of Rohs et al. [2006] who presented global tropospheric and stratospheric CH 4 growth values for and determined the growth in stratospheric water vapor (at several altitudes) attributable to tropospheric methane growth. Forexample,duringthis17yearperiod,CH 4 increased in the global troposphere and at 26 km in the northern midlatitudes by and ppmv, respectively, implying that 55% of the additional tropospheric CH 4 had been oxidized during transit from the tropical tropopause to 26 km in the northern midlatitudes. This produced 0.55 * * 2 = ppmv of additional water vapor. Growth in stratospheric water vapor from the oxidation of additional stratospheric H 2 was insignificant during the 1980s and 1990s because trends of tropospheric and stratospheric H 2 were both very small ( ppmv yr 1 )[Rohs et al., 2006]. [43] We interpolated the results of Rohs et al. [2006] at the centers of our 2 km altitude layers to calculate the fractions of tropospheric methane increases that produce additional stratospheric water vapor. These range from 3.5 ± 0.5% at km to 59 ± 12% at km, increasing nearexponentially with altitude. Tropospheric methane increases during periods 2, 3 and 4 are computed from a smoothedcurve representation of global tropospheric CH 4 mixing ratios from mid 1983 to the end of 2009 [Dlugokencky et al., 2009]. Calculation of the tropospheric methane increase for period 1 requires back extrapolation of this curve using the mean ± ppmv yr 1 tropospheric CH 4 trend for [Blake and Rowland, 1986]. [44] Following Rohs et al. [2006], tropospheric methane growth is calculated over time periods 4 years earlier than the trend periods for stratospheric water vapor to account for the 3 5 year mean ages of midlatitude air masses at km. Tropospheric CH 4 increased by ± ppmv from 1976 through 2005, with 59% and 32% of the growth occurring during periods 1 and 2, respectively (Table 3). Methane grew by ppmv during the first 2 years of period 3 ( ), stalled for the remainder of period 3 ( ), then weakly increased by ppmv during period 4 ( ) [Dlugokencky et al., 2009]. [45] As above, twice the tropospheric CH 4 increase during each trend period is multiplied by the altitude dependent fraction of additional CH 4 oxidized to determine the methaneattributable water vapor growth within each 2 km altitude layer (Table 3). These values vary considerably with altitude and from one period to another, with the largest values at the highest altitudes during periods 1 and 2 when tropospheric CH 4 growth was strong. Methane increases during periods 3 and 4 were quite small and contributed very little to stratospheric water vapor growth. Below 18 km CH 4 oxidation was always a weak contributor (<3%) to water vapor increases, even when tropospheric CH 4 growth was strong. [46] Methane growth provided enough additional stratospheric water vapor during period 1 to account for all water vapor growth above 22 km (Figure 4a). However, this was evidently not the only mechanism adding stratospheric water vapor below 22 km where methane attributable growth was only 2 46% of the observed water vapor growth if both sets of error bars are considered (Figure 4a). Coupling these altitude dependent methane growth contributions with an increase in tropical lower stratospheric water vapor would yield positive midlatitude stratospheric water vapor growth with an altitude dependence similar to that observed for period 1. [47] For period 2 there is good positive correlation between water vapor growth and methane attributable water vapor growth, but the latter can account for at most 14 ± 4% of the former (Figure 4b). The strong positive altitude gradient of period 2 water vapor growth (0.08 ± 0.02 ppmv km 1 ) indicates that the full remainder of growth could not emanate from a water vapor increase in the tropical lower stratosphere. Combining period 2 with period 1, when methane growth was also strong but water vapor growth diminished with altitude (Figure 4a), the maximum contribution of methane attributable water vapor growth during jumps to 28 ± 4% (Figure 4c). A similar maximum fraction of 30 year water vapor growth (25 ± 5%) is attributable to methane growth (Figure 4f) because CH 4 increased only weakly during periods 3 4 and the average net growth of water vapor during periods 3 4 was near zero. As for periods 2 and 4, the positive altitude gradient of 30 year water vapor growth (0.07 ± 0.04 ppmv km 1 ) cannot be explained by a simple combination of tropical water vapor changes and increases in tropospheric methane. It should be noted that these maximum relative contributions of methane growth to water vapor increases do not change significantly with the introduction of 5% systematic errors in the Boulder record, though in some cases the uncertainties of the relative contributions more than double. 7. Discussion [48] The water vapor time series and trends presented here for the northern midlatitude stratosphere are in good qualitative agreement with those recently presented by Fujiwara et al. [2010, Figure 11] for the tropical stratosphere based on data from balloon borne frost point hygrometers and satellite based sensors. These authors describe decadal 8of12

9 Table 3. Contributions of Methane Growth to Stratospheric Water Vapor Growth a Altitude Interval Period Dates b DCH 4 Troposphere km km km km km km (0.004) c (0.002) d (0.005) (0.009) (0.016) (0.026) (0.040) (0.002) (0.001) (0.002) (0.005) (0.008) (0.014) (0.022) (0.001) (0.000) (0.001) (0.001) (0.002) (0.003) (0.005) (0.001) (0.000) (0.000) (0.001) (0.001) (0.001) (0.002) (0.004) (0.002) (0.005) (0.010) (0.018) (0.029) (0.046) (0.005) (0.002) (0.005) (0.010) (0.018) (0.029) (0.046) a Stratospheric water vapor increases due to methane growth are determined using the approach and stratospheric CH 4 trends of Rohs et al. [2006]. An example calculation is given in section 6. Uncertainties are calculated from small uncertainties in tropospheric CH 4 growth and larger uncertainties in the calculated fractions of tropospheric CH 4 growth that contributed to stratospheric water vapor growth at the given altitudes. Dates defining the time periods of tropospheric CH 4 growth are 4 years earlier than those defining stratospheric water vapor trend periods to account for the 4 year mean age of midlatitude stratospheric air masses between 20 and 28 km. b Dates define the approximate time periods of tropospheric CH 4 growth that correspond to the water vapor trend periods for the six altitude intervals. c Tropospheric CH 4 growth values and their 95% confidence limits (in parentheses) are given in ppmv. d Values of stratospheric water vapor growth due to tropospheric CH 4 growth, and their 95% confidence limits (in parentheses), are given in ppmv. variations in tropical water vapor mixing ratios as higher and increasing in the 1990s, lower in the early 2000s, and probably slightly higher again or recovering after 2004, but did not perform a quantitative trend analysis. Interestingly, their time series of saturation mixing ratios at 100 hpa in the tropical western Pacific, based on temperature data from the ERA and JRA/JCDAS reanalyses [Fujiwara et al., 2010, Figure 13], exhibit many of the same ups and downs as the time series of water vapor moving averages over Boulder (Figure 2). [49] The magnitudes and altitude gradients of some of the water vapor trends presented here indicate that methaneattributable growth and tropical water vapor changes may not be the only mechanisms driving the observed midlatitude trends. Positive water vapor growth that strengthens with altitude during period 2, when methane contributions were 14 ± 4% or less (Figure 4b), cannot result exclusively from a water vapor increase in the tropical lower stratosphere. Water vapor growth during period 4 also strengthens with altitude in the absence of significant contributions from methane growth (Figure 4e). The period 3 water vapor decreases, which Randel et al. [2006] link to a drop in tropical tropopause temperatures, do not show the expected altitude dependency though this may be obscured by the large uncertainties in growth (Figure 4d). [50] Substantial efforts to attribute midlatitude stratospheric water vapor changes to water vapor perturbations in the tropical lower stratosphere have been made. Attempts to connect the stratospheric water vapor increase over Boulder to changes in tropical tropopause temperatures have been largely unsuccessful [e.g., Simmons et al., 1999; Randel et al., 2000; Zhou et al., 2001; Seidel et al., 2001; Randel et al., 2004; Fueglistaler and Haynes, 2005], promoting the idea that alternative mechanisms may be needed. Other perturbations that would moisten the lower tropical stratosphere have been proposed. These include a widening of the tropical upwelling region [Zhou et al., 2001; Rosenlof, 2002] and an increase in the relative fraction of air crossing the tropical tropopause during the northern summer when cold point temperatures are warmest [Rosenlof, 2002]. These tropical influences would undoubtedly increase midlatitude stratospheric water vapor, but they lack the altitude dependence needed to explain why period 2 water vapor growth over Boulder strengthens sharply with altitude. [51] One plausible mechanism for positive water vapor growth that strengthens with altitude is a slowing of the mass flux into the tropical upper stratosphere, as this would increase the residence times of air masses in this critical methane sink region [Rosenlof, 2002]. Engeletal.[2009] determined that the mean ages of northern midlatitude air masses at km altitude increased slightly over the last 3 decades, but the trend lacked statistical significance at the 90% confidence level. In apparent disparity, model studies [e.g., Butchart et al., 2010] suggest the stratospheric mean meridional circulation has strengthened in recent decades, especially at lower stratospheric altitudes. Ray et al. [2010] are able to reproduce the Engeletal.[2009] mean age trend as well as ozone trends over the past 3 decades by combining increased horizontal mixing into the tropics with a small strengthening of lower stratospheric circulation and a moderate weakening of the mass flux in the middle and upper stratosphere. [52] A weakening of the mean circulation in the middle and upper stratosphere and a strengthening of the mean circulation in the lower stratosphere would serve to reduce midlatitude stratospheric CH 4 abundance at higher altitudes by a larger relative fraction than at lower altitudes, expanding CH 4 mixing ratios differences between low and high altitudes over time. The Rohs et al. [2006, Figure 2] CH 4 mixing ratio trend curves for 23 and 30 km show a small ( 0.05 ppmv) divergence from 1978 to 2003, but this value is close to what is expected from the 0.3 ppmv growth in tropospheric CH 4 during Shorter term variations in the stratospheric CH 4 data may indicate transient changes in stratospheric dynamics, but these cannot be distinguished from the results of short term variations in tropospheric methane abundance, cross tropopause transport, or stratospheric oxidation chemistry. Transient changes like these should have little effect on our calculations of methane attributable water vapor growth (section 6) because our computations used net CH 4 growth values over a 17 year period. [53] Water vapor growth values and their altitude gradients during several trend periods between 1980 and 2010 cannot be adequately explained by a combination of methane growth and tropical water vapor changes. Intuitively, there is a need for at least one other important mechanism that increases midlatitude stratospheric water vapor with stronger contributions at higher than lower altitudes. Attribution of the 9of12

10 Figure 4. Observed stratospheric water vapor growth (black bars) and stratospheric water vapor growth attributable to methane growth (gray bars) for (a) period 1 ( ), (b) period 2 ( ), (c) periods 1 2 combined ( ), (d) period 3 ( ), (e) period 4 ( ), and (f) periods 1 4 combined ( ). The x axis scales in Figures 4a, 4b, 4d, and 4e are the same, but are expanded in Figures 4c and 4f to accommodate the larger ranges of values. Error bars span 95% confidence intervals of the growth values. For periods 1 2 combined and periods 1 4 combined the contributions of methaneattributable water vapor growth to observed water vapor growth range from 2% at lowest altitudes to 28% at the highest altitudes. observed water vapor changes over Boulder may be confounded by a complex and dynamic mixture of mechanisms that perturb midlatitude stratospheric water vapor. Several large, multiple year deviations in the Boulder time series (Figure 2) are poorly represented by the fitted trend curves. These deviations may or may not be driven by the mechanism(s) responsible for the underlying trend during that period. In other words, these fluctuations may have been caused by shorter term changes in the mechanism(s) responsible for the net growth during a given period, or may have resulted from changes in other mechanisms that influence stratospheric water vapor. [54] At present, in situ measurements of stratospheric water vapor are limited to a few locations worldwide, and only the Boulder record spans more than 20 years. Our lack of objective quantification of total FPH measurement uncertainties throughout the entire observational record imposes some qualifications on the interpretation of the observed water vapor changes over Boulder. The illustrative introduction of 5% systematic errors into the FPH measurement uncertainties quantitatively demonstrates these qualifications. Though in situ stratospheric water vapor measurements continue to improve at existing monitoring sites, it is critical for the longterm monitoring of stratospheric water vapor that new, reliable measurement programs are initiated around the globe and particularly at tropical locations. [55] Rates of stratospheric water vapor change during the last decade are on average 100% stronger than during the previous 2 decades. The rates of water vapor decreases during period 3 ( ), especially the rapid declines at the lower altitudes, indicate that reductions in tropical coldpoint temperatures were sudden and substantial. Positive trends during period 4 ( ) are on average stronger than the negative trends during period 3. Has the changing climate 10 of 12