A Microphysical
Connection Among Biomass Burning,
Cumulus Clouds, and Stratospheric Moisture
Science v.295, 15feb02
Steven Sherwood
Department of Geology and Geophysics, Yale University, New Haven, CT 06520, USA.
E-mail: steven.sherwood@yale.edu
Data collected during the last half-century appear to indicate an approximate doubling of stratospheric water vapor during this period, of which approximately half can be accounted for by increases in stratospheric production of water vapor by methane oxidation (1). The other half presumably results from increases in the moisture content of air entering the stratosphere through the tropical tropopause. However, temperatures near this entry point, which were thought to regulate stratospheric moisture levels, have not increased during recent decades (2, 3). This implies that either relative humidity near the tropopause has substantially increased, or other pathways exist whereby moisture can enter the stratosphere. The importance of this problem is underscored by recent findings that stratospheric moisture increases may be a significant contributor to global temperature trends (4) and may also interfere with the recovery of polar ozone by exacerbating destruction mechanisms (5).
I show that fluctuations in stratospheric humidity can indeed be caused by fluctuations in relative humidity just below the tropical tropopause, which in turn are governed by the sizes of ice crystals lofted in deep convective updrafts. The moisture content of air entering the stratosphere is thought to be controlled by condensation of vapor to the ice phase in transient lifting events outside of convective cells and/or phase changes within intense convective cells themselves (6). The relative importance of these two controlling factors is unknown. Convective moistening or drying should depend not only on temperature but also on the propensity of lofted ice to evaporate at a level high enough (>14 to 15 km) so that the vapor will enter the stratosphere rather than subsiding back into the troposphere (7, 8). It has been widely assumed that condensation outside of convection resets the water vapor to a lower value independent of convective influence, but the evidence presented here argues against this assumption.
This study uses data acquired from 1992 to 1998. Ice crystal “effective
diameter” De was retrieved near the tops of deep cumulonimbus
clouds (Cb) using the ISCCP (International Satellite Cloud Climatology Project)
B3 archive of radiance observations by the AVHRR (Advanced Very High Resolution
Radiometer) on board the NOAA (National Oceanic and Atmospheric Administration)
polar orbiter series (9, 10). Cb were defined as clouds whose 11-µm brightness
temperatures were less than 210 K, indicating tops in the region above 14 km
where the air is slowly rising toward the stratosphere (6). Water vapor
observations came from HALOE (Halogen Occultation Experiment) (11); available
vertical profiles from each month were interpolated to surfaces of constant
potential temperature 
and averaged over a tropical belt. Temperature fields near the tropical
tropopause were obtained by applying a specially designed analysis method to
radiosonde temperature data at 100 hPa (12). On the time and space scales
considered here, temperature anomalies at 100 hPa are expected to be
representative of those at other levels near the tropopause (13, 14).
The large (~5 to 10 K peak-to-peak) annual cycles in tropopause temperature, T, and saturation water vapor mixing ratio with respect to ice, qs (a function of T), cause a corresponding annual cycle in water vapor mixing ratio, q, of similar size and phase to that of qs (15, 16). Because changes in q not caused by temperature are sought, however, the relative humidity q/qs is considered. Strict thermodynamic control of water vapor would imply constant relative humidity. Because vapor concentrations can relax toward thermodynamic equilibrium only where condensed phases are present, the tropical <qs(t )> was computed here using a weighted mean with horizontal weighting proportional to the observed monthly mean 100- hPa frequency of cloud occurrence (17). An unweighted computation was also performed for comparison.
The results of this analysis are summarized in Fig. 1 (18). The quantity RH*
(t
) used to show humidity is defined as 
where q is the mean mixing ratio observed by HALOE within a tropical belt
(here, 20°S to 20°N) at the target level ,
and 
is the time required for air to rise from just below the tropopause (370 K) to
(19). Thus, RH* gives the ratio of an air layer’s eventual humidity (after it
has slowly lofted to )
to its initial saturation humidity at time t when the layer is just below
the tropopause. The precise mean value of RH* is not very meaningful because it
depends on the vertical weighting of the HALOE retrieval and the exact vertical
matching of the temperature and moisture data. The important feature in Fig. 1
is RH fluctuations of ~30%, which occur on a variety of time scales, notably
semiannual.
It takes several months or longer for air to rise the ~4 km from the 370 K level to the 450 K level. During this time, q at a given height often changes substantially. Nonetheless, RH* fluctuations are remarkably similar at the two levels. This “tape recorder” effect (16) is well known for seasonal thermodynamic changes but reappears here more generally. The 450 K level is above all significant dehydration. Moisture variations reaching this level enter what is known as the “tropical pipe” region (20) where they must ultimately affect the entire stratosphere. Dehydration or mixing on the way from 370 K to 450 K removed about 30% of the moisture regardless of the initial value, preserving moisture variations below the tropopause rather than erasing or clipping them.
Fluctuations of the observed mean effective diameter De of Cb ice crystals are also tightly correlated with those of RH* at both levels (Fig. 1). The correlations are highly significant regardless of the range of latitudes used in computing q or <qs> ( Table 1), strongly supporting a physical connection between De and RH. This is also true if Cb weighting of the temperature field is not implemented, although the correlation drops significantly (21); this result supports the contention that only temperatures in the presence of clouds affect ambient vapor amounts.
Fig. 1.
Effective ice radius De in Cb clouds and observed RH*
( plotted versus 370 K crossing time) at
= 370 K and
= 450 K (see text for definition). Gap in observations is due to excessive orbit
decay of NOAA-11. Scale for De at right decreases toward the
top. Measures of statistical association are given in Table 1.
De: effective diameter (microns)
q370: HALOE mixing ratio, 20N-20S, at 370 K (ppmv)
qs: radiosonde-estimated saturation mixing ratio (ppmv)sig( ): Standard error based on intramonth variability. This does not include interpolation errors, which for qs are probably
Supplemental Table 1.
similar or larger than sigma. It also excludes biases, which may be substantially larger than sigma for all observables.Year De singed) q370 sig(q370) qs sig(qs) 1992.042 28.001 0.090 0.000 0.000 4.499 0.044 1992.125 28.282 0.096 0.000 0.000 4.993 0.054 1992.208 27.975 0.119 0.000 0.000 5.347 0.064 1992.292 27.857 0.133 0.000 0.000 4.907 0.027 1992.375 27.682 0.098 0.000 0.000 5.044 0.029 1992.458 27.784 0.098 0.000 0.000 6.166 0.050 1992.542 28.071 0.092 0.000 0.000 9.374 0.095 1992.625 28.110 0.092 0.000 0.000 9.124 0.118 1992.708 28.088 0.104 0.000 0.000 0.000 0.000 1992.792 27.915 0.079 0.000 0.000 6.043 0.074 1992.875 27.465 0.087 0.000 0.000 4.621 0.058 1992.958 27.505 0.087 0.000 0.000 4.058 0.066 1993.042 28.071 0.086 4.905 0.096 5.497 0.080 1993.125 27.920 0.106 3.654 0.265 4.771 0.055 1993.208 28.170 0.106 4.668 0.145 0.000 0.000 1993.292 27.538 0.126 4.179 0.186 4.921 0.055 1993.375 27.247 0.110 5.900 0.171 5.531 0.032 1993.458 27.487 0.118 6.021 0.259 0.000 0.000 1993.542 27.952 0.105 7.841 0.139 9.350 0.118 1993.625 27.904 0.131 5.781 0.138 9.581 0.101 1993.708 0.000 0.000 5.790 0.122 0.000 0.000 1993.792 0.000 0.000 5.035 0.131 0.000 0.000 1993.875 0.000 0.000 4.112 0.066 0.000 0.000 1993.958 0.000 0.000 4.030 0.170 0.000 0.000 1994.042 0.000 0.000 3.173 0.059 0.000 0.000 1994.125 0.000 0.000 3.530 0.127 0.000 0.000 1994.208 0.000 0.000 3.991 0.099 0.000 0.000 1994.292 0.000 0.000 3.812 0.106 0.000 0.000 1994.375 0.000 0.000 4.596 0.064 0.000 0.000 1994.458 0.000 0.000 5.123 0.170 0.000 0.000 1994.542 0.000 0.000 5.711 0.063 0.000 0.000 1994.625 0.000 0.000 5.772 0.101 0.000 0.000 1994.708 0.000 0.000 5.344 0.100 0.000 0.000 1994.792 0.000 0.000 4.984 0.100 0.000 0.000 1994.875 0.000 0.000 4.496 0.075 0.000 0.000 1994.958 0.000 0.000 0.000 0.000 0.000 0.000 1995.042 0.000 0.000 3.601 0.085 0.000 0.000 1995.125 0.000 0.000 3.330 0.156 0.000 0.000 1995.208 27.886 0.080 3.437 0.154 4.598 0.052 1995.292 27.960 0.118 4.118 0.135 4.245 0.031 1995.375 27.562 0.089 4.538 0.081 5.411 0.033 1995.458 27.465 0.104 4.206 0.171 6.415 0.044 1995.542 27.695 0.087 5.587 0.067 8.936 0.066 1995.625 28.078 0.083 6.063 0.121 8.961 0.087 1995.708 28.012 0.091 5.956 0.120 7.657 0.054 1995.792 27.796 0.076 5.981 0.091 6.737 0.048 1995.875 27.142 0.087 5.775 0.109 4.384 0.062 1995.958 27.310 0.078 4.426 0.709 4.134 0.064 1996.042 27.421 0.081 4.120 0.071 4.400 0.066 1996.125 28.021 0.071 4.472 0.129 4.499 0.062 1996.208 27.754 0.076 4.020 0.079 4.314 0.043 1996.292 27.550 0.088 4.529 0.102 4.734 0.033 1996.375 27.556 0.097 4.696 0.190 5.138 0.028 1996.458 27.522 0.081 0.000 0.000 5.725 0.054 1996.542 27.555 0.101 6.015 0.096 6.569 0.110 1996.625 27.856 0.091 6.249 0.354 7.700 0.094 1996.708 27.568 0.097 5.731 0.117 6.197 0.062 1996.792 27.547 0.084 5.543 0.092 5.560 0.063 1996.875 27.042 0.070 5.544 0.204 4.464 0.053 1996.958 26.895 0.082 5.044 0.358 3.847 0.063 1997.042 27.294 0.063 4.309 0.108 4.144 0.068 1997.125 27.457 0.080 3.124 0.283 3.992 0.077 1997.208 27.230 0.101 4.162 0.107 4.487 0.065 1997.292 27.233 0.092 4.505 0.081 4.056 0.037 1997.375 27.199 0.102 4.284 0.176 4.137 0.031 1997.458 27.369 0.108 5.397 0.198 5.843 0.046 1997.542 27.917 0.111 5.940 0.068 6.969 0.083 1997.625 27.956 0.111 6.134 0.144 8.278 0.094 1997.708 27.866 0.119 6.838 0.083 7.601 0.049 1997.792 27.610 0.117 6.232 0.073 6.453 0.046 1997.875 27.358 0.092 0.000 0.000 5.787 0.038 1997.958 27.694 0.094 4.491 0.108 4.762 0.038 1998.042 27.687 0.107 4.198 0.117 4.752 0.038 1998.125 27.819 0.093 4.408 0.700 4.859 0.031 1998.208 28.003 0.109 3.725 0.076 5.342 0.030 1998.292 27.567 0.135 4.723 0.077 5.249 0.026 1998.375 0.000 0.000 5.212 0.164 0.000 0.000 1998.458 0.000 0.000 5.508 0.074 0.000 0.000 1998.542 0.000 0.000 5.860 0.094 0.000 0.000 1998.625 0.000 0.000 0.000 0.000 0.000 0.000 1998.708 0.000 0.000 6.676 0.066 0.000 0.000 1998.792 0.000 0.000 5.627 0.069 0.000 0.000 1998.875 0.000 0.000 0.000 0.000 0.000 0.000 1998.958 0.000 0.000 3.849 0.064 0.000 0.000
Table 1. Linear correlation coefficients r between monthly tropical means of RH* and De, with P values for each correlation. The P value indicates the probability that the observed correlation could have occurred by chance, assuming 20 degrees of freedom in the data (36). The standard case refers to RH*370 computed using HALOE data from 20°N to 20°S, with Cb weighting of qs as described in the text. Also listed are results at 450 K, results using HALOE data within different latitude limits, and results obtained without Cb weighting of qs (i.e., with uniform weighting). The standard and 450 K cases are those graphed in Fig. 1.
Conditions r P . Standard case 0.70 0.0003 450 K level 0.62 0.0020 HALOE (10°N to 10°S) 0.66 0.0008 HALOE (25°N to 25°S) 0.70 0.0003 qs unweighted 0.53 0.0012
The significance of the correlations, together with the lack of any common instrument between the observations of De and those of RH, enables us to reject the possibility of accidental association due to instrument sampling or accuracy issues (22). The only worry is if some atmospheric constituent happens to be contaminating both the ISCCP and HALOE retrievals in a similar, systematic manner. The only likely contaminants would be aerosols near the tropopause or stratospheric moisture itself (23). Aerosol is measured by the CLAES (Cryogenic Limb Array Etalon Spectrometer) instrument flying together with HALOE (24); examination of CLAES aerosol time series (25) showed essentially zero correlation with the traces in Fig. 1, ruling out aerosols as a spurious source of correlation. Radiative effects of stratospheric moisture cannot account for the magnitude or sign of the correlation (23). Thus, we cannot explain the data other than by accepting the De-RH relationship as physical.
Two straightforward interpretations of this relationship are possible. First, smaller ice crystals may increase RH; second, higher RH may reduce crystal size. These possibilities may be distinguished observationally: Any influence of RH on De would have to operate locally and would cause instantaneous spatial covariability of the two quantities as a result of the short lifetimes (~1 hour) of Cb, whereas influences of De on RH (thus q) would not appear locally because of the long lifetime (»1 day at these altitudes) over which q variations can spread out horizontally. The data show that the global relationship does not exist locally (Fig. 2), ruling out the second interpretation. Before accepting that De controls RH, however, we must ensure that the observed sensitivity is physically reasonable.
Observations of cumulus clouds indicate a distribution of particle sizes that usually drops off steeply below 10 µm and above 30 to 40 µm, with varying proportions within this range and occasional bimodality (26). The simplest way of treating this situation is to assume the presence of two size modes, taken here to be D1 = 10 µm and D2 = 30 µm. The populations of these two modes may be affected by variations in the number of particles nucleated at lower levels. In the deepest Cb, however, the total water content (ND3, where N is the number of crystals) of rising parcels should be unaffected by such influences because it is strongly regulated in Cb by cloud-dynamical factors. This assumption implies
for microphysical variations.
Smaller ice crystals should evaporate much more rapidly than larger ones and should settle more slowly, thus leaving vapor closer to the tropopause. If we treat the crystals as ice spheres, then Stokes flow applies at these sizes and their terminal fall speed scales as D2. Sublimation for these sizes is governed primarily by diffusion with only minor effects due to ventilation (27), so it will be approximately proportional to surface area ND2. Therefore, the amount of vapor q' sublimated into a thin layer near the top of the outflow by a unit mass of ice scales simply as
If we consider an increase 
in the proportion of small particles, then Eqs. 1 and 2 imply an increase in
sublimation rate of 
To relate variations of the source q' to those of the ambient amount q,
it is reasonable to assume that q = aq', where a is a constant
(28). The effective diameter
decreases for positive e by an amount that depends on the population ratio N1/N2.
For N1 = N2, De changes from 28 µm to (28 – 3.7
)
µm.
Putting these results together, we obtain an estimated 26% increase in q per 1.0-µm decrease in De (independent of a). This estimate is well within a factor of 2 of the empirical ratio in Fig. 1. The estimate can be altered by roughly a factor of 2, either by changing the mode population ratio by a factor of 10 or by changing the size distribution to a very wide or unimodal one. Although our calculation is a crude one, it shows that the interpretation of Fig. 1 as control of q by De is physically reasonable.
Fig. 2. Mean De versus RH. Plotted points were obtained by binning all retrievals in 1-month 3 5° latitude 3 10° longitude boxes and averaging boxes by RH*370. Land results include only boxes that are mostly land; ocean results include only boxes that are mostly ocean. The dashed lines show the regression slope from the comparison in Fig. 1.
Can this effect explain the observed increases in stratospheric moisture over time? For this to be so, mean De would have to show a decrease of ~1 µm over the past 50 years. Unfortunately, trends in tropical-mean De cannot be verified using AVHRR at this time because of the lack of an adequate absolute solar reflectivity calibration. However, indirect evidence for them does exist.
Aerosols are capable of nucleating new droplets in cumulus clouds, giving them the potential to increase ice particle numbers and reduce their sizes (29). A recent examination of De in Cb (9) found that spatial and temporal variations in De on interannual and longer time scales were closely associated with variations in smoke and/or dust generated in biomass burning regions. The semiannual variation of De in Fig. 1 is in phase with the semiannual variation in biomass burning, which occurs predominantly in the spring of both hemispheres. A variety of regional trends in De, apparently due to aerosol, were found in different continental regions and spanned a range of ~1.5 µm/decade. The trends lack absolute calibration, but if only a small majority of them were toward smaller De (more aerosol), their overall average could easily match the –0.2 µm/decade needed to explain the stratospheric moisture increase.
Estimation of the global mean trend in De may be attempted by multiplying its
empirical sensitivity to aerosol by the estimated trend in biomass burning.
According to one estimate, tropical biomass burning has increased by 50% since
the middle of the 19th century, with most of the increase occurring in the last
50 years (30). For this to explain the stratospheric moisture trend (assuming
linear sensitivity) would require 
De, the reduction in tropical-mean
De brought
about by all present-day burning, to be 2 to 3 µm. This would be moderate
compared to observed climatological De variations. However, quantifying De
accurately is very difficult without more extensive aerosol information, because
it is impossible to tell what De would be in the absence of burning sources,
especially over oceans. Half of Cb occur over oceans, and most continental Cb
occur after the local burning season, so De
would definitely fall short of the
required 2 to 3 µm unless aerosol sources were currently affecting clouds at
considerable distances. However, this may actually be the case if clouds in
unpolluted regimes (such as over oceans) are highly susceptible to small inputs
of smoke, as suggested by some observations (31). Nor is biomass burning the
only potential source of aerosol; airborne dust may also have increased as a
result of land-use impacts or circulation changes (32).
Further studies with more capable instrumentation will be vital to a better understanding of how aerosols and water vapor may be connected by cloud mechanisms. A finely calibrated, long-term record may prove necessary before we can draw more definitive conclusions concerning stratospheric moisture trends.
References and Notes
1. K. H. Rosenlof et al., Geophys. Res. Lett. 28, 1195 (2001).
2. D. J. Seidel, R. J. Ross, J. K. Angell, G. C. Reid, J. Geophys. Res. 106, 7857 (2001).
3. X. L. Zhou, M. A. Geller, M. H. Zhang, J. Geophys. Res. 106, 1511 (2001).
4. P. M. D. Forster, K. P. Shine, Geophys. Res. Lett. 26, 3309 (1999).
5. D. B. Kirk-Davidoff, E. J. Hintsa, J. Anderson, D. W. Keith, Nature 402, 399 (1999).
6. S. C. Sherwood, A. E. Dessler, Geophys. Res. Lett. 27, 2513 (2000).
7. H. S. Johnston, S. Solomon, J. Geophys. Res. 84, 3155 (1979).
8. S. C. Sherwood, A. E. Dessler, J. Atmos. Sci. 58, 765 (2001).
9. S. C. Sherwood, J. Clim., in press.
10. The Cb ice crystal size estimates were obtained by empirically decomposing AVHRR 3.7-µm Cb solar reflectances into the product of three terms: one function of viewing geometry, one function of time, and a residual (variance-minimized) term indicating the individual scene reflectance anomaly. The decomposition gives an accurate estimate of variations in effective diameter De (the mean ice crystal diameter weighted by surface area) from which orbital and scanning variations have been removed.
11. J. E. Harries et al., J. Geophys. Res. 101, 10205 (1996).
12. S. C. Sherwood, J. Geophys. Res. 105, 29489 (2000).
13. J. E. Frederick, A. R. Douglass, Mon. Weather Rev. 111, 1397 (1983).
14. E. J. Highwood, B. J. Hoskins, Q. J. R. Meteorol. Soc. 124, 1579 (1998).
15. E. M. Weinstock, E. J. Hintsa, A. E. Dessler, J. G. Anderson, Geophys. Res. Lett. 22, 3231 (1995).
16. P. W. Mote et al., J. Geophys. Res. 101, 3989 (1996).
17. The saturation mixing ratio qs(t ) is first averaged in monthly 5° X 10° longitude-latitude bins, then the bin values for each month are weighted by that month’s bin Cb populations (or not, for the unweighted case) and averaged. Condensed water may occur independently of Cb in the form of thin cirrus clouds, but observations indicate that the horizontal distribution of these clouds is similar to that of Cb, so the weighting should be appropriate regardless of which cloud type is most important.
18. A table of the individual monthly data and their sample uncertainties is available on Science Online at
www.sciencemag.org/cgi/content/full/295/5558/1272/DC1. (above)19. These lag times were computed from a monthly climatology of radiatively balanced vertical velocities calculated by K. Rosenlof (16).
20. R. A. Plumb, J. Geophys. Res. 101, 3957 (1996).
21. The difference between correlation coefficients (0.70 and 0.53) has only a 12% likelihood of being this large by chance, even under the overly conservative assumption that the experiments are completely independent.
22. Errors in De and RH* due to poor sampling or random instrumental errors contribute only to the uncorrelated variance and are accounted for in the cited significance figures.
23. Aerosols in the lower stratosphere are known to affect HALOE retrievals. These effects are removed as part of the retrieval algorithm, but removal may not be perfect. Stratospheric moisture slightly absorbs the radiation used to observe De, but absorption by higher humidity would lead to larger De, the opposite of the observed relationship.
24. S. T. Massie et al., J. Geophys. Res. 101, 9757 (1996).
25. S. Sherwood, data not shown.
26. H. Pruppacher, J. D. Klett, Microphysics of Clouds and Precipitation (Kluwer, Dordrecht, Netherlands, ed. 2, 1997).
27. W. Ji, P. K. Wang, J. Atmos. Sci. 56, 829 (1999).
28. The linearity assumption follows by noting from Fig. 1 that there was no apparent lag between variations in De and RH, and that the dehydration rate below 450 K was roughly proportional to q. The former circumstance implies a quasi-steady balance between sources and sinks of q near the tropopause, whereas the latter implies that the sinks (thus sources) must be proportional to q.
29. V. T. J. Phillips, T. W. Choularton, A. M. Blyth, J. Latham, Q. J. R. Meteorol. Soc., in press.
30. R. A. Houghton, in Global Biomass Burning, J. S. Levine, Ed. (MIT Press, Cambridge, MA, 1991), pp. 321–325.
31. D. Rosenfeld, Science 287, 1793 (2000).
32. C. Moulin, C. E. Lambert, F. Dulac, U. Dayan, Nature 387, 691 (1997).
33. C. S. Bretherton, M. Widmann, V. P. Dymnikov, J. M. Wallace, I. Blade´, J. Clim. 12, 1990 (1999).
34. G. H. Orcutt, S. F. James, Biometrika 35, 397 (1948).
35. M. S. Bartlett, J. R. Stat. Soc. 98, 536 (1935).
36. The number 20 was obtained by visually estimating the number of independent fluctuations in the data and subtracting 1. Computations using two common formulas based on autocovariance models (33) for general (34) and red-noise (35) processes yielded 18 and 37 degrees of freedom, respectively. All these estimates are high enough to yield strongly significant results.
37. I thank A. Dessler for generously supplying key interpolated data sets, K. Rosenlof for providing vertical velocity calculations, A. Heymsfield for useful advice, and H. Zeleznik for editorial assistance. The ISCCP AVHRR data were obtained from the NASA Langley Atmospheric Sciences Data Center. Supported NASA Earth Observing System Interdisciplinary Science program grant UPN 291-01-91.
6 August 2001; accepted 9 January 2002
|
If you have come to this page from an outside location click here to get back to mindfully.org |
