The PATMOS-x level 2b climatology, generated using three decades of AVHRR measurements, contains valuable information about the past global cloud record. We extract climatologies of tropical deep convective clouds from the PATMOS-x data set, based on the 10.30–11.30 μm brightness temperature. A comparison of the cross tropopause convective cloud frequency between ISCCP and PATMOS-x shows that PATMOS-x has a greater frequency of occurrence than does the ISCCP, and this enhanced frequency is attributed to greater horizontal resolution (2 km) in the PATMOS-x data. The high resolution makes this dataset suitable for a search for cross tropopause convection, which happens on length scales down to 1 km. We find there have been several changes in deep convective activity over land during the period 1982 to 2009. We explore specifically the epoch of the HALOE satellite, and find a correlation between land deep convective activity and anomalies in the HALOE stratospheric water retrievals. A simple model is able to predict stratospheric water vapor concentrations highly correlated to that observed using only frequency of deep convection. From this we conclude that deep convection over land contributes to moistening of the lowest tropical stratosphere on seasonal, annual and decadal timescales.
 Water vapor is an important constituent of the stratosphere. Despite its relatively small mixing ratio, stratospheric water vapor has a large influence on the radiation budget of the Earths atmosphere [Forster and Shine, 2002]. Recently it was estimated that a 10% decrease of stratospheric water vapor mixing ratio (WVMR) in 2001 and the following years may have reduced the rate of global warming by 25% [Solomon et al., 2010], and in this way partly explains the slowdown of global warming in the 21st century, discussed, e.g., by Swanson and Tsonis . The mechanisms regulating the stratospheric water vapor are currently under discussion [Schoeberl et al., 2008; Mote et al., 1996; Peter et al., 2006; Liu et al., 2007]. Essentially there are two concepts in the discussion, built on two different physical processes which are not mutually exclusive. The first is the dehydration process occurring at the tropical tropopause. During the slow ascent through the tropical tropopause air masses pass through ice clouds allowing excess water to sediment out [Fueglistaler and Haynes, 2005; Randel et al., 2006]. The second is the possible role of continental Deep Convective Clouds (DCCs) penetrating the tropopause [Nielsen et al., 2007; Khaykin et al., 2009; Liu et al., 2010; Chaboureau et al., 2007; Corti et al., 2008]. During such events water may be transported to the stratosphere. In this paper we investigate this last concept by pure observational means.
2. Data Analysis
 The PATMOS-x level 2b data contains global fields of the 11 μm brightness temperature, T11μm, derived directly from radiances measured by the polar orbiting AVHRR instrument. The instrument has a horizontal resolution of around 2 km, and is subsampled to fit a global equal-angle 0.1 degree grid. The down-sampling is done in a way that preserves statistical properties, while only a fraction of the measurements are used to produce the dataset. The 5-channel AVHRR record begins in 1981, while the 4-channel record goes back to 1978.
 Both the PATMOS-x (PATMOS, 2010, available at http://cimss.ssec.wisc.edu/patmosx/) (1982–2009) and the ISCCP (2010, available at http://isccp.giss.nasa.gov/products/products.html) (1983–2007) data sets possess the record-length required for analysis of multi-decadal trends in stratospheric deep convective cloud occurrence. Figure 1 shows the 2007 monthly diurnally corrected relative frequency (discussed below) of occurrence of clouds with top temperatures (CTT) less than 215 K for both climatologies. Qualitatively PATMOS-x and ISCCP are in fair agreement, with peak occurrence during boreal spring and fall and minimal occurrence in summer, but the relative occurrence of the cold cloud tops for ISCCP is roughly half that of PATMOS-x. The reason for this discrepancy is the pixel resolution of the records. The ISCCP B1 record has a pixel resolution of 10 km compared to 2 km at nadir for PATMOS-x. The coarser spatial resolution of the ISCCP product allows for averaging of deep convective cloud tops with warmer surrounding cloud, resulting in an overall warming of cloud top temperature measurements. For this reason we chose the higher resolution PATMOS-x record as a more apt fit for this analysis.
3. Time Series of Deep Convective Cloud Frequencies
 Monthly DCC frequencies are calculated from 45S to 45N for each 2 degrees latitude and 1 km altitude band separately for oceans and land. Due to drift in orbit the Local Time of Day (TOD) is not conserved over time.
 In Figure 2 the frequency of deep convective clouds with T11μm < 200 K (DCC<200K) are mapped for each TOD through the whole PATMOS-x era, for both land and ocean. The diurnal cycle is illustrated in Figures 2a and 2b, while the annual cycle is seen in Figure 2c. Clearly there is a much stronger DCC<200K frequency and variability in the land deep convection, than in the oceanic deep convection. In the following analysis we will focus on land deep convection.
 We perform a correction for the diurnal variability by dividing the frequencies with a diurnal cycle (specific for latitude, altitude and month), but which has been normalized to be equal to unity at 19:00 Local Time (19:00 LT), which is in the most intensely convective period of the day. The underlying assumption is that the shape of the diurnal variation is conserved for a specific month and location so by performing this normalization we shift the cloud frequency measured any time of day to the value it presumably would have had at a specified time of day.
 The satellite samples a band of local TODs for each overpass, seen as a band in Figure 2, but the DCC frequency is increased slightly at the edge of these bands. This is due to the fact that the edges are associated with larger satellite viewing angle, thus larger optical depth and consequently larger DCC frequency. Instead of compensating for this we choose to restrict our analysis to only include viewing angles below 30 degrees. An altitude proxy, ΔT = TTropopause(NCEP) − T11μm, is calculated from the brightness temperature of each incident. All incidences colder than the tropopause are counted as “tropical stratospheric clouds” (TSCs). Note that we choose the monthly averaged (climatic) local tropopause temperature from the NCEP model as reference, since by doing so we are not introducing possible interannual variability through the NCEP analysis. The rationale behind this decision is expanded in the auxiliary material [see also Gettelman et al., 2009]. A 4-d time/latitude/TOD/temperature histogram is then constructed for the whole period, and normalized to a monthly cloud frequency by dividing with the total number of measurements within each time-latitude-TOD-temperature box. In Figure 2c we show the frequency of continental TSCs, calculated with this method. The northern hemisphere August-September-October tropopause is warmer than the southern hemisphere February-March-April tropopause. This is the main reason for the dominating TSC intensity at the northern hemisphere, seen in Figure 2c.
 The question arises of whether TSCs, as defined in this study, are actually penetrating the tropopause or being constrained to the upper troposphere. To address this question we compare ΔT to an alternative altitude measure, suggested by Schmetz et al. , for a selected set of land tropical clouds in vicinity of the tropopause. Schmetz et al.  performed line-by-line calculations to obtain the difference between the 6.7 microns brightness temperature, T6.7μm, and the 11 microns brightness temperature, T11μm, for a simulated cloud rising through the troposphere and into the stratosphere. The contribution of stratospheric emissions is negligible in the 11 μm window, but not in the 6.7 μm window. This means that T6.7μm is higher than the physical cloud top temperature. A 6.7 μm channel is not included in the AVHRR sensor, but it does exist on the MODIS sensor. We have thus processed a few days of MODIS data for this comparison. Figure 3 shows the result of this analysis. The criterion that we use to identify TSCs place those clouds on the descending (right-side) tail of the scatter plot. This gives us confidence that the deep convective clouds chosen in this study as TSCs are in most cases penetrating into the stratosphere.
 The HALOE data set provides a continuous global record of stratospheric water content from 1993 to 2006. The HALOE version 19 water vapor data is reported as water vapor volume mixing ratio (WVMR) profiles, associated with position and altitude. Features of this dataset are well described by Randel et al. . Annual water vapor variations in stratosphere is dominated by an increased abundance between 16 and 18 km altitude in September October November (SON). The interannual variation is correlated with the Quasi Biennial Oscillation (QBO) phase, and there is a very dramatic drop in WVMR in 2001. We note that the annual cycle of the water vapor mixing ratio (WVMR) in the lowest stratosphere corresponds qualitatively with the TSC peak in boreal autumn, and that there is a drop of TSC intensity around 2001 as well.
4. Attribution of Stratospheric Water Variations to Variations of DCC
 We propose a model to account for the variation of WVMR in a layer of the lowest tropical stratosphere:
Here x is the mean WVMR at a given altitude/latitude range and a represents the background WVMR in the slowly ascending air entering the region just above the tropopause. By assuming a being constant we actually disregard any variability coming from changing tropopause temperatures. The purpose of this obvious over-simplification is to assess how much of the WVMR variability can be explained solely by deep convection. k is the inverse time given by the mean ascent velocity divided by the thickness of the layer, and c is a constant representing the efficiency of the hydration process, that is, we simply assume that the hydration is proportional to the number of observed TSCs (d(t)), counted as the number of incidences where the altitude is over a certain threshold, e.g., the tropopause. The model may be solved:
Luckily the last term vanishes because of the very long spin up time. (d(t) goes back to 1982, while the HALOE data starts in 1992.) In Figure 4 (top) we show as an example a fit of this model, where d(t) was calculated by summing continental TSC frequencies from latitudes between 4N and 20N. We search for the most likely geographical origin of convective moistening by fitting the modeled water vapor mixing ratio x, calculated from different latitudinal bands to the measured (HALOE) tropical water vapor in the same latitude bands. In Figure 4 (middle) we show the correlation coefficient r and the time lag for different combinations of latitude (TSC) and altitude (WVMR). The plot reveals a very good correlation in the tropical northern hemisphere. Here the time lag is also practically zero, thus the latitude range 10–20N is a possible candidate as source for water vapour variation. The southern part of the tropical belt (−25S to 0N), is less favorably correlated, and the time lag is several months.
 The peak at 24N may not contribute much in terms of water transport, since the TSC intensity is weak at that latitude. We note that there is a good correlation between 4N and 20N, where we also find the largest TSC intensity globally (Figure 2c). It is important to note that the correlation is independent of the values of the parameters c and a, thus there is only one parameter, k, which is found to 0.010 d−1 for a model layer thickness of 1 km. At 10N–14N one can obtain a correlation of 0.91. The good correlation is partly due to the annual variability. Fueglistaler and Haynes  showed the same water climatology to correlate with a modelled value derived from the zonal tropopause temperature (r = 0.81). The fact is that we cannot attribute the WVMR variations to one process just based on a phenomenological correlation. However, we note the coinciding latitude, altitude and time of year of the HALOE WVMR anomaly and PATMOS-x TSC frequency suggests a physical connection.
 In Figure 4 (bottom) we present the same analysis for the deseasonalized TSC frequencies and WVMR. In this case we include the 20S–20N latitude and 16–19 km altitude belt. The well known phenomenological connection between the WVMR and the QBO is seen in the deseasonalized data. With the QBO removed, a favorable fit is found of the model to the residual, with a correlation of 0.64. Thus it is plausible that the TSC intensity controls the interannual variations in WVMR. The drop of water vapor mixing ratio in 2001 is evidently preceded by a drop of TSC frequency in the lowest stratosphere.
 The PATMOS-x dataset contains, due to its high horizontal resolution and record length, unique historical information about tropical deep convection. The annual cycle of water vapor mixing ratio, in the lowest stratosphere, shows a pronounced peak in the lower UTLS in late northern summer, which in the literature is attributed to a large upwelling in the Himalayan monsoon [Gettelman et al., 2004]. In our analysis we find that the TSC peak in northern hemisphere autumn corresponds with the peak of the WVMR in the lowest stratosphere. The observed correlation of 0.91 between monthly WVMR and TSC values supports the notion that WVMR variations in the lowest tropical stratosphere is controlled by deep convection.
 We want to thank James Russell III and the HALOE team for helpfulness, and for providing data, and also Ken Knapp from the ISCCP project for providing the raw brightness temperatures for the comparison. The views, opinions, and findings contained in this report are those of the author(s) and should not be construed as an official National Oceanic and Atmospheric Administration or U.S. Government position, policy, or decision.