Abstract

Relative humidity fields from the High-Resolution Infrared Radiation Sounder (HIRS) flown on NOAA series satellites since 1979 have been used to study the seasonal aspects of the interannual variability of relative humidity in the tropical troposphere. The El Niño–Southern Oscillation (ENSO) is the only statistically identifiable physical mechanism of such variability. Boreal winter (December–February) relative humidity variations during an ENSO event follow patterns of anomalous convection and large-scale upper-level circulation. During El Niño (La Niña) regions of large negative (positive) relative humidity anomalies exist at subtropical latitudes over the Pacific Ocean. These are not always balanced by increases (decreases) in humidity near the equator. NCEP– NCAR reanalysis temperatures are used to separate observed changes in relative humidity into contributions from tropospheric temperature versus the contribution from changes in water vapor content. The authors find that at subtropical latitudes variations in temperature contribute between 50% and 70% of the observed change in relative humidity. It is also shown that large relative humidity anomalies exist over the equatorial Indian, Atlantic, and far east Pacific Oceans during the summer season (June–August) following an ENSO event. Ocean– atmosphere dynamics coupled with the seasonal cycle of relative humidity explain the existence of the long-lasting effects of ENSO in the atmosphere. The authors argue that observed linear trends in regional and tropical mean relative humidity are unlikely to be due solely to ENSO or a simple intensification of the hydrological cycle.

1. Introduction

The critical importance of water vapor in the troposphere to the climate system is well documented in the literature (e.g., Lindzen 1990; Shine and Sinha 1991; Del Genio et al. 1994; Peixoto and Oort 1996; Harries 1997; Held and Soden 2000). Variations in water vapor in the cold mid-upper troposphere can have a radiative effect comparable to or greater than equivalent changes in the lower troposphere despite containing a small fraction of the total column water vapor (Shine and Sinha 1991; Held and Soden 2000). Sensitivity of outgoing longwave radiation to changes in tropospheric relative humidity is also nonlinear such that subtropical dry zones are radiatively more sensitive to equivalent changes in humidity than the moist deep Tropics (Spencer and Braswell 1997). It is therefore important to understand the full spatiotemporal variability of this important atmospheric constituent.

The principal mode of interannual variability of upper troposphere relative humidity (UTRH) is recognized as being related to the El Niño–Southern Oscillation (ENSO; Bates et al. 1996). To first order the redistribution of UTRH during an ENSO event follows the patterns of changing convection over the Indo–Pacific domain. Extremes in mean UTRH of the zonal band bounded by 30°N and 30°S, which includes both ascending and descending branches of the Hadley circulation, preferentially occur during Northern Hemisphere winter and spring, and the largest observed tropical UTRH anomalies have occurred during ENSO events (Bates et al. 2001). Moistening through convection is limited by saturation, but drying in the surrounding areas has no such limit (Sun and Lindzen 1993); therefore reduced UTRH over the subtropical oceans during an El Niño event, for example, could lead to a net reduction in UTRH across the tropical region bounded by 30°N and 30°S (Fu et al. 1997). Previous analyses of UTRH and ENSO have concentrated on northern winter (Bates et al. 2001; Blakenship and Wilheit 2001) or zonal mean humidity (Fu et al. 1997) and often analyze isolated events. In this study we use a composite of El Niño and La Niña events to describe regional patterns of ENSO-related variability in UTRH in both the winter and summer seasons. We find that anomalies in UTRH exist over the Indian and Atlantic Oceans two seasons after the peak December–February (DJF) response in Pacific sea surface temperatures (SST).

Previous studies have assumed that in the Tropics seasonal to interannual UTRH variability is dominated by changes in the water vapor content (Peixoto and Oort 1996; Bates et al. 2001). We show that both regional and zonal wintertime extremes in UTRH as a response to ENSO are also strongly influenced by regional changes in tropospheric temperatures over the Pacific Ocean.

Finally we examine long-term trends in UTRH, reviewing the role of changes in ENSO and the hydrological cycle in describing the observations.

2. Data

We analyze a product of the High Resolution Infrared Radiation Sounder channel 12 (HIRS12) brightness temperatures as described in Stephens et al. (1996) and Bates et al. (1996), and modified and extended by Bates et al. (2001).

HIRS12 measures upwelling radiation from the strong 6.7-μm emission band of water vapor, sensitive to relative humidity and temperature in a broadband layer between approximately 500 and 200 hPa, with a peak sensitivity in the tropics at approximately 300 hPa. We will refer to the observations as being upper troposphere relative humidity. We acknowledge that there are significant contributions from much of the mid to upper troposphere, but we wish to maintain some consistency with terminology used in previous studies (see references in previous paragraph). We prefer the acronym UTRH rather than the commonly used UTH as a constant reminder that the observed variable is not only sensitive to changes in water vapor, but also temperature.

HIRS is an infrared sensor and therefore requires cloud clearing. This introduces an unknown systematic dry bias and possible sensitivity to changes in regional cloud characteristics. For the purposes of this study we will assume that this sampling bias does not significantly affect the interannual variability.

The single largest source of uncertainty in the HIRS UTRH dataset is the different response functions of each instrument used over the observing period, coupled with the lack of an in situ standard of suitable quality to provide absolute calibration. Bates et al. (1996, 2001) have corrected the UTRH dataset with respect to the instruments aboard NOAA-10 and NOAA-7 satellites to produce a self-consistent dataset of global observations on a 2.5° × 2.5° latitude–longitude grid for the period March 1979–May 1998.

A number of ancillary data have been used to quantify the dynamical mechanisms controlling UTRH. From the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) 40-year reanalysis dataset (Kalnay et al. 1996) velocity potential at 0.2σ (close to 200 hPa) level, wind vectors at 300 hPa and temperature at 400 and 300 hPa have been used. These levels were chosen as they are closest to the peak in the sensitivity of the HIRS channel 12. We have also used the National Oceanic and Atmospheric Administration's (NOAA's) interpolated outgoing longwave radiation (OLR; Liebmann and Smith 1996) and the Met Office Hadley Centre Sea Ice and Sea Surface Temperature (HadISST) dataset (Rayner et al. 2003) for sea surface temperatures.

ENSO years were identified based upon the Niño-3.4 index of central Pacific SST anomalies in the region bounded by 5°N and 5°S, 170°W and 120°W (Kousky 1996; Trenberth 1997). ENSO is a coupled ocean–atmosphere phenomenon; therefore, we used an oceanic index for ENSO in our analysis of the atmospheric response.

3. Method

a. EOF and composite analysis

We initially conducted an empirical orthogonal function analysis (not shown) on mean fields of each season December–February (DJF), March–May (MAM), June– August (JJA), and September–November (SON) and found that in each case ENSO variability is the only uniquely identifiable large-scale physical mechanism for interannual variability of UTRH. This mode describes approximately 20%–25% of the total interannual variance in each season for the tropical region bounded by 30°N and 30°S. Limitations of this statistical approach were reached in JJA when the first two modes were found to be mixed (North et al. 1982). The principal component time series of JJA UTRH correlated positively with the JJA mean Niño-3.4 index of Pacific SSTs with a correlation coefficient of +0.75, while the second had no significant correlation to coincident JJA Niño-3.4 but a correlation coefficient of +0.72 with the preceding DJF Niño-3.4. This suggested that these modes represented patterns of onset and decay of ENSO, respectively, during JJA.

Having identified ENSO as the only major pattern of interannual variance in UTRH we are able to statistically determine, we present our results as composite anomaly fields of the difference between the mean of four El Niño and three La Niña events for the boreal seasons DJF (winter) and JJA (summer) to quantify aspects of this mode of variability. Taking composites in this manner is a useful measure of the strength of regional ENSO variability in UTRH, as we do not calculate anomalies from the long-term average, which is itself made up of a non-even mix of El Niño and La Niña events, and do not restrict the analysis to a single event or pair of events.

Representative El Niño events within the observed period are 1982/83, 1986–88, 1991/92, and 1994/95. In this study the period of 1987/88 was used instead of 1986/87, as we are particularly interested in capturing the decay phase of El Niño. Similarly, 1998 was not included as the dataset extends only to May of that year. There is only one strong La Niña event in the observed period during 1988/89, and two weak La Niña events during 1984/85 and 1995/96. For clarity, unless stated otherwise, results will be discussed as representing a warm El Niño event, where increased UTRH is observed over the central equatorial Pacific and decreased UTRH in the subtropical Pacific.

It should be noted that we are limited to a small population and that 75%–80% of the interannual variance is not directly related to ENSO. Composite differences are shown only where they are significant at the 95% level based on a t test, estimating the standard deviation from the complete 20-yr period. Visual comparison of the composite difference fields with the individual El Niño and La Niña mean fields (not shown) gives us confidence that the significant features presented are manifest in both El Niño and La Niña and are of opposite sign. There is also strong agreement between the composites and the leading EOF modes of variability. For example, compare Fig. 1a (DJF composite) with Bates et al. (2001, Fig. 2b).

Fig. 1.

Composite difference (significant at the 95% level) of HIRS channel 12 UTRH (filled contours) between the mean of four El Niño and three La Niña events for (a) DJF and (b) JJA. Overplotted are 0.2σ level velocity potential contours, solid contours at 0.5, 1, 2, 3, and 4 × 106 m2 s−1; dashed contours at 0, −0.5, −1, −2, −3, and −4 × 106 m2 s−1. Arrows showing 300-hPa vector wind are plotted with reference magnitude of (a) 5 m s−1 and (b) 2.5 m s−1

Fig. 1.

Composite difference (significant at the 95% level) of HIRS channel 12 UTRH (filled contours) between the mean of four El Niño and three La Niña events for (a) DJF and (b) JJA. Overplotted are 0.2σ level velocity potential contours, solid contours at 0.5, 1, 2, 3, and 4 × 106 m2 s−1; dashed contours at 0, −0.5, −1, −2, −3, and −4 × 106 m2 s−1. Arrows showing 300-hPa vector wind are plotted with reference magnitude of (a) 5 m s−1 and (b) 2.5 m s−1

b. Separating temperature and water vapor

We used NCEP–NCAR reanalysis monthly mean temperatures at 300 and 400 hPa (these levels being closest to the peak sensitivity of the HIRS12 UTRH) to create datasets of monthly mean saturated vapor pressure (es), calculated from the Goff–Gratch formulas (Balkan and Fischer 1987).

Relative humidity is defined as the ratio of vapor pressure (e), a function of dewpoint temperature, and saturated vapor pressure (es), a function of temperature, of an air parcel. We define relative humidity (RH) in Eq. (1) below as represented by a mean state (overbar— in this case representing the climatological mean of each of the 12 months of the year) and some deviation (Δ) from that mean:

 
formula

The mean relative humidity is defined as

 
formula

We are interested in deviations from the background mean state rather than time rates of change. We must therefore consider sources of systematic error in our computations due to the nonlinear relationships among the moisture variables. We are computing monthly mean vapor pressure from previously averaged monthly mean temperature and relative humidity. This will result in a bias compared with the monthly mean of daily (point) vapor pressure calculated from daily (point) measurements of temperature and humidity. This bias has been examined by Gaffen et al. (1991) from radiosonde observations. They find that for tropical stations this bias is small and positive (of order 1%). At low latitudes, the temperature range is small; this means that the saturation vapor pressure curve can be reasonably approximated by a line segment. Gaffen et al. conclude that within 30° of the equator the use of monthly mean temperature and relative humidity to estimate changes in specific humidity is a good approximation. Similarly mean RH is not equal to the ratio of the means of the vapor pressure components [see Eq. (2)]. We investigated the difference between climatological mean RH and the ratio of the climatological means of the vapor pressure components for all months and all grid boxes. We find that the bias is typically less than 2% of the mean RH for all grid boxes within 30° of the equator.

We can now make the following assumptions: first, that the use of monthly means does not introduce a large bias and, second, that in Eq. (1) ΔRH is small when Δes and Δe are zero. These assumptions allow us to confidently manipulate Eq. (1) to study deviations from climatology. First, consider the situation of a change in relative humidity at fixed dewpoint (water vapor content). From (1) we get

 
formula

where ΔRHT is the expected change in relative humidity due to a known change in saturated vapor pressure (temperature). The mean vapor pressure (e) is an unknown and has been substituted with esRH.

Alternatively we can assume that the temperature remains constant in Eq. (1). The change in relative humidity as a result of changing dewpoint (ΔRHw) can be estimated as

 
formula

Again the change in vapor pressure has been estimated from es and RH.

We find that ΔRH, the observed deviation from climatology, can be determined from ΔRHT and ΔRHw as

 
formula

The final term (ɛ) on the right-hand side of Eq. (5) represents the errors discussed above and is small, typically less than 2% of the value of RH. The third term, or product term as we will refer to it, is not necessarily small. However, when the monthly mean deviations are averaged over longer time scales, as in the composite analysis, we find that this term is generally small and less than 5% of the total sum.

Equations (3)–(5) provide a basis from which we can determine monthly mean temperature and water vapor contributions to RH anomalies from saturated vapor pressure and RH.

4. Composite ENSO signals

Figure 1 shows ENSO composite difference fields of HIRS12 UTRH for (a) DJF and (b) JJA. Light shading indicates regions where we observe reduction in UTRH, significant at the 95% level, and dark shading indicates an increase. Near the equator negative (positive) anomalies of velocity potential are taken to correspond to regions of anomalous upper-level divergence (convergence) related to enhanced (suppressed) large-scale convection. There is striking agreement between the position of anomalies of velocity potential and UTRH along the equatorial region in both seasons. It is clear that the changes in UTRH related to ENSO through the deep Tropics, 10°S–10°N, are a direct response to changes in location and intensity of tropical deep convection.

a. Winter

During DJF anomalous convection is found over the central Pacific and increased subsidence over the Maritime Continent and continental South America. This pattern is clearly represented in Fig. 1a. The magnitude of UTRH differences over the central equatorial Pacific and Maritime Continent, for example, are in the range 10%–15%.

Equatorial upper-level winds east of the date line in the Pacific are anomalously easterly. On initial inspection this might be interpreted as convergence into the high UTRH region in the vicinity of the date line. The anomaly here, however, represents a weakening, not a reversal, of upper-level equatorial westerlies. Velocity potential and divergence fields (not shown) confirm that there is net divergence close to the date line in the composite. The westerly subtropical jets of both the Northern and Southern Hemispheres are intensified, and we observe large-scale anomalous upper-level anticyclonic systems on either side of the equator over the Pacific. These are regions of anomalous subsidence and consequently result in strong negative UTRH difference of order −10% south of the equator and −15% north of the equator. Bates et al. (2001) suggest that variations in Rossby wave activity related to the changes in the base-state circulation during an ENSO event modulate the water vapor flux into the tropical upper troposphere, affecting the mean Tropics-wide relative humidity. Intensified subtropical westerlies during El Niño might also be expected to advect climatologically dry air from the west. We return to this discussion in section 4c.

Over the equatorial Atlantic convection is suppressed. Upper-level westerly wind anomalies along the equator and increases of UTRH in the subtropical Atlantic are also observed. Humidity increases are found over Western Australia and the southern Indian Ocean and east of China at around 25°N.

b. Summer

Figure 1b shows the composite response to ENSO during the boreal summer, JJA, season following an ENSO event. Overall there is little significant response during this season; however centers of significant UTRH anomalies lie in the Atlantic and Indian Oceans and over the far eastern equatorial Pacific. Two seasons following the peak in Pacific Ocean SST anomalies, regional differences in UTRH exist of order 10% outside the Pacific, while the large-scale characteristic ENSO signature over the Pacific has largely disappeared. These tropical changes follow patterns of anomalies in velocity potential, suggesting that there are changes in the character of tropical convective processes over remote oceans many months following the peak of an ENSO event. The largest UTRH signal is found over the western Indian Ocean north of Madagascar.

ENSO composite difference fields for (a) SST and (b) OLR, both for JJA, are presented in Fig. 2. We observe warming of the tropical Indian Ocean during post-ENSO JJA, particularly in the north and west. Changes in atmospheric circulation accompanying ENSO events induce changes in local cloud and evaporation characteristics and lead to changes in the surface net heat flux over remote oceans (Klein et al. 1999, their Fig. 2c). This can be seen as a positive SST difference in the Indian Ocean and South China Sea in Fig. 2a. Ocean dynamics also result in localized SST anomalies. In the tropical south Indian Ocean SSTs are strongly influenced by subsurface thermocline variability and ENSO is a dominant forcing of this. An El Niño forces a westward propagating downwelling Rossby wave in the south Indian Ocean, resulting in delayed warming of the southwest Indian Ocean (Xie et al. 2002). There exist mechanisms for atmosphere–ocean dynamics to result in a warm Indian Ocean following an El Niño. The peak lag correlation for the Indian Ocean warming is about 3 months, but there is evidence that these anomalies persist into the summer (JJA) season (see Klein et al. 1999, Fig. 6c; Xie et al. 2002, Fig. 11b).

Fig. 2.

As in Fig. 1 but for the JJA season of (a) SST, contours at −1°, −0.75°, −0.5°, 0.5, 0.75°, and 1°C and (b) OLR, contours at −10, −5, −3, 3, 5, and 10 W m−2

Fig. 2.

As in Fig. 1 but for the JJA season of (a) SST, contours at −1°, −0.75°, −0.5°, 0.5, 0.75°, and 1°C and (b) OLR, contours at −10, −5, −3, 3, 5, and 10 W m−2

Negative differences in OLR (light shading in Fig. 2b) are associated with increased convection and cloudiness during El Niño events. The general pattern of OLR and UTRH differences are similar; where UTRH differences are observed we find OLR changes of the expected sign. Accordingly, reduced OLR is found over the western Indian Ocean and the central and far east Pacific, and positive values over Brazil.

During MAM (not shown) there are positive UTRH differences of approximately 5% over the south Indian Ocean at 20°S and around 80°–90°E, somewhat weaker than changes observed during JJA despite the larger SST anomalies in this season. Warming of the Indian Ocean is not in itself sufficient to explain the large anomalies in UTRH that occur in JJA in this region. In order to understand this relative moistening over the western Indian Ocean, normally a dry, relatively clear sky region, we consider the UTRH climatology. During JJA (Fig. 3) strong horizontal humidity gradients exist along approximately 10°S and 10°–15°N in the climatology. A tongue of low humidity extends across the equator along the East African coast. UTRH is very high along much of the equatorial and north Indian Ocean. Such strong humidity gradients do not exist in other seasons. Greater westerly extent or horizontal advection of high humidity air from the eastern Indian Ocean would be expected to have a dramatic effect along these humidity fronts. We interpret the observations as a northeast–southwest extension of the region of high humidity associated with the north Indian Ocean during JJA, supported by warmer surface temperatures in the west. This is coupled to the climatology characterized by large humidity gradients that result in large UTRH anomalies in the summer season during the decay of El Niño.

Fig. 3.

JJA 1979–97 climatology of HIRS channel 12 UTRH

Fig. 3.

JJA 1979–97 climatology of HIRS channel 12 UTRH

Further evidence of a relationship between UTRH anomalies in JJA over the western Indian Ocean and ENSO variability is seen by analyzing time series of JJA mean UTRH in the area bounded by 10°S and 10°N, 50° and 70°E and SSTs in the extended region 10°S– 10°N, 50°–90°E (SST anomalies are more basinwide than those seen in UTRH). The correlation coefficient between time series of JJA SST and the preceding DJF Niño-3.4 is +0.7, supporting Klein et al. (1999). A correlation coefficient of +0.6 exists between JJA UTRH and JJA SST over the period studied and a correlation of +0.49 between JJA UTRH and DJF Niño-3.4. No significant correlation was found between western Indian Ocean UTRH and the intensity of the Indian summer monsoon based upon time series of all-India rainfall (Parthasarathy et al. 1995). This leads us to conclude that ENSO is the principal driving mechanism behind this observed extension of high humidity at upper levels over the western Indian Ocean.

UTRH differences of 5%–10% exist in the far eastern equatorial Pacific Ocean (Fig. 1b). Again we observe positive SST differences, perhaps due to persistence of the classic El Niño pattern. It is also in a region of strong mean latitudinal humidity gradient and suggests a more southerly extent of the Central American monsoon high humidity region. Reduced humidity over South America and the Atlantic is accompanied by weak anomalous southeasterlies on the background westerly flow, suggesting that advection of dry air from the east and increased subsidence result in the observed reduction in UTRH in this region.

c. Temperature

We have investigated the impact of temperature changes in the upper troposphere on relative humidity during ENSO events. During JJA the composite temperature difference for the tropical region is between 0.5° and 1°C everywhere. This contributes to a small change in the saturated vapor pressure, and to a small net reduction in the relative humidity composite everywhere. We can therefore be confident that the patterns of UTRH anomalies during this season are largely due to regional changes in water vapor content.

In the northern winter (DJF) season this is not the case. Larger regional composite temperature differences of order 2°–3°C exist in the subtropical anticyclone regions of the central-east Pacific. Figure 4 shows the composite difference of (a) RHT and (b) RHw, as estimated from Eqs. (3) and (4) using reanalysis temperatures at 300 hPa. The sum of Figs. 4a and 4b approximate to the observed humidity composite in Fig. 1a since the final terms in Eq. (5) are small everywhere in the composite. Over the east Pacific we estimate that temperature changes have a strong impact on the relative humidity field. For example, in the region roughly bounded by 10° and 20°N, 150° and 130°W we find that the UTRH composite difference is of order 10%–15% (Fig. 1a), while RHT (Fig. 4a) is of order 6%–12% and RHw (Fig. 4b) is of order 3%–8%. On average temperature variations alone can explain 50%–70% of the observed changes in UTRH in the subtropical Pacific region. This range represents the sensitivity of our estimates to using temperatures at 300 hPa (Fig. 4) or 400 hPa (not shown) and to the exact latitude–longitude range considered. We conclude that large deviations in UTRH in the subtropical east Pacific region during El Niño (La Niña) are the result of a combination of upper-level warming (cooling) and drying (moistening), the principal effect being that of temperature change. Temperature changes would also appear to be an important consideration over the Amazon basin, again due to subsidence warming, and off the coast of China between 25° and 30°N where the reanalysis suggests upper-level cooling. Over the equatorial Pacific significant changes in UTRH are observed and are the result of large changes in the water vapor content as a result of the shift in location of deep convection.

Fig. 4.

As in Fig. 1a but for estimated change in UTRH resulting from (a) temperature changes (RHT) and (b) changes in specific humidity (RHw)

Fig. 4.

As in Fig. 1a but for estimated change in UTRH resulting from (a) temperature changes (RHT) and (b) changes in specific humidity (RHw)

Does ENSO have an observable impact on the mean relative humidity of the zonal band bounded by 30°N and 30°S? This zonal band, which we will refer to as “Tropicswide,” incorporates both the ascending and descending branches of the tropical circulation. Changes in the exact latitude range used had little effect on the results. We do not observe a significant net change in Tropicswide UTRH in the composites. Table 1 details the 30°N–30°S mean UTRH and its breakdown into components on the right-hand side of Eq. (5). Table 1 also includes a number of ENSO years that were not included in the composite analysis. For reference the standard deviation of DJF UTRH is 0.84%. The picture is somewhat confused. In general, we expect temperature variations to result in a Tropicswide reduction (increase) in UTRH during an El Niño (La Niña), and water vapor variations to act in the opposite sense. There are a number of exceptions to this, notably the El Niño winter of 1994/95, the lowest humidity DJF season in the observed period, which we estimate to be a result of tropospheric drying that does not occur in any other El Niño winter. We find that the net Tropicswide UTRH in the strong El Niño of 1982/83 and La Niña of 1988/ 89 is a result of tropical moistening (drying) being insufficient to balance large changes in the UTRH due to regional changes in the temperature field. It should be noted that the intensity of the ENSO event does not have a simple relationship with tropical mean humidity. For example, the large El Niño of 1998 was accompanied by near-normal tropical mean UTRH during DJF as changes in the tropical water vapor content balanced the effect of warming.

Table 1.

Anomalies in UTRH for the DJF season. Year refers to the year of Jan and Feb. Seasonal anomalies are given for the Niño-3.4 index of Pacific SSTs. ΔRH is the observed UTRH anomaly; ΔRHT and ΔRHw the estimated temperature and water vapor components of the UTRH anomaly. The final column is the error term from Eq. (5). Rows in bold are those years that were included in the composite analysis

Anomalies in UTRH for the DJF season. Year refers to the year of Jan and Feb. Seasonal anomalies are given for the Niño-3.4 index of Pacific SSTs. ΔRH is the observed UTRH anomaly; ΔRHT and ΔRHw the estimated temperature and water vapor components of the UTRH anomaly. The final column is the error term from Eq. (5). Rows in bold are those years that were included in the composite analysis
Anomalies in UTRH for the DJF season. Year refers to the year of Jan and Feb. Seasonal anomalies are given for the Niño-3.4 index of Pacific SSTs. ΔRH is the observed UTRH anomaly; ΔRHT and ΔRHw the estimated temperature and water vapor components of the UTRH anomaly. The final column is the error term from Eq. (5). Rows in bold are those years that were included in the composite analysis

In summary, of the five ENSO winters when the Tropicswide mean UTRH anomaly exceeded 1%, three winters (1982/83, 1983/84, and 1988/89) were dominated by temperature rather than water vapor changes, and two (1984/85 and 1994/95) were dominated by a net reduction in water vapor mixing ratio, in the case of 1994/95 an unexpected result considering the sign of water vapor changes in all other ENSO events. These results do not support the idea that observed changes in transient mixing in the Pacific explain Tropicswide extremes in UTRH, rather that the Tropicswide net change is some imbalance between changes in relative humidity due to both temperature and water vapor changes. ENSO describes a pattern of UTRH variability but has no simple association with the Tropicswide mean UTRH.

5. Trends

Linear trends in HIRS UTRH are discussed in Bates and Jackson (2001, hereafter BJ). They identify trends in regional and zonal mean UTRH and suggest that increased frequency of warm ENSO events in the 1990s versus the 1980s and an intensified hydrological cycle may explain the observed linear trends.

Figure 5 shows results of linear trend analysis on (a) regional and (b) zonal monthly mean anomalies of HIRS12 UTRH, for the reference period March 1979 through August 1997. Trends are only shown where they are significant at the 95% level. Positive trends of greater than 0.3% yr−1 are observed in the equatorial western Indian Ocean region and parts of Africa and the Amazon (Fig. 5a). Negative trends of similar magnitude are found along the Southern Ocean convergence zones and over oceanic regions between 25° and 45° latitude in both hemispheres (Fig. 5b). Trends in zonal mean humidity between 10°N and 10°S (+0.08% yr−1), and 25° and 45°S (−0.1% yr−1) are significant at the 95% level, but not at the 99% level. These trends of opposite sign result in no detectable trend in area averages of the whole tropical region bounded by 30°N and 30°S.

Fig. 5.

Trends in (a) regional (where significant at the 95% level) and (b) zonal mean UTRH for Mar 1979–Aug 1997, in % yr−1. Dotted lines in (b) represent significance at the 95% level accounting for autocorrelation of the data

Fig. 5.

Trends in (a) regional (where significant at the 95% level) and (b) zonal mean UTRH for Mar 1979–Aug 1997, in % yr−1. Dotted lines in (b) represent significance at the 95% level accounting for autocorrelation of the data

Analysis of HIRS channel 4 (sensitive to tropospheric temperature) in BJ suggests that the humidity changes cannot be reconciled with temperature changes. The magnitude of tropospheric temperature trends is, however, currently a source of much debate (see e.g., Christy et al. 2000; Vinnikov and Grody 2003; Mears et al. 2003).

We can estimate time rates of change of temperature and humidity by using a simple formulation for es (Balkan and Fischer 1987):

 
formula

where a and b are constants. This approximation has a relative error of less than 1% of es. Setting vapor pressure e to be constant we can estimate the time rate of change of relative humidity from

 
formula

where y = RH b/T2, and T is in kelvin. We calculated the mean and variance of the coefficient y using temperatures from the reanalysis at 300 hPa, and b = 6147.795. The mean of y is 4.4 and the standard deviation of the estimate 1.1 for the Tropics. Using a range of values for y between 2.2 and 6.6 we then estimated temperature trends that would result in relative humidity trends of greater than 0.1% yr−1 assuming constant water vapor. The temperature trend required for this must be of a magnitude greater than 0.15 K decade−1, with cooling close to the equator and warming at subtropical latitudes. This is outside the current range of estimates of temperature trends in the tropical region from the Microwave Sounding Unit (MSU) satellite records created by either Christy et al. (2000) or Mears et al. (2003). If real, the observed trends in UTRH are representative of low frequency variations in tropical water vapor rather than temperature.

a. ENSO

In this study we did not include the 9-month period September 1997 through May 1998 so that the linear trend analysis would not be affected by the presence of a large El Niño event at the end of the series. Comparing our Fig. 5a with Fig. 1b of BJ, we note that small regional trends over the equatorial Pacific, east Indian Ocean, and Maritime Continent are not captured by our analysis for this reason.

Regions of strongest trends are not located in regions most sensitive to ENSO variations, most notably over the Pacific Ocean. We removed El Niño and La Niña years, as identified in section 3, from the dataset and recalculated the trends. We found the trends in zonal mean UTRH to be largely unaffected by this crude attempt to remove the effects of ENSO. We conclude that occurrence of such events in the period is not sufficient to explain observed trends in UTRH.

b. The hydrological cycle

Trends in the equatorial domain are found in all seasons, while the southern subtropical trend is largest in the local winter season (JJA). Similarly the small (not significant) negative trend at northern subtropical latitudes (Fig. 5b) is largest during DJF and MAM. The sign, distribution, and seasonality of the trends suggest that the trends may represent a change in the intensity of the tropical hydrological cycle.

In Fig. 6 we present time series of mean UTRH within the latitude range (a) 10°N–10°S and (b) 25°–45°S where the strongest trend in the zonal means are observed. We find that an upward shift in UTRH in the equatorial region is largely confined to the period from 1984 to 1988, and a downward shift at subtropical latitudes during 1987 to 1994. The segmental changes in Fig. 6 are responsible for the linear trends presented in Fig. 5b. These changes of opposite sign are not contemporaneous. They therefore cannot represent compensating changes, as would be expected to accompany a change in the overturning circulation.

Fig. 6.

Time series of monthly mean UTRH anomalies with respect to mean annual cycle for Mar 1979–Aug 1997, smoothed with a 5-month running mean. Means bounded by (a) 10°N and 10°S and (b) 25° and 45°S. Solid straight line indicates region of greatest trend, defined as the periods (a) Jan 1983–Jan 1989 and (b) Jun 1986–Jun 1994

Fig. 6.

Time series of monthly mean UTRH anomalies with respect to mean annual cycle for Mar 1979–Aug 1997, smoothed with a 5-month running mean. Means bounded by (a) 10°N and 10°S and (b) 25° and 45°S. Solid straight line indicates region of greatest trend, defined as the periods (a) Jan 1983–Jan 1989 and (b) Jun 1986–Jun 1994

c. Discussion

We argue that the dynamical mechanisms presented in BJ are unable to adequately explain the observations from HIRS12, despite observed trends in UTRH following an intuitively sensible pattern of increasing humidity over the equatorial region and decreasing humidity in the subtropical belts.

Observed trends in UTRH are likely an artifact of the dataset's limited length, as is evidenced by the time series presented in Fig. 6. We have not rejected the hypothesis that these decadal changes are real. Other decadal-scale ocean–atmosphere variability (e.g., Folland et al. 2002; Trenberth and Hurrell 1994) may lead to changes in the distribution of UTRH. In this case we are limited by the short observational period and, on longer time scales, would expect UTRH to remain approximately constant.

The HIRS dataset is constructed by merging data from multiple instruments over the observed period. Therefore decadal-scale variability is sensitive to intersatellite calibration. This is a problem common to all long-term observing systems used in climate research. Details of the procedure to create the UTRH product and its validation are outlined in Bates et al. (1996, 2001) and assessed in Jackson and Bates (2001). The HIRS12 UTRH dataset has proved adequate in identifying interannual variability. On longer time scales, however, satellite datasets are sensitive to the treatment of individual instruments within the time series. This results in uncertainties in the magnitude of trends. In the case of the MSU temperature datasets, these uncertainties have been shown to stem from the merging of the MSU mounted on NOAA-9, where there was little overlap with other instruments (Mears et al. 2003). This highlights a requirement for multiple treatments of long-term datasets that are to be used for climate research in order to assess this sensitivity to dataset development methodology. Without this, it is not possible to estimate the uncertainty in decadal variability. Intuitively one might not expect instrumental biases to result in trends of different signs in different locations, but the biases may be related to cloud clearing processes or changes in local cloud properties and amount. Satellite drift results in changes in sampling through the diurnal cycle, which may also have a regional sensitivity.

6. Conclusions

ENSO is identified as the leading mode of UTRH variability for all seasons. We show through composite analysis that the anomalies in UTRH follow the expected patterns of changing tropical convection and subsidence through the Tropics during the winter (DJF) season of peak Pacific SST anomalies. During an El Niño, in the subtropical Pacific, changes in the general overturning can result in large regional decreases in regional UTRH and increases in temperature. Estimates of upper-troposphere temperatures from the NCEP– NCAR reanalysis suggest that between 50% and 70% of the observed subtropical Pacific changes in UTRH may be a direct result of changes in the temperature of the air there. Although the temperature estimates carry their own uncertainty, we assume that the order of magnitude estimate is correct. We expect water vapor to change in this region but attribute the largest proportion of the UTRH change in the subtropical Pacific to atmospheric temperature.

There is no statistically significant net effect of ENSO on Tropicswide UTRH. Of the five ENSO winters with the largest deviation from normal, three were identified as largely a response to changes in temperature that were not balanced by the net change in water vapor, while two events were the reverse. This does not support a single mechanism of net UTRH change due to transport of water vapor through transient mixing or other processes.

We have shown that northern summer (JJA) UTRH in the Tropics has a delayed response to the ENSO cycle. Although not widespread, local composite differences were of similar magnitude to those found in DJF. The seasonality of steep humidity gradients supported by anomalous SSTs results in large UTRH anomalies over the western Indian Ocean about the equator. UTRH changes are also observed over the Atlantic and far eastern Pacific.

Observed trends in zonal and regional means of UTRH are likely an artifact of the record's limited length and might not be expected to survive in longer-term records. We argue that a simple intensification of the hydrological cycle or increased frequency of ENSO events cannot describe the observed changes. We suggest that some other decadal-scale ocean–atmosphere variability projected onto the relatively short time series, or sensitivity to intersatellite calibration, are potential explanations of the observations. In order to validate the physical nature of the low frequency variability, further evidence of changes in dynamical mechanisms that control humidity distributions on these time scales, and uncertainty estimates as a result of intersatellite calibration, are required. The importance of these observations of decadal variability to climate change is unclear at this time, and we urge caution when attempting to interpret them in the context of the tropical energy budget.

Acknowledgments

Thanks to Simon Tett for constructive advice and comments throughout, and to Rob Allan and Tara Ansell for advice on the effects of ENSO. The work was funded by the Department for Environment, Food and Rural Affairs under Contract PECD 7/12/37. The NCEP reanalysis data and interpolated OLR were retrieved from the NOAA–CIRES Climate Diagnostics Center (http://www.cdc.noaa.gov/index.html). All-India rainfall data were obtained online from the Indian Institute of Tropical Meteorology (http://www.tropmet.res.in/;).

REFERENCES

REFERENCES
Balkan
,
S.
, and
G.
Fischer
,
Eds.,
.
1987
:
Meteorology: Physical and Chemical Properties of the Air.
Vol. 4b, Landolt-Bornstein: Numerical Data and Functional Relationships in Science and Technology—New Series, Group V, Springer-Verlag
.
Bates
,
J. J.
, and
D. L.
Jackson
,
2001
:
Trends in upper-tropospheric humidity.
Geophys. Res. Lett
,
28
,
1695
1698
.
Bates
,
J. J.
,
X.
Wu
, and
D. L.
Jackson
,
1996
:
Interannual variability of upper-troposphere water vapor band brightness temperature.
J. Climate
,
9
,
427
438
.
Bates
,
J. J.
,
D. L.
Jackson
,
F-M.
Breon
, and
Z. D.
Bergen
,
2001
:
Variability of tropical upper tropospheric humidity 1979–1998.
J. Geophys. Res
,
106
(
D23
)
32271
32281
.
Blakenship
,
C. B.
, and
T. T.
Wilheit
,
2001
:
SSM/T-2 measurements of regional changes in three-dimensional water vapor fields during ENSO events.
J. Geophys. Res
,
106
(
D6
)
5239
5254
.
Christy
,
J. R.
,
R. W.
Spencer
, and
W. D.
Braswell
,
2000
:
MSU tropospheric temperatures: Dataset construction and radiosonde comparisons.
J. Atmos. Oceanic Technol
,
17
,
1153
1170
.
Del Genio
,
A. D.
,
W.
Kovari
Jr.
, and
M-S.
Yao
,
1994
:
Climatic implications of the seasonal variation of upper troposphere water vapor.
Geophys. Res. Lett
,
21
,
2701
2704
.
Folland
,
C. K.
,
J. A.
Renwick
,
M. J.
Salinger
, and
A. B.
Mullan
,
2002
:
Relative influences of the Interdecadal Pacific Oscillation and ENSO on the South Pacific Convergence Zone.
Geophys. Res. Lett., 21, 1643, doi:10.1029/2001GL014201
.
Fu
,
R.
,
R. E.
Dickenson
, and
B.
Newkirk
,
1997
:
Response of the upper tropospheric humidity and moisture transport to changes of tropical convection. A comparison between observations and a GCM over an ENSO cycle.
Geophys. Res. Lett
,
24
,
2371
2374
.
Gaffen
,
D. J.
,
T. P.
Barnett
, and
W. P.
Elliott
,
1991
:
Space and time scales of global tropospheric moisture.
J. Climate
,
4
,
989
1008
.
Harries
,
J. E.
,
1997
:
Atmospheric radiation and atmospheric humidity.
Quart. J. Roy. Meteor. Soc
,
123
,
2173
2186
.
Held
,
I. M.
, and
B. J.
Soden
,
2000
:
Water vapour feedback and global warming.
Ann. Rev. Energy Environ
,
25
,
441
475
.
Jackson
,
D. L.
, and
J. J.
Bates
,
2001
:
Upper tropospheric humidity algorithm assessment.
J. Geophys. Res
,
106
(
D23
)
32259
32270
.
Kalnay
,
E.
, and
Coauthors
,
1996
:
The NCEP/NCAR 40-Year Reanalysis Project.
Bull. Amer. Meteor. Soc
,
77
,
437
471
.
Klein
,
S. A.
,
B. J.
Soden
, and
N-C.
Lau
,
1999
:
Remote sea surface temperature variations during ENSO: Evidence for a tropical atmospheric bridge.
J. Climate
,
12
,
917
932
.
Kousky
,
V. E,
,
Ed.,
.
1996
:
Climate Diagnostics Bulletin.
Vol. 96, Climate Prediction Center, NOAA
.
Liebmann
,
B.
, and
C. A.
Smith
,
1996
:
Description of a complete (interpolated) outgoing longwave radiation dataset.
Bull. Amer. Meteor. Soc
,
77
,
1275
1277
.
Lindzen
,
R. S.
,
1990
:
Some coolness concerning global warming.
Bull. Amer. Meteor. Soc
,
71
,
288
299
.
Mears
,
C. A.
,
M. C.
Schabel
, and
F. J.
Wentz
,
2003
:
A reanalysis of the MSU channel 2 tropospheric temperature record.
J. Climate
,
16
,
3650
3664
.
North
,
G. R.
,
T. L.
Bell
,
R. F.
Cahalan
, and
F. J.
Moeng
,
1982
:
Sampling errors in the estimation of empirical orthogonal functions.
Mon. Wea. Rev
,
110
,
699
706
.
Parthasarathy
,
B.
,
A. A.
Munot
, and
D. R.
Kothawale
,
1995
:
All India monthly and seasonal rainfall series: 1871–1993.
Theor. Appl. Climatol
,
49
,
217
224
.
Peixoto
,
J.
, and
A. H.
Oort
,
1996
:
Climatology of relative humidity in the atmosphere.
J. Climate
,
9
,
3443
3463
.
Rayner
,
N. A.
,
D. E.
Parker
,
E. B.
Horton
,
C. K.
Folland
,
L. V.
Alexander
,
D. P.
Rowell
,
E. C.
Kent
, and
A.
Kaplan
,
2003
:
Global analysis of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century.
J. Geophys. Res., 108, 4407, doi:10.1029/2002JD002670
.
Shine
,
K. P.
, and
A.
Sinha
,
1991
:
Sensitivity of the Earth's climate to height-dependent changes in the water vapour mixing ratio.
Nature
,
354
,
382
384
.
Spencer
,
R. W.
, and
W. D.
Braswell
,
1997
:
How dry is the tropical free troposphere? Implications for global warming theory.
Bull. Amer. Meteor. Soc
,
78
,
1097
1106
.
Stephens
,
G. L.
,
D.
Jackson
, and
I.
Wittmeyer
,
1996
:
Global observations of upper-tropospheric water vapor derived from TOVS radiance data.
J. Climate
,
9
,
305
326
.
Sun
,
D. Z.
, and
R. S.
Lindzen
,
1993
:
Distribution of tropical tropospheric water vapor.
J. Atmos. Sci
,
50
,
1643
1660
.
Trenberth
,
K. E.
,
1997
:
The definition of El Niño.
Bull. Amer. Meteor. Soc
,
78
,
2771
2777
.
Trenberth
,
K. E.
, and
J. W.
Hurrell
,
1994
:
Decadal atmosphere–ocean variations in the Pacific.
Climate Dyn
,
9
,
303
319
.
Vinnikov
,
K. Y.
, and
N. C.
Grody
,
2003
:
Global warming trend of mean tropospheric temperature observed by satellites.
Science
,
302
,
269
272
.
Xie
,
S-P.
,
H.
Annamalai
,
F. A.
Schott
, and
J. P.
McCreary
Jr.
,
2002
:
Structure and mechanisms of south Indian Ocean climate variability.
J. Climate
,
15
,
864
878
.

Footnotes

Corresponding author address: M. P. McCarthy, Hadley Centre for Climate Prediction and Research, Met Office, Fitzroy Road, Exeter EX1 3PB, United Kingdom. Email: mark.mccarthy@metoffice.com