A clear trend of tropical precipitation changes induced by global warming is found in hemispherical averages of most climate model simulations as well as from observation. It is observed that in response to global warming, an asymmetric pattern develops between tropical precipitation changes in the northern and southern hemispheres, and this asymmetry is locked with the seasonal cycle of tropical convection. In boreal summer (winter), the northern hemispherical average departure from tropical mean increases (decreases), while the departure of the southern hemispherical average decreases (increases). This implies an enhanced seasonal precipitation range between rainy and dry seasons and an increased precipitation difference between northern and southern hemispheres.
 In this study, we propose another perspective to examine the impact of global warming on tropical precipitation. In the previous studies [Neelin et al., 2003; Chou and Neelin, 2004], several mechanisms have already been proposed to explain how global warming affects regional tropical precipitation. These mechanisms provide a fundamental basis for examining a possible hemispherical asymmetry of tropical precipitation changes. We use the latest climate model simulations in the A2 scenarios supported by the Program for Climate Model Diagnosis and Intercomparison for the Fourth Assessment Report (AR4) of the Intergovernmental Panel on Climate Change (IPCC) to evaluate this asymmetry. Since the A2 scenario is at the higher end of the IPCC emission scenario, the warming is larger than other scenarios, such as the A1B scenario. However, we expect similar results of the tropical precipitation asymmetry no matter what scenario we choose. One realization from each of the 16 models is used. Descriptions of the climate model simulations used here can be found at http://www-pcmdi.llnl.gov/ipcc/model_documentation/ipcc_model_documentation.php.
2. Climate Model Simulations
Figure 1 shows the hemispherical average of tropical precipitation changes from 16 coupled atmosphere-ocean climate model (CGCM) simulations. The result is obtained from separately averaging precipitation changes for the northern and southern hemispheres over the tropics. Each hemispherical average is then subtracted by the tropical mean (30°S–30°N) of precipitation changes to obtain a departure from the tropical mean. This allows the southern hemispherical average to mirror the northern hemispherical average. Therefore, only the northern hemispherical average is shown here. To calculate the tropical precipitation trend in 1960–2099, hemispherical departure from tropical mean is computed for each month by a rank regression method, i.e., minimizing a product of usual variable times its rank in a centered ranking system [Hollander and Wolfe, 1999; Neelin et al., 2006]. All simulations except one, HadCM3, show a similar seasonal variation in the hemispherical-mean precipitation trend (Figure 1a). The northern hemispherical trend reaches maximum in late boreal summer (August–October; ASO) and is at its minimum in late boreal winter (February–April; FMA), and vice versa for the southern hemispherical average. This variation is consistent with the seasonal movement of the tropical convection zone which has its maximum convection zone over the summer hemisphere. The maximum (minimum) trend of the multi-model ensemble is around 0.02 (−0.02) mm day−1 per decade, with a range of (−)0.006–(−)0.04 mm day−1 per decade for different model simulations. At this rate, the seasonal range of the hemispherical-mean precipitation (difference between ASO and FMA) will be increased by 0.4 (0.12–0.82) mm day−1 at the end of the 21st century (2070–2099), a 15% (5–40%) increase with respect to the period between 1961–1990 (Figures 1b and 1c). This increase is much greater than that of the tropical-mean precipitation, which is around 0.13 mm day−1, a less than 4% increase. When the positive tropical-mean precipitation change is not removed, the enhancement of the hemispherical-mean precipitation becomes a little stronger, while the reduction of the hemispherical-mean precipitation becomes a little weaker or even unchanged (P.-H. Tan et al., Mechanisms of global warming impacts on robustness of tropical precipitation asymmetry, submitted to Journal of Climate, 2007, hereinafter referred to as Tan et al., submitted manuscript, 2007). In other words, tropical precipitation tends to increase in a rainy (summer) season and decrease slightly or remain unchanged in a dry (winter) season. The results in Figure 1 also indicate enhanced precipitation over the summer hemisphere dominated by convection and slightly reduced precipitation over the winter hemisphere dominated by subsidence. In addition, when the tropical-mean precipitation is not removed, the enhanced and reduced precipitation also has different amplitudes due to the positive tropical-mean precipitation change. By the end of the 21st century, the difference in tropical precipitation between the northern and southern hemispheres will be widened by approximately 0.4 (0.16–0.93) mm day−1 in both ASO and FMA. This is 15% (4–38%) of the precipitation difference in 1961–1990.
 To investigate mechanisms that induce the asymmetry of tropical precipitation changes, an atmospheric model with intermediate complexity coupled with a mixed-layer ocean (QTCM) [Neelin and Zeng, 2000; Zeng et al., 2000] is used. In this model, we are able to suppress the mechanisms to show their importance. In control equilibrium-doubled-CO2 experiments, the tropical precipitation changes do show a seasonal dependence (Figure 2) with similar amplitude to the CGCM simulations shown in Figure 1a. Based on moisture budget, we first examined the effect of vertical moisture advection via mean circulation −∂pq′ on this asymmetry. Here, q is the specific humidity, ω is the pressure velocity, and ∂p is a partial derivative in pressure p. denotes climatology in the current climate (1961–1990) and ( )′ is a departure from the current climate. We conducted regular-CO2 and doubled-CO2 experiments with suppression of this effect by using the climatology of vertical moisture distribution, i.e., ∂pq′ = 0, which is obtained from the control regular-CO2 run. As a result, the asymmetry of the tropical precipitation changes is reduced substantially. This demonstrates that −∂pq′ does affect the asymmetry. −∂pq′ can affect tropical precipitation directly from increasing the vertical moisture transport, a direct moisture effect [Chou and Neelin, 2004]. It can also affect tropical precipitation via feedback of anomalous vertical motion induced by atmospheric heating, i.e., −ω′∂p, which is referred to as the rich-get-richer mechanism [Chou and Neelin, 2004]. These two effects could not be distinguished in this experiment. The analysis of Tan et al. (submitted manuscript, 2007) shows that −∂pq′ is the dominant term for the precipitation asymmetry, not −ω′∂p, which is associated with the changes of the intertropical convergence zone and the Hadley circulation. We further suppress the horizontal moist static energy (MSE) advection −v · ∇(T + q), which is dominated by horizontal moisture advection −v · ∇q in the tropics, by using the same suppression technique, since this effect can also modify the vertical velocity. Here, v is horizontal velocity and T is air temperature. The result shows little change in the asymmetry (the difference between green and red curves in Figure 2), so the effect of the horizontal MSE, which is referred to as the upped-ante mechanism [Neelin et al., 2003; Chou and Neelin, 2004], does not vary the asymmetry of tropical precipitation changes much. Other effects, such as evaporation and net heat flux into the atmosphere, do not affect the asymmetry significantly either because their spatial distribution is roughly symmetric to the equator. Overall, −∂pq′ is the most dominant factor for inducing the asymmetry of the tropical precipitation changes in the QTCM1 simulations. Since water vapor increases more at the lower troposphere than at the higher troposphere under global warming, vertical moisture transport is enhanced over ascending regions (the summer hemisphere) and reduced over subsidence regions (the winter hemisphere). This creates the hemispherical asymmetry of tropical precipitation changes that varies with season.
4. Examining the Period of 1979–2005
 The next investigation into the global warming impact on tropical precipitation is over the period of 1979–2005 when observed global precipitation data is available. Figure 3 shows the ensemble mean of 16 CGCM simulations along with one observation, the Global Precipitation Climatology Project (GPCP) [Adler et al., 2003]. The ensemble mean of the tropical precipitation trend shows a distinct seasonal variation that is similar to Figure 1a. However, this seasonal variation becomes much more scattered for each individual simulation. For example, some show the seasonal variation, such as GFDL_CM2.1 and MIROC3.2(medres), but some do not. The observation also shows a tendency toward such seasonal variation. The asymmetry of the hemispherical precipitation changes is more apparent in the boreal summer than in the boreal winter. This might be due to a stronger El Niño-Southern Oscillation (ENSO) influence on the asymmetry of tropical precipitation in the boreal winter [Chou and Lo, 2007]. In a time series, hemispherical-mean tropical precipitation changes in ASO and FMA also show similar season-dependent trends in the model ensemble mean and in the observations, although the trend of these observations has a larger magnitude (Figures 3b and 3c). The trend of northern (southern) hemispherical averages for the model ensemble mean is +(−)0.024 mm day−1 per decade in ASO and −(+)0.0094 mm day−1 per decade in FMA. The trend in northern (southern) hemispherical changes for the GPCP precipitation is +(−)0.035 mm day−1 per decade in ASO and −(+)0.036 mm day−1 per decade in FMA. Because the observed global precipitation is only available for a relatively short time period (1979–2005), these trends scatter in amplitude. The ASO trends both in observations and in the model ensemble have passed the 95% statistical confidence level of a Spearman-rho test, but the FMA trends have not, due to strong influences of El Niño in the boreal winter [Chou and Lo, 2007]. Nevertheless, they do show a tendency of asymmetry in the hemispherical-mean tropical precipitation changes. According to the multi-model ensemble mean, the FMA precipitation trend will exceed the 95% confidence level in 2010 and 99% in 2015.
 Our study shows that under global warming, two consistent signals are found among climate model simulations. First, the seasonal precipitation range is increased; secondly, the precipitation difference between northern and southern hemispheres is widened in the boreal winter and summer. These signals imply a tendency that the wet season gets wetter and the dry season gets slightly drier or remains unchanged. Furthermore, the signals found in this study are much stronger than in a commonly used index, global (tropical) mean precipitation. Thus, they constitute a better index and can be potentially used for early detection of global warming impacts on the global hydrological cycle. For the short period with available global observation data (1979–2005), the asymmetric pattern of hemispherically-averaged tropical precipitation changes shows a similar tendency, but with less confidence in the boreal winter. Figure 4 shows a time series of the ocean-only tropical mean vertically integrated water vapor from the version 6 of SSMI [Wentz, 1997] and the ensemble mean of the CGCM simulations. A clear moistening trend is found in both the observation (1.63% per decade) and the multi-model ensemble mean (1.37% per decade), using the same rank regression method described in section 2. This moistening trend of the column integrated water vapor is strongly associated with sea surface temperature warming [Wentz and Schabel, 2000; Trenberth et al., 2005]. Since the increase of water vapor is the main mechanism for inducing the hemispherical asymmetry of tropical precipitation changes, this asymmetry should become more apparent and detectable as the Earth continues to warm up.
 We acknowledge the international modeling groups for providing their data for analysis, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) for collecting and archiving the model data, the JSC/CLIVAR Working Group on Coupled Modelling (WGCM) and their Coupled Model Intercomparison Project (CMIP) and Climate Simulation Panel for organizing the model data analysis activity, and the IPCC WG1 TSU for technical support. The IPCC Data Archive at Lawrence Livermore National Laboratory is supported by the Office of Science, U.S. Department of Energy. This work was supported by the National Science Council Grant 95-2111-M-001-001. We thank James Fan, Shirley Chao, and S.-C. Lung for help during the preparation of this manuscript.